Language, Cognition and Space

Advances in Cognitive Linguistics

Series editors: Benjamin K. Bergen, Vyvyan Evans & Jörg Zinken
The Cognitive Linguistics Reader
Aspects of Cognitive Ethnolinguistics. Jerzy Bartmiński. Edited by: Jörg Zinken

Forthcoming titles in the series:

Language and Representation. Chris Sinha
Language, Cognition and Space:
The State of the Art and New Directions

Edited by

Vyvyan Evans and Paul Chilton

Published by

Equinox Publishing Ltd

UK: Unit 6, The Village, 101 Amies St, London, SW11 2JW
USA: DBBC, 28 Main Street, Oakville, CT 06779

www.equinoxpub.com

Language, Cognition and Space

First published 2010

© Vyvyan Evans, Paul Chilton and contributors 2010

All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage or retrieval system, without
prior permission in writing from the publishers.

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

ISBN 978-1-84553-252-9 (hardback)

ISBN 978-1-84553-501-8 (paperback)

Library of Congress Cataloging-in-Publication Data

Language, cognition, and space : the state of the art and new directions
/ edited by Vyvyan Evans and Paul Chilton.
p. cm. -- (Advances in cognitive linguistics)
Includes bibliographical references and index.
ISBN 978-1-84553-252-9 (hb) -- ISBN 978-1-84553-501-8 (pb) 1.
Cognitive grammar. 2. Space perception. I. Evans, Vyvyan. II. Chilton,
Paul A. (Paul Anthony)
P165.L38 2009
415--dc22
2008054445

Typeset by Catchline, Milton Keynes (www.catchline.com)

Contents
Introduction
Paul Chilton 1

Part I: Perception and space 19

1 The perceptual basis of spatial representation
Vyvyan Evans 21

Part II: The interaction between language and

spatial cognition 49
2 Language and space: momentary interactions
Barbara Landau, Banchiamlack Dessalegn and
Ariel Micah Goldberg 51
3 Language and inner space
Benjamin Bergen, Carl Polley and Kathryn Wheeler 79

Part III: Typological, psycholinguistic and neurolinguistic

approaches to spatial representation 93
4 Inside in and on: typological and psycholinguistic perspectives
Michele I. Feist 95
5 Parsing space around objects
Laura Carlson 115
6 A neuroscientific perspective on the linguistic encoding of categorical spatial
relations
David Kemmerer 139

Part IV: Theoretical approaches to spatial representation

in language 169
7 Genesis of spatial terms
Claude Vandeloise 171
8 Forceful prepositions
Joost Zwarts 193
9 From the spatial to the non-spatial: the ‘state’ lexical concepts of in, on and at
Vyvyan Evans 215
 Part V: Spatial representation in specific languages 249
10 Static topological relations in Basque
Iraide Ibarretxe-Antuñano 251
11 Taking the Principled Polysemy Model of spatial particles beyond English:
the case of Russian za
Darya Shakhova and Andrea Tyler 267
12 Frames of reference, effects of motion, and lexical meanings of Japanese
front/back terms
Kazuko Shinohara and Yoshihiro Matsunaka 293

Part VI: Space in sign-language and gesture 317

13 How spoken language and signed language structure space differently
Leonard Talmy 319
14 Geometric and image-schematic patterns in gesture space
Irene Mittelberg 351

Part VII: Motion 387

15 Translocation, language and the categorization of experience
Jordan Zlatev, Johan Blomberg and Caroline David 389
16 Motion: a conceptual typology
Stéphanie Pourcel 419

Part VIII: The relation between space, time and modality 451
17 Space for thinking
Daniel Casasanto 453
18 Temporal frames of reference
Jörg Zinken 479
19 From mind to grammar: coordinate systems, prepositions, constructions
Paul Chilton 499
Index 515
List of contributors
Benjamin Bergen (University of California, San Diego)
Johan Blomberg (Lund University),
Laura Carlson (University of Notre Dame)
Daniel Casasanto (MPI for Psycholinguistics, Nijmegen)
Paul Chilton (University of Lancaster)
Caroline David (Université de Montpellier 3)
Banchiamlack Dessalegn (Johns Hopkins University)
Vyvyan Evans (Bangor University)
Michele Feist (University of Louisiana at Lafayette)
Ariel Micah Goldberg (Johns Hopkins University)
Iraide Ibarretxe-Antuñano (Universidad de Zaragoza)
David Kemmerer (Purdue University)
Barbara Landau (Johns Hopkins University)
Yoshihiro Matsunaka (Tokyo Polytechnic University)
Irene Mittelberg (RWTH Aachen University)
Carl Polley (University of Hawai‘i, Mānoa)
Stéphanie Pourcel (Bangor University)
Darya Shakhova (Georgetown University)
Kazuko Shinohara (Tokyo University of Agriculture and Technology)
Leonard Talmy (University at Buffalo, State University of New York)
Andrea Tyler (Georgetown University)
Claude Vandeloise (Louisiana State University)
Kathryn Wheeler (University of Hawai‘i, Mānoa)
Jordan Zlatev (Lund University)
Joost Zwarts (Utrecht University)
Jörg Zinken (Portsmouth University)
 Introduction
Paul Chilton

You have carried out a complex series of non-linguistic spatial tasks when you picked
up this book. You were not aware of all of them but they included, perhaps, navigat-
ing your way visually to, into and around a book shop or library. Perhaps you used
some specifically space-related language: ‘Where can I find that book on language
and space’?
This book is to a large extent about how these and other unconscious spatial activi-
ties – in the mind and by way of the body – relate to language and how language relates
to space. This is an area of enquiry that has interested linguists, philosophers, and
psychologists for a very long time and for a variety of reasons. In recent years, say
over the past fifteen years, cognitively oriented linguists have devoted more and more
research effort into trying to understand the space-language relationship. The present
book aims to give an overview of some aspects of this effort, its current state, and the
directions in which it is heading.

The long view

The concern with language and space can be seen in a historical perspective, one that
begins with physical space per se rather than its relationship with language. The study
of space emerged among the ancient Babylonians and Greeks and was received by
European civilisation in the form of Euclidean geometry. Aristotle added a view of
space that saw it as places embedded in containers, rather than relations, ambiguously
using the same Greek word, topos, for ‘place’ as well as ‘space’. The next conceptual
breakthrough was probably the development of analytic geometry by Descartes and
projective geometry by Desargues. It was not until the nineteenth century that non-
Euclidean geometries were developed – in effect extending the concept of ‘space’ beyond
what could be intuited through everyday perception.
The concept of space posed problems treated in philosophical discourse before space
was investigated by scientific methods that were both empirical and also depended on
extensions of mathematics. According to Kant (1781/1963), propositions about space
are synthetic a priori. This means that propositions about space are not analytic, that is,
self-defining. Nor are they dependent on sense-experience, or a posteriori. Rather, Kant
conceives space as intrinsically incorporated in human understanding. The implication

1
2 LANGUAGE, COGNITION AND SPACE

is that human minds are innately endowed with a Euclidean perception and concep-
tion, or one might say ‘construction’, of physical reality. While Kant’s philosophical
formulation leaves many details unaccounted for, it is consonant with modern views
that the human mind has evolved in a way that provides a construction of space fitted
to human survival. It is important to bear in mind that such a construction may or
may not correspond directly to the objective nature of the physical. Indeed, from the
beginning of the twentieth century, mathematics and physics have shown that space
is not structured in terms of Euclidean space, although Euclidean geometry serves on
the local and human scale.
Somewhat later in the twentieth century, neuroscientists began to put together
a picture of the way the human brain may construct space at the neuronal level. The
neuro-scientist John O’Keefe, who has contributed pioneering work to mammalian
spatial cognition, argues that the three-dimensional Euclidean construction is inherent
in the human nervous system. He further argues that it is this necessity that leads to
the indeterminacies that appear to exist at the quantum level. (O’Keefe 1999: 47–51).
In the present volume, Chapter 1 by Vyvyan Evans, as well as Chapter 6 by David
Kemmerer, outlines further details of what many other researchers in cognitive science,
psychology and neuroscience have discovered about the human embodiment of spatial
experience. Among the findings most relevant from the linguistic point of view is the
functional distinction between egocentric frameworks of spatial conceptualisation
(neurally embodied in the parietal cortex) and allocentric frameworks instantiated in
the hippocampus and surrounding structures.
Language enables humans to communicate about many things – to stimulate
conceptualisations among members of a group sharing a language. We can assume
that communicating about spatial locations and movements is one area that has
particular significance in the evolution of language and languages. And possibly it
is the most fundamental area. Indeed, it appears that while not all languages have
a word for ‘space’, there is no language that does not have various classes of words
that refer to spatial experience. Some of these words have very general referential
power, related to the relative position of speaker or addressee – demonstratives like
this and that for example, which are learned very early by children (cf. Haspelmath
2003). It is equally clear that the human brain has neuronal modules specialised for
the perception and cognitive processing of various physical phenomena, such as
shape, distance, direction, location and locomotion. As already noted, neuroscientists
have accumulated and are accumulating evidence of what exactly these systems are.
However, what remains very much less clear is the relationship between linguistic
expressions for space, in all their variability and similarity across the world’s lan-
guages, and the various interacting non-linguistic systems of spatial cognition. This
relationship is in fact hotly debated among linguists. In order to make headway it is
important to be as clear as we can about the precise questions we need to ask about
the language-cognition relationship.
 INTRODUCTION 3

How can we think about the relationship between language and

space?

The meanings we find coded in linguistic structures relate not to the physicist’s culturally
developed understanding of space but to the naturally evolved (and possibly limited) way
in which the human brain constructs space for purposes of survival. However, stating
the issue in this way makes an important assumption that needs to be questioned. To
speak of the evolved system of spatial representation in the human brain is to imply that
the system is universally the same for all human individuals. And also it implies that
whatever the language they happen to speak it is the same for all human individuals. It
seems natural to think that all humans would negotiate their physical surroundings in
the same way but this view is seriously challenged by scholars who emphasize cultural
differences and by scholars who take seriously Whorf ’s well known view that different
languages influence or determine conceptualisation, including spatial conceptualisation
(Whorf 1939/2000). After the period in which Whorf ’s claims were widely seen as
discredited, ‘neo-Whorfian’ research has become intellectually respectable (cf. Gumperz
and Levinson 1996, Levinson 1996, Levinson 2003 and many other works) and provides
the framework for several chapters of the present volume.
An important question in this relativistic perspective concerns the limits of varia-
tion in the way languages encode concepts of space. Variation may not be random but
follow universal patterns. If this is the case, then we should further want to explain these
patterns in terms of properties of the human space-representing apparatus. It remains
logically possible that it is physical objective space itself that structures the human
experiential apparatus that variably structures linguistic meanings. It might nonetheless
be said, following O’Keefe’s arguments (O’Keefe 1999), that it is likely that the human
apparatus in some sense imposes its way of seeing things on the physical world. Even so,
this does not rule out some form of a modified realist view of the relationship between
human cognition and an objective physical universe. In any event, we need to pose at
least the following questions.
How exactly do languages encode spatial concepts? That is, what spatial meanings
does a language enable its users to exchange among one another? It is important to
distinguish language from a language, for languages may in principle differ in the spatial
meanings that they encode. A working hypothesis is that languages do indeed differ in
this regard and various empirical investigations have been undertaken in an attempt to
prove or disprove it. We need to ask whether differences between languages in regard
to the way they express spatial relationships are random and unconstrained. It is quite
possible, in a purely logical sense, that variation could mean much overlap and small
differential features. It cannot be ruled out of course that even small differences in spatial
encoding among languages could correspond to significant cognitive distinctions. Two
crucial questions for the contributors to this volume are therefore: Do differences in
linguistic encoding of spatial concepts affect non-linguistic conceptualisation of space?
And, if so, which elements of spatial encoding are involved? There is now a growing
body of empirical research aiming to answer this kind of question. The present volume
reports on a wide range of empirical investigations that give varying answers.
4 LANGUAGE, COGNITION AND SPACE

There is a further dimension of the space-language question that is rather different,

and also more speculative, though has been for some time influential in cognitive
linguistics. This line of thinking concerns the possibility that spatial conceptuali-
sations provide the basis for non-spatial expressions, including abstract ones. The
theories associated with such a perspective go much further than, for example, a
view of language that simply postulates a separate module of language to deal with a
separate module of human spatial cognition (cf. the models of Jackendoff 1993, 2002).
According to such theories, spatial cognition motivates the coding of concepts that are
not in themselves self-evidently spatial and in many parts of linguistic structure. This
perspective is perhaps clearest in the case of lexical meanings. The lexicalisation of
spatial meanings to express temporal meanings is the best known case. For example,
temporal terms such as before and after are etymologically spatial terms, as noted by
Traugott (1975), while the moving ego and moving time metaphors have been much
discussed since at least Lakoff and Johnson (1980). But similar observations for other
domains (cf. ‘his behaviour went from bad to worse’) have been noted since Gruber
(1965) and discussed in detail by Talmy (2000) under the rubric of ‘fictive motion’.
But spatial concepts may well be more deeply integrated with linguistic structure,
including grammatical constructions. In the 1970s, some linguists went a considerable
distance along this road under the banner of ‘localism’, treating grammatical categories
such as tense and aspect as spatially grounded, as well as, for example, causatives,
modals, transitivity, instrumental adverbs, possessive and existential constructions
(cf. Lyons 1977: 718–724; Anderson 1971). Although the term ‘localism’ is no longer
used, many approaches in cognitive linguistics are consistent with this idea. In some
of its manifestations cognitive linguistics is heavily dependent on spatially iconic
diagrams for the purpose of describing a very wide range of linguistic phenomena
(e.g. Langacker 1987, 1991). Further developments in formalising the spatial basis
of language structure are reflected in Chilton, current volume and 2005). It is also
worth noting that O’Keefe’s pioneering work in mammalian spatial cognition utilises
a geometrical framework that he links speculatively with the evolution and structure
of human language (O’Keefe 1996 and 2003).

Overview of the present volume

The interest in space and language culminated in the ground-breaking collection of

papers by Bloom et al. published in 1996. That volume contained an interdisciplinary
perspective with contributions from psychologists, cognitive scientists and biologists, as
well as linguists. Some of these contributors are also contributors to the present volume.
But since the editors’ aim, in the present volume, was not only to sketch the state of the
art but also to break new ground and explore new directions, there are many important
papers from a new generation of researchers working in cognitive linguistics or in areas
overlapping with its concerns.
Following the overview of biological mechanisms involved in the non-linguistic
perception of spatial relationships in Section I, Section II opens up the questions
 INTRODUCTION 5

that cluster around the Whorf hypothesis, questions that also come up in Section
III. Barbara Landau and her associates in chapter 2 propose a solution to the prob-
lem of whether the structure of a particular language can influence or determine
non-linguistic spatial cognition. Spatial language and visual representations of space
organise spatial experience in distinct ways, though the two systems overlap. Landau
and colleagues propose that this overlap can be considered in terms of two possible
mechanisms – selection and enrichment.
An example of selection is the way in which some languages ‘choose’ to code
direction rather than manner in verbs of motion – a topic that is controversial and
much researched, as will be see from later chapters of this book (Section VII). Another
example of selection is frames of reference – also an important topic and one that has
led to strong Whorfian claims (Levinson 1996, 2003, challenged by Li and Gleitman
2002). An example of enrichment is the use of spatial language to facilitate orientation
tasks by combining geometric and non-geometric information. The chapter describes
experimental evidence indicating that language can influence non-linguistic cognitions
that can lead to erroneous spatial judgements.
For both selection and enrichment, the question is whether language has a per-
manent impact on non-linguistic cognition. The authors argue that if it has, then such
effects are to be found ‘in the moment of a task’, when it is possible that language is
temporarily modulating attention. This is an approach to Whorfian claims that is
similar to Slobin’s (1996) notion of ‘thinking for speaking’.
Chapter 3, by Benjamin Bergen and colleagues also addresses the possibility that the
way a particular language encodes spatial expressions may influence the non-linguistic
spatial systems, either in the long term or for the task in hand. If a language can have
a particular effect, one has to ask what becomes of universalist assumptions about
languages. Like the previous chapter, this chapter also takes the view that the variation
in respect of spatial expressions lies within limits and that languages overlap. Further,
Bergen and colleagues consider the relationship between spatial and abstract meanings.
The issue is whether using such expressions as ‘the winter is coming’ means that speakers
are activating spatial concepts or lexicalised non-spatial meanings. In this chapter the
authors review a number of experiments, including their own, that strongly indicate
two kinds of relationship between language and cognition. The first kind of relation-
ship is one in which linguistic spatial expressions activate the same neuro-circuitry
as non-linguistic spatial cognition. These effects seem to apply not only to primary
spatial expressions such as prepositions but to spatial components of verbs and nouns.
The second kind of relationship is one in which lexicalised expressions metaphorically
derived from spatial concepts, such as those for time, also activate spatial cognition,
though evidence from brain lesions is reported that may modify this finding. The chapter
assesses also in what sense cross-linguistic differences in the space-time metaphor
influence non-linguistic temporal thinking. Whatever the linguistic expression, we
also want to know to what extent processing language about space uses the same parts
of the brain as processing space per se. Here and also in Kemmerer’s chapter (Chapter
6), the reported evidence suggests substantial overlap, although the exact nature and
extent of this overlap remains for further research.
6 LANGUAGE, COGNITION AND SPACE

What methodologies can we use to investigate the relationship between spatial

expressions in language and in non-linguistic spatial cognition? What can we learn
from these various methodologies? In Section III we present three different approaches
to the descriptive analysis of linguistically activated spatial concepts. Scientific models
of space, as noted earlier, have depended on mathematical formulations, from Euclid’s
geometry, Cartesian coordinates, projective geometries, to topology, spacetime and the
seemingly strange behaviour of location and time at the quantum level. As was also
noted, it may be the case that Euclidean dimensional geometry, coordinates and vectors
are intrinsic to the human systems of spatial cognition. However, although language
systems also draw on geometrical concepts – as they must if the claims of the last two
chapters are right and if it is also true that non-linguistic cognition is fundamentally
geometrical, it does not follow that this is all there is to linguistic encoding of spatial
concepts. Section III opens by addressing some of the issues that confront linguists
when they focus on what exactly it is that constitutes spatial meaning in human lan-
guages. In general, the picture that is emerging is that linguistic expressions for spatial
concepts involve not only concepts that can be described geometrically but also other
concepts useful to humans. Michele Feist’s chapter (Chapter 4), for instance, argues that
the variation and the commonality found among the world’s languages with respect
to spatial expressions cannot be adequately described by geometrical means alone.
Rather there are three factors, one of which is geometry, the other two being ‘functional’
attributes and ‘qualitative physics’. The relevance of functional attributes has been widely
commented on (by, for example, Talmy, Vandeloise, Tyler and Evans) and this factor
implies that language users draw on various cognitive frames in addition to spatial ones
when producing and interpreting spatial utterances. Similarly, Feist claims, speakers
draw on naïve physics involving force-related cognition such as support and control of
movement. By what means can linguists investigate and describe such dimensions? In
addition to psycholinguistic experimentation, drawn on extensively in the preceding
chapters, Feist highlights the importance of language-typology studies, and reports
findings for the semantics of words corresponding to in and on across 27 languages
from 9 language families.
Nonetheless, geometrical approaches remain a fundamental concern. In Chapter
5, Laura Carlson proposes a new methodology for investigating how several factors
modify geometrical frames of reference. The term ‘frame of reference’ has become
standard among researchers into human conceptualisation, though with some variation
in application (cf. the review in Levinson 2003: 26). Frames of reference are essentially
three-dimensional coordinate systems with a scalar (not strictly metric) quality. They
involve direction and (non-metric) distance. The coordinate systems vary in the way (or
ways) a particular language requires them to be set up. The origin and orientation of the
three axes may be located on the self, another person or fixed by the environment (e.g.
landscape or earth’s magnetic field). The axis system may be geometrically transformed,
e.g. rotated, reflected or translated from their origo located at some reference point,
typically the speaker. Each of these possibilities constitutes a reference frame. It should
be noted that in addition to coordinate geometry, linguists have often invoked, though
 INTRODUCTION 7

loosely, the mathematical notion of topological relations, to describe spatial meanings

such as those of containment and contact.
So much is taken for granted. Carlson’s method, however, shows how spatial terms
such as prepositions have meanings that can be defined by ‘spatial templates’ (another
term already in the literature) or spatial regions within the three dimensional coordinate
systems. What is crucial here is that Carlson’s methodology shows, consistently with
some of Feist’s points, how such spaces can be influenced by non-spatial linguistic
meanings. Different kinds of object in different kinds of settings have varying regions
projected by a particular preposition. In different cases, what counts as being in front
of something is shown to vary depending, amongst other things, on what the reference
object actually is, the speaker’s typical interaction with it, the presence of other objects,
and the particular reference frame being used.
Linguistic analysis and psycholinguistic experimentation give only indirect access
to what is happening at the level of brain structure. David Kemmerer’s chapter (Chapter
6) surveys the work of cognitive neuroscientists, focussing on one aspect of linguisti-
cally mediated spatial cognition, location. Like the authors of the preceding chapters,
Kemmerer takes into consideration the language-typological questions concerning
variation and commonality, as well as the possible Whorfian effect of language on
non-linguistic spatial cognition. Kemmerer poses the question ‘what types of categori-
cal spatial relations are encoded by language?’ Here ‘categorical’ is the term used by
some psycholinguists and cognitive scientists to refer to the class of linguistic terms
devoted to spatial relations (e.g. prepositions). Among categorical relations Kemmerer
makes important distinctions between three kinds of spatial relation. The first is deictic
relations as found in demonstratives, which non-metrically divide space into proximal
or distal zones, varying in number from two (most commonly) to four, depending on
the language. These systems may be centred on the speaker (again, most commonly),
addressee or some geographical feature. The second is topological relations of the
kind alluded to earlier. Reviewing the cross-linguistic evidence, Kemmerer tentatively
concludes that there is a universal conceptual space with strong ‘attractors’. These are
kinds of topological relation, e.g. containment, which languages are statistically likely to
encode. The third is ‘projective’ relations – that is, reference frames in the sense outlined
above. While acknowledging the cross-linguistic variety and what may seem prodigious
cognitive feats associated with the use of an ‘absolute’ (or geocentric) reference frame,
Kemmerer notes the relatively small number of core spatial concepts encoded cross-
linguistically.
But what are the neuro-anatomical correlates? The answers come from the field of
cognitive neuro-science, a field that draws its evidence from brain-lesion studies, which
work by inference, and brain scanning techniques, typically PET (positron emission
tomography) and fMRI (functional magnetic resonance imaging), which are capable
of providing information of activated brain regions in on-line linguistic and cognitive
tasks. The state of the art in this still developing field is summarised by Kemmerer
in terms of brain regions likely to be implicated. Many studies indicate that the left
inferior parietal lobule is involved – the region into which projects the ‘dorsal’ or so-
called ‘where’ pathway of the visual system, involved in precise locating of objects for
8 LANGUAGE, COGNITION AND SPACE

sensorimotor interaction. (The other pathway is the ‘ventral’ or ‘what’ pathway, which
identifies objects.) Linguists have variously speculated about correlations between these
two pathways and aspects of linguistic organisation. See Landau and Jackendoff 1993
and Hurford 2003a and 2003b) Further, Kemmerer reports studies that suggest that
categorical space relations are processed in the adjacent regions of the supramarginal
gyrus and possibly the angular gyrus. With regard to deictic, topological and projective
relations several brain regions in addition to those mentioned may be involved, perhaps
the inferotemporal cortex (linked to the ventral ‘what’ pathway).
The final part of chapter 6 addresses the Whorfian question, drawing evidence
from brain-lesion data and from psycholinguistic experiments on normal individuals.
One set of findings indicates that linguistic and perceptual-cognitive representations of
categorical spatial relations are to some extent distinct. Another set of findings, however,
seems to give some plausibility to the claim that linguistic representation of space can,
in a certain sense, influence perceptual-cognitive representation of space. This may
happen in two ways. The acquisition of a certain language may decrease an infant’s
sensitivity to certain categorical spatial distinctions (e.g. mirror image differentiation
in Tzeltal speakers) or it may lead to the converse, an increase in sensitivity to certain
categorical spatial distinctions. An example of the latter is the finding that Tzeltal
speakers utilize an absolute (geocentric) frame of reference in non-linguistic cognitive
tasks (cf. Levinson 1996, 2003). Kemmerer is cautious, however, in interpreting such
findings as indicating permanent influence on spatial cognition, suggesting a similar
interpretation to that put forward by Landau and colleagues in Chapter 2 – namely,
that particular linguistic encoding has its primary effect in the moment of executing
a task. It is premature to close the debate about Whorfian effects: this is an area for
future interdisciplinary research.
It is clear that investigating the relationships between linguistic expressions for space
and non-linguistic spatial cognition requires an adequate description on the linguistic
side. What are the descriptive methods? What theoretical frameworks are required for
such methods? And in which new research directions do these frameworks point? Part
IV offers three theoretical approaches to spatial representation in language.
In Chapter 7 Claude Vandeloise builds on his previous work to outline a theory
of spatial expressions that makes claims in a diachronic perspective as well as a cross-
linguistic synchronic perspective. The theoretical starting point is Berlin and Kay’s
implicational scale for colour terms (modified by MacLaury) and, for spatial terms, the
hierarchical classification of adpositions proposed by Levinson and Meira (see Chapter 7
for detailed references). This chapter argues that the Levinson-Meira model has several
problematic features and an alternative hierarchy is proposed. To do this Vandeloise
uses several of the theoretical ways of categorising spatial expressions that have been
introduced in the previous sections of this book. While the Levinson-Meira model
includes only topological kinds of expression, excluding projective (frame of reference)
expressions, Vandeloise includes projective and dynamic concepts as well as topological
ones. The dynamic concepts are of some significance, since they rest on physical notions
of force. Vandeloise does, however, like Levinson and Meira exclude ‘kinetic’ expressions
such as English from and to (which might also be called directional). The most striking
 INTRODUCTION 9

result of this reanalysis is the claim that the most abstract spatial distinction is between
‘topological’ expressions and ‘control’, the latter involving concepts of force and energy.
Sub-divisions of the proposed hierarchy include projective relations under ‘location’,
while ‘containment and ‘support’ come under ‘control’. The hierarchy is claimed to apply
to languages universally and also to have relevance for the sequential genesis of spatial
terms in the history of particular languages.
Is it possible to have a unified model combining both geometric and force dynamic
properties? In Chapter 8, Joost Zwarts proposes a way to model forces that uses the same
notational and theoretical framework as is used for coordinate geometry. However, he
extends this framework by introducing the elementary mathematics of vectors. This
approach may seem unusual to linguists who have worked on spatial expressions,
but it is a natural complement to the notion of frames of reference. In fact, O’Keefe
(1996, 2003) has already proposed a ‘vector grammar’ for spatial expressions and other
researchers, including Zwarts himself, have pursued the idea (see references in Chapter
8 and also in van der Zee, E. & Slack, J. (eds) 2003). Chilton (2005 and Chapter 19
this volume) takes this framework in a more abstract direction. Because vectors are
conventionally used in the applied sciences to model forces, as well as locations in
axis systems, Zwarts is able to address ‘force-dynamic’ prepositions such as against,
which are not in included in the other classifications, as well as the ‘control’ type and
the ‘support’ type of prepositions. Further, he is able to address spatial verbs that are
both directional and ‘forceful’ like push and pull and also to accommodate semantic
notions such as Agent and Patient. Zwarts’s vector-based framework offers a way of
representing and combining spatial relations with the general notion of ‘force dynamics’
that is much invoked by cognitive linguists and sometimes regarded as distinct from
geometry-based descriptions.
Vyvyan Evans’s approach (Chapter 9) to three English prepositions (in, on and
at) maintains the distinction between spatio-geometrical and ‘functional’ aspects of
spatial meaning, where ‘functional’ covers ‘humanly relevant interactions’ with objects
in particular spatial configurations. It does this within a cognitive-linguistic theoretical
framework that proposes an explicit and wide-ranging theory of linguistic meaning,
addressing some of the central issues outlined earlier, in particular, the questions con-
cerning the nature of the interface between linguistically-encoded concepts and non-
linguistic concepts. This theoretical framework, which Evans calls the Theory of Lexical
Concepts and Cognitive Models (LCCM Theory), is a refinement of Tyler and Evans’s
earlier Principled Polysemy theory and addresses the important question of extended
word meanings. The incorporation of a diachronic perspective is a crucial element of
this new theory. What is at issue is the problem of accounting for the emergence of
non-spatial meanings of originally spatial terms – such as in love, on alert, and the like.
In regard to the question of the relationship between linguistic meaning and
non-linguistic conceptualization, LCCM Theory starts from the claim that language
encodes abstracted, schematic or ‘skeletal’ concepts, referred to as ‘lexical concepts’,
independently of non-linguistic representations stored as in cognitive models, which
are richer. There is a further distinction in LCCM Theory, the distinction between
closed-class and open-class forms, which have already been given importance in the
10 LANGUAGE, COGNITION AND SPACE

work of Leonard Talmy (cf. also his chapter, Chapter 13 in the present volume). Both
types are pairings of linguistic form and schematic concepts, but the open class forms
provide ‘access sites’ to cognitive models that tend to be more complex, richer and
variable, while closed-class forms (including prepositions) are associated with concepts
that are relatively more schematic.
Evans’s main concern in the present chapter is with prepositions. Evans’s claim is
that prepositional concepts are made up from a constellation of conventionalised mean-
ings clustered around a central or prototypical ‘spatial scene’, a concept that includes
both a spatio-geometric and a functional element. He further claims that the central
concept gives rise to ‘parameters’ (or is ‘parameterized’) over time. A parameter is a kind
of sub-schema attached to the central or prototypical schema. Furthermore, language
only encodes knowledge via parameterization, in Evans’s sense of the term. According
to the theory, parameterization arises from the extension in use of the functional
(rather than the spatial) ingredients of the central concept. Parameters are abstractions
over functional concepts resulting from the use of the central concept in particular
human contexts. Over time, such uses become associated with particular functional
parameters of the central lexical concept, itself associated with a particular language
form: whence the phenomenon of polysemy. Thus, for example, the preposition in has
physical enclosure as its central and earliest meaning, but eventually produces various
distinguishable ‘state’ parameters, for example ‘psychosomatic state’, as observed in
expressions like in love. The non-physical meanings of in are thus not computed for
each expression on line, but are entrenched meanings associated with the linguistic
form in for speakers of English – which is not to say that new meaning parameters are
not so computed, for this is the very mechanism by which polysemy becomes extended
diachronically. In sum, ‘states are locations’, as Conceptual Metaphor Theory showed
us, but the LCCM account puts forward a refined and more detailed explanatory and
descriptive framework.
Theoretical frameworks such as those just summarised are an integral part of the
overall research endeavour. The empirical investigations of individual human languages
makes no sense unless the framework pf description is made explicit and this is why
much of the literature on language and space has been, and continues to be, taken up
with theoretical refinements. But the reverse is also true – theory has to be comple-
mented and integrated with cross linguistic evidence. Sections V and VI of this volume
constitute a sample of such evidence. Since it is the descriptive detail that is crucial,
we shall summarise only briefly the content of these chapters, leaving the individual
chapters to speak for themselves.
In these chapters examples of spatial expressions are examined in four languages:
Basque, Russian, Japanese and American Sign Language. So far we have referred in
this Introduction mainly to the English encoding of spatial concepts in prepositions.
However, the world’s languages (including to some extent also English) distribute spatial
concepts across various morpho-syntactic categories. In Chapter 10 Iraide Ibarretxe-
Antuñano shows how, in Basque, spatial concepts are distributed across case inflections,
‘spatial (or locative) nouns’, and motion verbs. Ibarretxe-Antuñano investigates the
statistical frequencies of choices from these categories made by Basque speakers in
 INTRODUCTION 11

responding to pictorial images of the kinds of spatial relationship that have been termed
topological. In addition to the findings of these experiments, Ibarretxe-Antuñano notes
that two further dimensions may be involved in Basque spatial expressions, namely
dynamicity and agentivity. Such concepts may of course also be relevant for research
into other languages and for the general theory of linguistic spatial concepts.
Russian is also a language whose encoding of spatial concepts includes case-marking
and rather rich polysemy. In Chapter 11, Darya Shakhova and Andrea Tyler take Russian
linguistic spatial expressions as a test case for the theory of Principled Polysemy, already
discussed in Chapter 9. If a theory of spatial expressions is to be of interest, it needs
to make universal claims about human language – that is, it needs to be applicable to
individual language systems. In this chapter Shakhova and Tyler claim that Principled
Polysemy theory does indeed make descriptive predictions that are borne out when the
Russian data for the preposition za are examined in detail. This, as we have noted, means
that functional concepts play a crucial role in combination with geometrical (in this case
projective) concepts. A detailed analysis of the polysemy network of za emerges. One
particularly interesting finding, relevant for other case-marked languages, concerns the
use of Russian instrumental case with verbs like those corresponding to English follow.
A similar general approach is adopted in Chapter 12, by Kazuko Shinohara and
Yoshihiro Matsunaka, who investigate the concepts associated with three spatial preposi-
tions in Japanese: mae, ushiro, saki. These prepositions have some conceptual similarity
with English in front of, and so raise questions about their relationship to frames of
reference and the geometric transformations of such frames. Again, the aim is not only
to enrich our knowledge of the semantics of a particular language but also to test certain
theoretical claims. There are two theoretical claims at issue. The first concerns reference
frames, which Levinson and others have claimed are encoded linguistically while others
(see references in chapter 12 to Svorou, Carlson-Radvansky and Irwin) have asserted
that reference frames are not coded linguistically. The second theoretical issue is what
Tyler and Evans, following earlier work by Dominiek Sandra, have called the ‘polysemy
fallacy’: the attribution of unnecessarily many separate meanings to a single lexical item
when meaning differences can be explained in terms of contextual inference. This kind
of proliferation is typical of some of the earlier cognitive linguistics approaches, e.g.
Lakoff ’s multiple image schemas for the preposition over. Shinohara and Matsunaka
investigate a particular kind of contextual condition on the conceptualisations associ-
ated with mae, ushiro, saki – namely, the effect of the motion of the perceiver and the
motion of the perceived object. What the authors of Chapter 12 find, on the basis both
of linguistic analysis and psycholinguistic experiments, is that the Japanese prepositions
have a minimal specification that includes reference frame information and that the
contextual conditions in which they are used makes an important contribution to the
associated conceptualisations. In general, their findings uphold Levinson’s claims and
also those of Tyler and Evans.
The accumulation of empirical findings within a coherent theoretical perspective
may eventually lead to a deeper understanding both of universals and of variation in
the world’s languages. The implicit aim is to gain indirect evidence about the human
language system itself and about its relationship with the non-linguistic systems of the
12 LANGUAGE, COGNITION AND SPACE

human brain. We need logically to decide what we mean by ‘language’. The prevailing
tendency is to focus on spoken language and spoken languages. However, there are two
ways (apart from written representations of the spoken) in which language exceeds
the boundaries of the spoken, namely by the use of gesture in conventionalised sign
languages where the spoken word is absent, and the use of gesture as an integrated
accompaniment of spoken languages themselves. Section VII of the volume consists
of two chapters that address these two additional aspects in the context of some far-
reaching questions concerning the nature of the human language ability.
Leonard Talmy’s contribution, Chapter 13, compares the ways in which American
Sign Language expresses spatial concepts with the way in which spoken languages
(principally American English) do so. This comparative approach is held to have deep
theoretical consequences bearing on the long standing issue of the existence and hypo-
thetical nature of a dedicated language module in the human brain. Like Evans, Talmy
distinguishes between closed-class language forms and open-class forms, the former
including many elements linked to spatial concepts. In signed languages, there is a
subsystem, known as ‘classifier expressions’, which deals specially with the location
or motion of objects with respect to one another. Talmy’s chapter compares spoken
closed-class spatial expressions with signed language classifier expressions. In order
to compare these two space-expressing systems in the two languages, Talmy outlines a
detailed theoretical framework that aims to provide the ‘fundamental space-structuring
elements and categories’ universally available in spoken languages. Sign language is then
compared with this set. Talmy’s claim for spoken languages is that all spatial expressions
in spoken languages are associated with conceptual schemas made up of combinations
of conceptual elements, organised in categories, and pertaining to spatial scenes. Most
of these categories appear to be mainly describable in geometric or force dynamic or
other physical terms, while non-geometric properties, such as affective state, figure less
prominently in this model. Moreover, the notion of ‘functional’ meaning is not invoked
in the way it is in the models of Vandeloise, and Evans, discussed above. Some of the
spatial schemas made up of the spatial elements are more basic than others and can
be extended by certain regular processes that, to a certain extent, resemble geometric
transformations in Talmy’s account. How do sign languages compare? What Talmy
finds is that the two language modalities, spoken and signed, share a ‘core’ in the general
design principles governing the spatial concepts for which they have forms. However,
sign language differs markedly, both qualitatively and quantitatively, and with a high
degree of iconicity. The most general claim is that sign language closely parallels the
processing of spatial scenes in visual perception. These findings lead to Talmy to propose
a new neural model for language. His hypothesis is that the common ground between
the two language modalities (including spatial expressions) results from a single neural
system that can be considered the fundamental language module of the human brain.
The properties of this module are highly abstract and are primarily concerned with
categorising and combinatorial principles. What is new is Talmy’s emphasis on the
linkage between the human language ability and the visual system, a linkage which is
crucial in the case of signed languages.
 INTRODUCTION 13

Chapter 14, by Irene Mittelberg, is complementary to the approach of the chapter by

Talmy in two respects. First, rather than considering universals, it examines data from
particular speech events representing a particular genre of discourse, namely academic
lectures on linguistics. Second, the way in which gesture is considered in this chapter
concerns the way in which space is used to communicate meaning, rather than the way
in which space is represented in communication. Mittelberg’s analytic descriptions
show that hand shapes and motion patterns recur across speakers. These hand shape
configurations, and traced patterns in the air, may constitute a kind of common sense
geometry. Many of these patterns also appear to be iconic visual representations of the
kinds of image schemas that are postulated in the cognitive linguistic literature. The
hand shapes and hand motions, which are finely synchronised with speech, are not all,
however, representations of objects perceived as having the shapes represented. In fact,
most of them co-occur with spoken lexical items associated with abstract meanings.
Even when these abstract items are not recognisably metaphorical (as many abstract
concepts often are), they are synchronised with concrete gesture patterns describ-
ing a concrete visual geometric shape. For example, in the lecture discourse studied,
the concept of ‘category’ may co-occur with a container-like cupping of the hands; a
‘subcategory’ may co-occur with the same gesture located relatively lower in the gesture
space relative to the body of the speaker. Or a relatively small space between the finger
tips may stand for a relatively small object that is being referred to in the accompanying
speech. A fundamental iconic principle is clearly at work and is apparently spontaneously
applied by both speaker and hearer. The relationship between the geometric shape of
the gesture and the spoken abstract concept is not simply visual, but also kinaesthetic. It
is important to note that embodied interaction with objects, specifically manipulation,
is what is involved. Moreover, metaphoricity as well as iconicity is at work, since the
iconically represented container concept and the iconically represented ‘sub’ relationship
are themselves metaphorically related to the abstract concept of subcategory. The analysis
of linguistic data alone has in the past been used to show that image schemas are used
in the understanding and coding of abstract concepts. The analysis of concrete gesture
data can be claimed to reinforce this claim.
The papers summarised so far predominantly concern spatial representations that
are static. Of course, in classical physics, nothing is static except in relation to some
frame of reference. That is, motion is relative. Now we have already commented on the
various attempts to deal with spatial concepts in terms of classical Euclidean geometry.
Similar issues arise when linguists turn to the concept of motion. Clearly, languages
contain expressions that are associated with the concepts of motion. But what exactly
are these concepts? Are they like the classical Newtonian laws? The two chapters in
Section VII include alternate, and to some extent competing, attempts to establish the
appropriate conceptual framework for the description and investigation of the ways
in which different languages encode concepts of movement through space. As Jordan
Zlatev and his co-authors point out, in Chapter 15, a well justified descriptive framework
is essential if we are to make headway with some of the intriguing questions that have
already emerged in work on the linguistic expression of motion.
14 LANGUAGE, COGNITION AND SPACE

These questions revolve around three related issues already mentioned as focal in
the study of space and language. Are there any universal tendencies in the linguistic
encoding of spatial motion? How do languages vary in this respect? And to what extent
does linguistic encoding affect, if at all, non-linguistic cognition? From the beginning
of the contemporary interest in this area, the linguistic perspective has had typological
implications. Some languages (like English and other Germanic languages) encode
the manner of motion in a main verb, while the direction of motion is expressed in a
‘satellite’ (e.g. prepositional phrase; cf. Talmy 2000). Other languages (e.g. Romance
languages) appear to prefer the converse. Thus while French has Jean traverse la rue
en courant, ‘John is crossing the street running’), English has John is running across the
street. Slobin’s empirical work (e.g. Slobin 1996) followed up the potential Whorfian
implications – does different linguistic encoding of direction and manner imply differ-
ent ways of non-linguistic cognizing? The chapter by Zlatev and associates takes on a
twofold challenge. First, they take issue with Talmy’s classification of motion events and
propose a new taxonomy. Secondly, they use their new taxonomy as a basis for exploring
the Whorfian questions experimentally. It is clear that human conceptual taxonomy is
not that of classical mechanics but a rather fine-grained conceptual characterisation
humanly relevant to motion that includes such contrasts as self-motion as opposed to
caused motion. They find that the binary typology alluded to above is inadequate to
capture the range of variation in the syntactic encoding of direction and manner of
motion. In a series of experiments the authors investigate possible effects of coding in
three languages (French, Swedish and Thai), finding that the results do not support a
strong Whorfian account that would claim different languages entail entirely different
views of the world. Indeed, they strongly suggest that the similarities between languages
in the relevant respect are greater than their differences.
Like Zlatev and colleagues, Stéphanie Pourcel is among the newer generation of
cognitive linguists probing and advancing the pioneering work of Talmy and Slobin. In
Chapter 16, Pourcel, again like Zlatev, offers a conceptual revision of motion categories
as well as experimental explorations with a neo-Whorfian angle. Her logical starting
point is the observation that hitherto experimental studies investigating the possible
impact of linguistic categories on non-linguistic cognition have tended to rely on motion
categories that are drawn from particular languages – a flaw that she considers a mani-
festation of ‘linguacentrism’. The first part of her paper is thus an attempt to provide a
theoretical typology of the cognitive domain of motion that is independent of language
and that can then be used as a basis for empirical work. The second part of the paper
then describes experiments designed to test the language-independent validity of the
hypothesized motion categories. As far as the typology is concerned, it is compatible to
an extent with Zlatev’s in its recognition of the considerable complexity of the conceptual
domain of motion. Among other features, motion types involve directionality, causality,
telicity and force dynamics. The most important part of Pourcel’s typology, however, is
probably her insistence on the way the ‘existential status’ of the Figure (the moving or
moved entity) constrains the manners of motion associated with it. In this perspective,
the conceptualisation of motion is determined by the animacy, agentivity and causal
capacity of the entity involved in a motion event. This makes it possible to allow for the
 INTRODUCTION 15

effect of conceptualising biological types and physical forces. Pourcel also notes blended
types of moving Figure constructed in, for example, fiction and myth. The empirical
investigations reported in the second half of the chapter, including earlier experiments
conducted with Anetta Kopecka, lend support to Pourcel’s major hypothesis that the
conceptualisation of the domain of motion is centred around the Figure schema rather
than ground, path, manner or causal motivations. The most general claim here is that a
figure-based typology of motion is universal, irrespective of the way particular languages
syntactically code motion and manner.
The final part of the volume, Section VIII, contains three papers that go beyond the
conceptualisations of spatial relations between physical objects. If spatial conceptualisa-
tion is somehow fundamental in the human mind, then should we not expect to find it
motivating conceptual domains that are not themselves to do with the physical domain?
This question has been behind the observations noted by psychologists and linguists at
least since the 1970s. (Interestingly, as Daniel Casasanto notes in his chapter, a similar
idea is to be found in an 1898 essay by Jules Lafargue, the son-in-law of Karl Marx.)
In Chapter 17, Casasanto takes up some of the key questions that we have seen recur-
ring throughout this book. In particular, Casasanto addresses the question of spatial
metaphors in language and their relationship to non-linguistic cognition, using spatial
metaphors for time as his ‘test bed’. Despite the abundance of linguistic observations
in the linguistics literature, we still need to know whether people think about time in
spatial terms even when they are not using language, thus whether there are purely
mental spatial metaphors for time. We need to know if mental metaphors are universal
and whether using a particular language to think or speak about time makes us think
in ways constrained by the way those languages encode time concepts. We also need to
know if any similarities between linguistic and non-linguistic metaphors for time are
simply that, similarities, or causally related. In addition to time metaphors, Casasanto
extends his empirical investigation of spatial metaphors to the experience of musical
pitch. These particular cases serve to help us understand how humans manage to come
up with abstract concepts in the first place, an ancient puzzle for the philosophy of
mind. If the answer is indeed that it is a capacity for metaphor that facilitates abstract
conceptualisation, then this result could be intriguingly consistent with the notion of
exaptation in evolutionary biology. According to Casasanto, the experimental evidence
points clearly both towards the use of spatial metaphors in thinking about time (but not
vice versa) and towards the influence of particular linguistic encoding on non-linguistic
spatial metaphors, and this not merely in a ‘thinking for speaking’ sense.
Chapter 18 also concerns the nature of spatial metaphors for time. As has been seen,
many investigators assume that frames of reference, which essentially are Cartesian
coordinate systems, provide the basis for the analysis of spatial cognition, whether
linguistically expressed or not. Jörg Zinken raises the question of whether frames of
reference are therefore also relevant to the description of the abstract conceptualisation
of time. He also asks, as Zlatev and Pourcel do for motion, whether the existing typolo-
gies for the concept of time are adequate for further cognitive-linguistic investigation.
The answer to this last question is that existing assumptions need to take into account
the rich anthropological literature concerning cultural variation in the understanding
16 LANGUAGE, COGNITION AND SPACE

of time, as well as two views of time that are commonly found in the philosophical
literature. These two views are, roughly, the experiencer-centred view of time (one event
comes to us after another and fades into the past), and the experiencer-independent
view, according to which events remain for ever strung out in a sequence. With respect
to the question whether frames of reference are relevant, Zinken argues that they are,
and shows how the three reference frame types formulated by Levinson (2003) have
consequences not just for spatial cognition and linguistic expression but also for spatial
metaphors for time. The upshot is a proposal for a new typology of temporal concepts
that combines reference frames and the distinction between experiencer-centred and
experiencer-independent time concepts. This more detailed framework is required,
in Zinken’s view, in order to make progress in the empirical exploration of space-time
metaphors across the world’s languages in relation to non-linguistic cognition. The
approach outlined in this chapter is not, however, entirely couched in terms that can be
characterised as broadly geometrical. While other contributors to the volume emphasise
the ‘functional’ factor in linguistically encoded conceptualisation of physical space,
Zinken focuses on the possible contribution of cultural factors. To what extent is the
English tendency to conceptualise immediate and distant future time as ‘in front of ’ the
speaker explained by a culture of forward planning and manipulation of events? Such
far-ranging questions point to new goals in cognitive-linguistic research.
The final chapter of the volume is a theoretical speculation concerning the possible
extension of spatial concepts to the description of more abstract aspects of language
structure, including grammatical constructions. In this short chapter I extend the notion
of reference frames, a notion that, as we have seen, emerges as fundamental in language-
and-space research. I take frames of reference to be Cartesian coordinate systems defining
a three dimensional space but I apply them not to physical space (or even a metaphorical
target domain such as time in Zinken’s paper) but to what I call the abstract ‘discourse
space’ (for more details see Chilton 2005). Of course, since Descartes, n-dimensional
spaces have been defined and explored extensively by mathematicians, but the three
Euclidean dimensions might be especially significant for humans, as pointed out at the
beginning of this Introduction. The model I propose is defined in three axes. These are:
discourse distance (essentially Figures are ‘closer’ to the speaker than Grounds); time
(some events in both past and future are ‘closer’ than others) and epistemic modality
(epistemically more certain events are ‘close’ to the speaker and counterfactual ones are
‘remote’). One further ingredient is added, namely simple vectors, which have distance
and direction. Consistently with a major component of the account of spatial prepositions,
the abstract axis systems can be transformed (cf. the ‘projection’ of Levinson’s ‘relative
frames’). Using geometrical diagrams for a large part of the argument, I suggest perhaps
surprising aspects of viewing certain syntactic and semantic phenomena in terms of
geometric transformation. This might appear to be pushing the geometric approach too
far, and I certainly do not wish to ignore the ‘functional’ components that are treated by
various authors in the present volume. However, the geometrical description of space
provides the essential scaffolding in all accounts of spatial expressions, as we have seen.
This is not surprising, if the notion of embodiment is taken seriously. And this is also
why my chapter, and indeed this whole volume, begins with a review of the grounding
of spatial perception and conception in biological systems.
 INTRODUCTION 17

References
Anderson, J. M. (1971) The Grammar of Case: Towards a Localist Theory. Cambridge:
Cambridge University Press.
Bloom, P., Peterson, M. A., Nadel, L. and Garrett, M. F. (eds) (1996) Language and
Space. Cambridge, MA: MIT Press.
Chilton, P. A. (2005) Vectors, viewpoint and viewpoint shift: Toward a discourse
space theory. Annual Review of Cognitive Linguistics, 3: 78–116.
Gruber, J.S. (1965/1976) Studies in Lexical Relations. PhD thesis MIT. Reprinted in
Studies in Lexical Relations. Amsterdam: North Holland.
Gumperz, J. J. and Levinson, S.C. (eds) (1996) Rethinking Linguistic Relativity.
Cambridge: Cambridge University Press.
Haspelmath, M. (2003) The geometry of grammatical meaning: Semantic maps
and cross-linguistic comparison. In M. Tomasello (ed.) The New Psychology of
Language (Vol. 2, pp. 211–242). Mahwah, NJ: Lawrence Erlbaum Associates.
Hurford, J. (2003a) The neural basis of predicate-argument structure. Behavioural
and Brain Sciences, 22(4): 261–283.
Hurford, J. (2003b) Ventral/dorsal, predicate/argument: The transformation from
perception to meaning. Behavioural and Brain Sciences, 26(3): 301–311.
Jackendoff, R. (1993) Semantics and Cognition. Cambridge, MA: MIT Press.
Jackendoff, R. (2002) Foundations of Language: Brain, Meaning, Grammar, Evolution.
Oxford: Oxford University Press.
Kant, I. (1781/1963) Critique of Pure Reason. (Second edition.) Translated by
M.K.Smith. London: Macmillan and Co.
Lakoff, G. and Johnson, M. (1980) Metaphors We Live By. Chicago: Chicago
University Press.
Landau, B. and Jackendoff, R. (1993) ‘What’ and ‘where’ in spatial language spatial
cognition. Behavioural and Brain Sciences, 16: 217–238.
Langacker, R. (1987) Foundations of Cognitive Grammar vol. I. Stanford: Stanford
University Press.
Langacker, R (1991) Foundations of Cognitive Grammar vol II. Stanford: Stanford
University Press.
Levinson, S. (1996) Language and space. Annual Review of Anthropology, 25:
353–382.
Levinson, S. (2003) Space in Language and Cognition. Cambridge: Cambridge
University Press.
Li, P. and Gleitman, L. (2002) Turning the tables: Language and spatial reasoning.
Cognition, 83(3): 265–294.
Lyons, J. (1977) Semantics. (2 vols.) Cambridge: Cambridge University Press.
O’Keefe, J. (1996) The spatial prepositions in English, vector grammar and the
cognitive map theory. In P. Bloom et al. (eds) Language and Space (pp. 277–316).
Cambridge, MA: MIT Press.
O’Keefe, J. (1999) Kant and the sea-horse: An essay in the neurophilosophy of space.
In N. Eilan, R. McCarthy and W. Brewer (eds) Spatial Representation: Problems in
Philosophy and Psychology. Oxford: Oxford University Press.
18 LANGUAGE, COGNITION AND SPACE

O’Keefe, J. (2003) Vector grammar, places, and the functional role of the spatial
prepositions in English. In E. van der Zee and J. Slack (eds) Representing
Direction in Language and Space. Oxford: Oxford University Press.
Slobin, D. I. (1996) From ‘thought and language’ to ‘thinking for speaking’. In J. J.
Gumperz and S. C. Levinson (eds) Rethinking Linguistic Relativity. Cambridge:
Cambridge University Press.
Talmy, L. (2000) Fictive motion in language and ‘ception’. In Toward a Cognitive
Semantics (Vol. I pp. 99–176). Cambridge, MA: MIT Press.
Traugott, E. C. (1975) Spatial expressions of tense and temporal sequencing.
Semiotica, 15: 207–230.
Van der Zee, E. and Slack, J. (eds) (2003) Representing Direction in Language and
Space. Oxford: Oxford University Press.
Whorf, B. L. (1939/2000) The relation of habitual thought and behavior to language.
In J. B. Carroll (ed.) Language, Thought and Reality: Selected Writings of Benjamin
Lee Whorf (pp. 134–159). Cambridge, MA: MIT Press.
Zwarts, J. (1997) Vectors as relative positions: A compositional semantics of modified
PPs. Journal of Semantics, 14: 57–58.
 Part I
Perception and space

19
1 The perceptual basis of spatial representation
Vyvyan Evans

Overview

The human experience of space includes knowledge relating to the size, shape, loca-
tion and distribution of entities in a stable three-dimensional environment. In this
introductory chapter I address the perceptual systems and processes that facilitate this:
the sense-perceptory and brain mechanisms that process perceptual information giving
rise to spatial experience. I also examine the processes whereby perceptual experience
is redescribed into rudimentary representations of space. That is, I examine primitive
concepts which form the bedrock of our ability to think, reason and talk about space
and, indeed, more abstract realms. Thus, this chapter is concerned primarily with i) the
perception of space, and the way in which spatial experience is ‘constructed’ by virtue
of our sense-perceptory systems and brain mechanisms, and ii) how spatial experience
is ‘redescribed’, giving rise to foundational spatial concepts prior to the emergence of
language from around one year onwards.
The chapter begins by examining the distinction between spatial representations,
exploring the difference between percepts and concepts. I then examine the human
perceptual systems which facilitate the detection of sensory stimuli from the external
environment. I then look at perceptual theories which attempt to explain how the
brain constructs spatial experience from this sensory input. I then turn to the human
mapping ability: an innate mechanism that allows us to construct spatial or cognitive
‘maps’ based on locational information. This ability is essential for wayfaring, which
is to say navigating in space. I then examine how percepts are redescribed as the basic
spatial primitives, known as image schemas.

1 Introduction: perception vs conception

My main concern in this chapter is to review the way in which space is experienced and
constructed by the human sensory (or sense-perceptory) systems, and the brain. I also
review the way in which these objects of spatial perception known as percepts give rise to
rudimentary spatial representations (or concepts) known as image schemas. Accordingly,
at this point I briefly review the distinction between perception (and percepts), and
conception (and concepts).
Perception consists of three stages: i) sensation ii) perceptual organisation and iii)
identification and recognition. Sensation concerns the way in which external energy, such
as light, heat, or (sound) vibrations are converted into the neural codes which the brain
recognises. Perceptual organisation concerns the way in which this sensory informa-

21
22 LANGUAGE, COGNITION AND SPACE

tion is organised and formed into a perceptual object, a percept. Identification and
recognition relates to the stage in the process whereby past experiences and conceptual
knowledge is brought to bear in order to interpret the percept. For instance, a spherical
object might be identified and recognised as a football or a coin, or a wheel, or some
other object. That is, this stage involves meaning, which is to say understanding the
nature, function and significance of the percept. As such, a previously-formed concept
is employed in order to identify and categorise the percept.
Table 1. Three stages in perception

Sensation external energy stimuli are detected and converted into neural codes
perceptual organisation integration of neural codes by the brain to form a percept

identification the percept is categorised, which involves matching with stored experiences
and recognition

The distinction between percepts and concepts relates to distinctions in representa-

tional formats: how experience is presented at the cognitive level and how it is stored.
Percepts constitute coherent representations which derive from sensory experience,
and arise from multiple modalities. That is, they derive from information which is
integrated from a number of different sensory systems, discussed in more detail in
the next section. Percepts are typically available to conscious experience. That is,
they are the product of on-line processing, resulting from a stimulus array perceived
in the ‘here-and-now’. A consequence of this is that they consist of specific informa-
tion relating to the specific stimulus array that they are derived from. Thus, they are
episodic in nature.
Concepts, on the other hand, represent schematisations, formed by abstracting
away points of differences in order to produce representations which generalise over
points of similarity. Thus, the concept car, for instance, is a schematisation derived
by generalising across many different sorts of specific (episodic) experiences relating
to automobiles in order to form a single representation. Of course, this greatly simpli-
fies things, and I emphasise that concepts, while stable schematisations are not static
and unchanging. Indeed, they continue to be updated and thus evolve as the human
perceiver continues to be exposed to new experiences. A consequence of the schematic
nature of concepts is that, unlike percepts, concepts are representations in the sense
of re-presentations. That is, they are stored in memory and can be activated during
off-line processing. That is, they can be recalled in the absence of the percept(s) which
may have given rise to them.
A further important point is that while percepts relate primarily to the sensory
details of a given entity, concepts include a much greater range of information types,
including the nature and function of the entity which is being represented, as well as
how it relates to other concepts. Thus, concepts are related to one another in a systematic
way, and form a structured knowledge ‘inventory’, what I will refer to as the human
conceptual system. Thus, concepts constitute ‘theories’ concerning a particular entity,
 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 23

and as such bring meaning to bear with respect to any given percept (for discussion
see Mandler 2004).
This said, how do percepts and concepts arise? Percepts arise from a process termed
scene analysis (e.g., Bregman 1990). Scene analysis is the process whereby the perceptual
stimulus array is segregated into coherent percepts. This is achieved by both bottom-up
processing and top-down processing.
Bottom-up processing relates to the processing and integration of perceptual
‘details’ that make up, for instance, object percepts, such as a vase or a ball. I will
consider two sorts of perceptual details later in the chapter which are termed textons
and geons. Top-down processing relates to the integration of perceptual information
which is guided by global principles. Such principles have been proposed, for instance
by Gestalt psychology, an important and influential movement that I will consider in
detail below.
Bottom-up and top-down processing cross-cut another important distinction which
relates to primitive segregation versus schema-based segregation. That is, scene analysis
proceeds by making use of both innate and learned constraints. Primitive segregation is
segregation of the stimulus array based on innate, which is to say, pre-given, primitives.
Such primitives, which include, for instance figure-ground segregation, discussed below,
derive from invariants in the stimulus array which have, through evolutionary processes
come to be ‘hard-wired’ in the human brain. In contrast, schema-based segregation
involves scene analysis which employs learned constraints.
Before concluding this section, it is necessary to briefly say something about the
relationship between spatial concepts and percepts. In fact, this is an issue I address in
greater detail when I present the work of developmental psychologist Jean Mandler later
in the chapter. However, for now I note that spatial concepts derive from, in the sense
of being ‘redescribed’ from, perceptual experience. This process, which Mandler refers
to as perceptual meaning analysis, uses spatial percepts as the basis for the formation
of rudimentary spatial concepts: image schemas. I will have more to say about these
basic spatial concepts later.

2 Sensory systems

In this section I review the mechanisms that facilitate the processing of energy signals
from the environment, the stimulus array, and how this information is detected by our
sensory systems, and processed. I begin by examining the sensory organs and systems
which serve as our windows on our spatial environment.

2.1 The visual system

The crucial organ for the visual system is the eye. The brain and the eye work together
to produce vision. Light enters the eye and is changed into nerve signals that travel
along the optic nerve to the brain. As light enters the eye it is brought into focus on
the rear surface of the eyeball. Light enters at the cornea (see Figure 1), which helps
24 LANGUAGE, COGNITION AND SPACE

to bend light directing it through the pupil: the small dark circle at the centre of your
eye. The amount of light that enters the pupil is controlled by the iris – often coloured
brown or blue and encircles the pupil – which expands or contracts making the iris
larger of smaller. Behind the pupil is a lens, a spherical body, bringing light waves into
focus on the retina, the rear of the eyeball. The retina consists of a thin layer of light
receptors known as photoreceptors. There are two kinds of photoreceptors: cones and
rods. Cones allow us to see in colour and provide our perception in daylight. Rods
facilitate vision under dim conditions and allow only black and white perception.
That part of the retina which is most sensitive is called the macula, and is responsible
for detailed central vision. The part of the macula which produces clearest vision is
the fovea. It is a tiny area densely packed with cone cells. Accordingly, when we look
ahead, light reflected from objects in our ‘line of sight’ is directed onto our fovea, and
objects occupying this area of the macula are perceived by virtue of what is termed
foveal vision. Objects at the edge of the visual field are perceived less clearly. Vision of
this kind is known as peripheral vision.

Figure 1. The eye

‘What’ and ‘where’ visual systems

The photoreceptor cells on the retina convert light energy into neural information.
However, this information from different parts of the retina is carried along two different
pathways or ‘streams’, connecting different parts of the visual cortex – that part of the
brain responsible for vision – and providing distinct sorts of information. The visual
cortex occupies about a third of the (cerebral) cortex, the outer layer of the cerebrum
(consisting of four lobes, see Figure 2).
 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 25

Parietal lobe Frontal lobe

Occipital lobe
Temporal lobe

Cerebellum

Figure 2. Diagram showing the four lobes of the cerebrum, and the cerebellum (The cebral cortex
is the outer layer of the cerebrum. Note: The brain is seen from the right side, the front of the brain,
above the eyes, is to the right.)

The visual cortex is divided into approximately thirty interconnected visual areas. The
first cortical visual area is known as the primary visual cortex or V1. V1 sends informa-
tion along two separate pathways or ‘streams’ through different parts of the visual cortex,
giving rise to two separate visual systems each providing different kinds of information
(Ungerleider and Mishkin 1982). The primary visual system, known as the focal system
sends information from the macula along the pathway known as the ventral stream
(ventral means ‘lower’). This system, often referred to as the ‘what’ system, provides
information relating to form recognition and object representation. That is, it allows
us to identify and recognise objects, including the recognition of attributes such as
colour, for instance.
The second system, known as the ambient system sends information from both
the macula and more peripheral locations on the retina along a pathway known as
the dorsal stream (dorsal means ‘upper’). This system, also known as the ‘where’
system, provides information relating to where an object is located in body-centred
space, rather than with details of the object itself. Thus, light signals in the eye are
transformed by the brain providing two distinct sorts of information relating to
‘what’ and ‘where’.
More recently Milner and Goodale (1995) have demonstrated that the distinction
between the two ‘streams’ does not strictly relate to the type of percept (‘what’ versus
‘where’) that visual processing provides, in the way conceived by Ungerleider and
Mishkin. Rather, while the ventral stream provides information that allows humans to
perceive particular objects (‘what’), the dorsal stream provides functional information
which facilitates readiness for action in order to interact with objects and other entities
in the world. In other words, the ventral stream provides information leading to the
conscious understanding of objects and other entities in the physical environment,
while the dorsal stream serves to facilitate motor programming.
Important evidence for these two distinct visual systems comes from the phenom-
enon known as blindsight. Some blind individuals appear to be able to localise and orient
to objects without actually being able to see them. In other words, some blind people
appear to be able to locate objects without knowing what the objects are, that is, without
26 LANGUAGE, COGNITION AND SPACE

being able to identify the object. This suggests that in such cases while the focal system
is damaged, the ambient system, mediated by the dorsal stream allows them to make
correct orientation judgments and responses, providing compelling evidence for two
distinct kinds of visual information.
Recent work on spatial representation in language suggests that the ‘what’ and
‘where’ systems may have linguistic reflexes. For instance, Landau and Jackendoff (1993)
argue that spatial relations, as encoded by prepositions, and objects as encoded by count
nouns roughly approximate the pre-linguistic representations deriving from the ‘where’
and ‘what’ systems respectively. Similarly, Hurford (2003) argues that the ‘where’ and
‘what’ systems provide neurological antecedents for predicate-argument structure in
language.

2.2 The vestibular system

The vestibular system, or orienting sense is the sensory system that provides information
relating to our sense of balance, and is the dominant system with respect to sensory
input about our movement and orientation in space. Together with the cochlea, the
auditory organ, discussed below, the vestibular system, is situated in the vestibulum in
the inner ear (Figure 3).
As our movements in space consist of rotations – circular motion, as when we turn
around – and translations – linear motion, as when we walk along a path (horizontal
motion), or climb a ladder (vertical motion or gravity) – the vestibular system comprises
two components. The first component consists of semicircular canals which detect
rotations. These are interconnected fluid-filled tubes which are located in three planes
at right angles to one another. The inner surface of the canals also contain hairs. As the
fluid moves in response to rotational movement the hairs detect motion of the fluid and
transduce this into neural code. The three distinct canals serve to provide rotational
information from three axes.
The second component consists of two fluid-filled sacs, the utricle and the saccule.
These chambers contain otoliths – literally ‘ear stones’ – which are heavier than the
fluid in the sacs and respond to linear and vertical motion, including both left-right,
forward-back motion and gravity (vertical motion). As before both the utricle and sac-
cule contain hairs which detect movement of the otoliths in response to linear motion.
This information is transduced into neural code which is transmitted to the brain for
processing.
The vestibular system sends signals primarily to the neural structures that control
our eye movements, and to the muscles that keep us upright. One important function
of the vestibular system is to coordinate body and head movement with the detection
of motion by the visual system. This is referred to as the vestibulo-ocular reflex (VOR),
which is necessary for vision to remain clear. This works during head movement by
producing an eye movement in the direction opposite to head movement, thus preserv-
ing the image on the centre of the visual field. For example, when the head moves to the
right, the eyes move to the left, and vice versa. Since slight head movements are present
all the time, the VOR is very important for stabilising vision.
 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 27

Figure 3. The vestibular system and cochlea

The vestibular system is, in phylogenetic (i.e., evolutionary) terms, one of the first systems
to have developed. In ontogenetic (i.e., developmental) terms it is the first to fully develop,
by six months after conception.

2.3 The auditory system

The vestibular system, and the key auditory organ, the cochlea, are closely linked,
both occupying the ear bone. It is widely believed that the cochlea evolved from the
phylogenetically earlier sensory structures responsible for detecting bodily orientation.
The auditory system works by transforming sensory information first from air to
fluid and then to electrical signals that are relayed to the brain. One important function
of the ear is to amplify sound vibrations, in preparation for the transformation from
air to fluid. The folds of cartilage that comprise the outer ear on the side of the head are
called the pinna (see Figure 4). The sound waves enter the ear canal, a simple tube which
starts to ampify the sound vibrations. At the far end of the ear canal is the eardrum
which marks the beginning of the middle ear.
The middle ear includes the ossicles – three very small bones shaped like a hammer,
an anvil, and a stirrup. The ossicles further amplify the sounds by converting the lower-
pressure eardrum sound vibrations into higher-pressure sound vibrations. Higher pres-
sure is necessary because the inner ear contains fluid rather than air. The signal in the
inner ear is then converted to neural code which travels up the auditory nerve.

Figure 4. Anatomy of the ear

28 LANGUAGE, COGNITION AND SPACE

The auditory nerve takes the neural code to that part of the brainstem known as the
cochlear nucleus. From the cochlear nucleus, auditory information is split into two
streams, similar to the way in which the visual signal is split into ‘where’ and ‘what’
streams. Auditory nerve fibres going to the ventral cochlear nucleus preserve the timing
of the auditory signal in the order of milliseconds. Minute differences in the timing of
signals received by both ears allow the brain to determine the direction of the sound.
The second, dorsal, stream analyses the quality of sound. It does this by virtue of
detecting differences in frequencies and thus allows differentiation of phonemes, such
as the distinction between set versus sat.

2.4 The haptic system

The haptic system includes the combined sensory input from the receptors for touch
in the skin and proprioception receptors in the body’s muscles and joints. Together
sense-perception from the haptic system gives rise to perceptual information from a
broad range of contact encounters between the body and environment that are sent to,
and processed by, a region of the cerebral cortex known as the somatosensory area. The
haptic system – deriving from hapsis which is Greek for ‘to grasp’ – provides perception
of geometric properties including the shape, dimension, and proportions of objects. It
also gives rise, through the proprioceptive receptors, to the felt sense of co-ordinated
movement, and thus is responsible, in part, for our perception of being distinct from the
environment which surrounds us. I review in more detail below the two key components
that make up the haptic system, the skin, and proprioception.

The skin

The skin is the largest organ, covering the entire body. It contains specialised nerve
endings which can be stimulated in different ways providing different sensations and
thus different sorts of sensory information. The sensory effect resulting from stimulation
of the skin is known as cutaneous sensitivity. There are three main cutaneous qualities:
pressure (also known as touch), temperature and pain. The somatesensory cortex in the
brain represents different skin regions as well as different cutaneous qualities. Thus, the
brain is provided with information relating to where on the skin a particular stimulus
is being received and what sort of quality is associated with it.
In terms of touch there is an important distinction to be made between active touch,
and passive touch. In active touch, the experiencer actively controls sensory stimulus
activation by virtue of picking up an object, for instance. By contrast, passive touch
occurs without the reception of the stimulus being controlled by the experiencer, as
when an object is placed in contact with the skin. Although the entire surface of the
skin responds to touch, the most sensitive receptors are the ‘exploratory’ parts of the
body. These include the fingers and hands, parts of the mouth and the tip of the tongue,
as well as the genitalia.
 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 29

Proprioception

Proprioception – from the Latin proprius which means ‘one’s own’ – relates to the sense
of body part position and movement. That is, it concerns the posture, location and
movement of the arms, legs and other parts of the human skeleton. Another commonly-
used term for proprioception is kinaesthesis – or kinaesthesia, from the Greek kineo, ‘to
move’. Proprioception is essential for a whole range of coordinated movements. To get
a sense of how it functions close your eyes and then touch your nose with a finger tip.
Your ability to do this comes from proprioception.
Proprioceptive receptors are known as mechanoreceptors. There are two types.
The first type provides sensory stimuli for joint information. The second provides
information deriving from mechanoreceptors founds in muscles and tendons. The
mechanoreceptors for joint information are stimulated by contact between the joint
surfaces. This occurs when the angles at which bones are held with respect to one another
change, due to movement. The mechanoreceptors in the muscles and tendons respond
to changes in the tension of muscle fibres when movement occurs.

3 Spatial perception: how we experience space

In this section I review the perception of objects, form, movement and three-dimensional
space. Perhaps unsurprisingly, given the importance of the visual modality for primates
in general and humans in particular, much of the work on various aspects of spatial
perception has traditionally focused on visual cues. Indeed, visual perception is perhaps
the best studied of the sensory systems. Accordingly, in this section I will primarily focus
on the role of visual perception in the experience and construction of spatial percepts.

3.1 Texture and object perception

Before objects can be identified visual details must be processed and integrated by the
visual system. Variations in visual scenes, in terms of i) light intensity, i.e., adjacent
regions of light and dark areas – known as contrast phenomena – ii) patterns and iii)
colour, form repeated patterns known as visual texture. The patterns, for instance, curly
versus straight hair, or a tiger’s stripes versus a leopard’s spots, are often the result of
the physical surface properties such as differentially oriented strands, and direction of
light and direction of motion.
One important bottom-up theory of visual texture perception is known as Feature
Integration theory. This theory assumes that there are two major stages involved in the
perception of visual texture. The first stage, known as the preattentive stage, involves the
unconscious processing of visual texture. In a seminal paper, psychologist Bela Julesz
(1981) proposed that the preattentive stage serves to process textural primitives, the
fundamental components of visual texture. These he labelled textons.
30 LANGUAGE, COGNITION AND SPACE

Textons are distinct and distinguishable characteristics of any given visual display.
For instance, textons include straight lines, line segments, curvature, widths, lengths,
intersections of lines, and so on. According to Julesz, the first stage of visual texture
perception involves discriminating between the range of textons in a visual display.
The second stage in visual texture perception is the focused attention stage. This
involves conscious processing in order to integrate the textons into complex unitary
objects.
Just as textons have been proposed as the primitive elements of visual texture
perception, a related bottom-up theory has been proposed to account for object
identification. This theory, associated with the work of Biederman (1987) is called
recognition by components. Biederman’s essential insight is that the identification of
objects involves the combination of a set of primitive three-dimensional geometric
components which he labels geons, short for ‘geometric icons’. Geons are simple
volumes such as cubes, spheres, cylinders, and wedges (see Figure 5). Biederman has
proposed 36 geons which can be combined in a range of ways giving rise to complex
objects. Biederman argues that object perception crucially relies upon recognising
the components which make up an object, the geons. Figure 6 illustrates how a
perceived object is comprised of a range of constituent geons. The image on the left
corresponds to the perceived object (a desk lamp), and the image on the right to the
constituent geons.

Figure 5. Some examples of geons (After Biederman 1987)

Figure 6. Geons in object perception

 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 31

3.2 Form perception

In the previous section I briefly looked at primitive elements that have been proposed for
textual perception and the identification of objects. However, in addition to identifiable
components of images and objects, there are also higher-level processes involved that are
essential for the perception of forms and the grouping of objects. Moreover, these appear
to be innate. I discuss two sorts of such organising principles below, figure-ground
segregation, and the Gestalt grouping principles.

Figure-ground perception

A fundamental way in which we segregate entities in our environment, thereby perceiv-

ing distinct objects and surfaces, comes from the our ability to perceive certain aspects
of any given spatial scene as ‘standing out’ from other parts of the scene. This is known
as figure-ground organisation.
The phenomenon of figure-ground organisation was pointed out by the Danish
psychologist Edgar Rubin in 1915. He observed that in visual perception we see parts
of a given spatial scene as being made up of well-defined objects, which ‘stand out’
from the background. That is, we see objects as three-dimensional entities which stand
out from the terrain in which they are located. For instance, in Figure 7, the image of
the lighthouse, the figure, stands out from the grey horizontal lines, the ground, as a
recognisable and distinct image.

Figure 7. Figure-ground segregation

Rubin proposed a number of perceptual differences between the figure and ground.
These are summarised in table 2.
32 LANGUAGE, COGNITION AND SPACE

Table 2. Distinctions between figure and ground

Figure Ground
Appears to be thing-like appears to be substance-like
a contour appears at edge of figure’s shape relatively formless
appears closer to the viewer, and in front of the ground appears further away and extends behind the figure
Appears more dominant less dominant
better remembered less well remembered
more associations with meaningful shapes suggests fewer associations with meaningful shapes

In addition, figure-ground perception appears to be innate. For instance, photographs

which lack depth, being two-dimensional surfaces, are perceived in three-dimensional
terms. That is, the figure-ground organisation associated with photographs is an illusion.
A particularly well-known illusion made famous by Rubin is the vase-profile illusion
(Figure 8).

Figure 8. The vase/profile illusion

The vase/profile illusion is an ambiguous figure-ground illusion. This is because it can be

perceived either as two black faces looking at each other, on a white background, or as
a white vase on a black background. In other words, it undergoes spontaneous reversal.
This illusion shows that perception is not solely determined by an image formed on
the retina. The spontaneous reversal illustrates the dynamic nature of the perceptual
processes. These processes illustrate that how the brain organises its visual environ-
ment depends on our innate ability to segregate images on the basis of figure-ground
organisation. As this image contains the same percentage of black and white, that part
of the image which is assigned the role of figure determines whether a vase or faces
are perceived.
Figure-ground organisation appears to be an evolutionary response to our physical
environment. Our visual system, for instance, has evolved in order to be able to perceive
three-dimensional objects as distinct from the surrounding terrain in which they are
embedded. Figure-ground organisation thus constitutes a hard-wired response to this
imperative.
 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 33

Gestalt grouping principles

Gestalt psychology was a movement which emerged in the first decades of the twentieth
century. Its primary concern, and those of its three leading proponents, the German
psychologists Max Wertheimer, Kurt Koffka and Wolfgang Köhler was to investigate
why some elements of the visual field form coherent figures, and others serve as the
ground. Gestalt is the German term for ‘form’, or ‘shape’ or ‘whole configuration’. The
Gestalt psychologists proposed a number of innate grouping principles that enable us
to perceive forms. Some of these, based on the work of Max Wertheimer (1923) are
presented below.

Principle of Proximity (or nearness)

This principle states that the elements in a scene which are closer together will be seen
as belonging together in a group. This is illustrated in Figure 9. The consequence of the
greater proximity or nearness of the dots on the vertical axis is that we perceive the dots
as being organised into columns rather than rows.

Figure 9. Column of dots

If the scene is altered so that the dots are closer together on the horizontal axis, then
we perceive a series of rows, as illustrated in Figure 10.

Figure 10. Rows of dots

Principle of Similarity

This principle states that entities which share visual characteristics such as size, shape
or colour will be perceived as belonging together in a group. For example, in Figure 11,
we perceive columns of shapes (rather than rows). In fact, the shapes are equidistant on
34 LANGUAGE, COGNITION AND SPACE

both the horizontal and vertical axes. It is due to our innate predisposition to organise
based, here, on similarity that similar shapes (squares or circles) are grouped together
and, consequently, are perceived as columns.

Figure 11. Columns of shapes

Principle of Closure

This principle holds that incomplete figures are often ‘completed’, even when part of the
perceptual information is missing. For instance, in Figure 12 we perceive a circle, even
though the ‘circle’ is incomplete. That is, there is a tendency to close simple figures, by
extrapolating from information which is present.

Figure 12. An incomplete figure subject to perceptual closure

A related perceptual process is illustrated by the following. In Figure 13, a white triangle
is perceived as being overlaid on three black circles, even though the image could simply
represent three incomplete circles. This phenomenon is known as the perception of
subjective or apparent contours. It resembles closure, in so far as there is the appearance
of edges across a blank area of the visual field.

Figure 13. Subjective contour: A white triangle

 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 35

Principle of Good Continuation

This principle states that human perception has a preference for continuous figures.
This is illustrated in Figure 14. Here, we perceive two unbroken rectangles, one passing
behind another, even though this is not what we actually see. In fact, the shaded rectangle
is obscured by the first, so we have no direct evidence that the shaded area represents
one continuous rectangle rather than two separate ones.

Figure 14. Two rectangles

Principle of Smallness

The Principle of Smallness states that smaller entities tend to be more readily perceived
as figures than larger entities. This is illustrated in Figure 15. We are more likely to
perceive a black cross than a white cross, because the black shading occupies a smaller
proportion of the image.

Figure 15. A black cross

Principle of common fate

The final principle I consider here is the Principle of Common Fate. This states that
elements that move in the same direction are perceived as being related to one another.
For instance, assume that we have two rows of 4 small squares. If the middle two squares
from the bottom row begin to move down the page, as depicted by the arrows in Figure
16, they are perceived as belonging together and thus form a separate group from those
that remain stationary.
36 LANGUAGE, COGNITION AND SPACE

Figure 16. Motion in the same direction

The Gestalt grouping principles I have surveyed conform to the general Gestalt Principle
of Good figure, also known as the Law of Prägnanz. This states that we tend to perceive
the simplest and most stable of the various perceptual possibilities.
While I have primarily focused in this section on visual perception, it is important
to emphasise that the principles I have discussed, both figure-ground and grouping
principles, manifest themselves in other modalities. For instance, Kennedy (1983;
Kennedy and Domander 1984), present evidence that figure-ground perception, includ-
ing the analogues to the ambiguous profile/vase illusion occur in the tactile (touch)
modality, based on experiments involving raised-line drawings of reversible figures.
Similarly, Bregman (1990) has argued that the Gestalt principles apply equally to
auditory scene analysis. He makes the point, for instance, that the ability to perceive
a series of musical notes as forming a tune is an instance of a gestalt par excellence.

3.3 The perception of movement

Our ability to detect movement is essential for the survival of the species. Below I
discuss a number of different systems for the detection of motion, and different kinds
of motion. I begin with the visual detection of motion.

Two visual systems

Motion detection appears to have evolutionary priority over shape detection (Gregory
1998). Indeed, as observed by Gregory, the evolution of the eye emerged in the first place
in order to detect motion. Indeed, only eyes relatively high up the evolutionary scale
produce stimulus in the absence of motion. The evolutionary development of vision
and the detection of motion are represented in the human eye:

The edge of our retinas are sensitive only to movement. You can see this by getting
someone to wave an object around at the side of your visual field where only the
edge of the retina is stimulated. Movement is seen, but it is impossible to identify the
object, and there is no colour. When movement stops the object becomes invisible.
This is as close as we can come to experiencing primitive vision. The extreme edge of
the retina is even more primitive: when it is stimulated by movement we experience
nothing; but a reflex is initiated, rotating the eye to bring the moving object into
central vision… (Gregory 1998: 98)
 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 37

The human visual system involves eyes which can move in the head, as when we keep
our heads stationary and move our eyes from side to side or up and down. Consequently,
our visual system has two distinct ways of detecting motion.
The first involves image-retina movement. This involves the eye ball remaining
stationary. In this situation the image of moving objects run sequentially across
adjacent photoreceptors on the retina. That is, the detection of movement occurs as
different photoreceptors are understood by the brain as relating to different locations
in space. The second method involves eye-head movement. This relates to movement
of the eyes in the eye-ball socket when we follow an object in motion. In this situation,
an object is not running across different photoreceptors as the eye moves in order to
track the object. Rather, information from the eye muscles, which stretch in response
to the movement of the eye, is understood by the brain as relating to motion of the
tracked object.

Optic flow

In normal vision when we move our eyes, the world remains stable. That is the visual
world doesn’t spin around. This follows as, during normal eye movements, signals from
the image-retina and eye-head systems cancel each other out, such that the world is
perceived as stable. While the two visual systems just described relate to the detection
of the movement of objects, another source of movement detection comes from the way
in which the human experiencer moves about during the world. As we move around
the location from which we view our environment changes. The consequence of this is
that there is a continuous change in the light stimulus which is projected on the retina.
Following the pioneering work of psychologist James Gibson (e.g., 1986), this changing
stimulus is known as optic flow.
Optic flow relates to a radial pattern which specifies the observer’s direction of
self-motion and is essential for successful navigation through the environment. As we
travel through the world, and as we approach objects, they appear to move towards
us, flowing past behind us as we move beyond them. Moreover, different objects at
different points in the visual field appear to move towards and past us at different rates.
For instance, imagine sitting on a train and travelling through the countryside. Distant
objects such as clouds or mountains appear to move so slowly that they are stationary.
Closer objects such as trees appear to move more quickly while very close objects appear
to whiz by in a blur. This motion, the optic flow pattern, provides important cues as
to distance. Moreover, the optic flow varies depending on the relationship between
viewing angle and direction of travel. For instance, objects which are dead-ahead and
thus centred in the visual field will appear to remain stationary, while objects which are
more peripheral in the visual field will appear to move more rapidly. However, because
the edges of centred objects will not be in foveal vision, the edges will have optic flow
associated with them. Thus, optic flow patterns provide important information about
both distance and direction of travel.
38 LANGUAGE, COGNITION AND SPACE

Biological motion

The requirement of being able to rapidly detect the motor activities of humans and
other organisms is essential for survival. Indeed, under certain lighting conditions,
such as at dusk, details relating to the precise nature of the animal in question may not
be readily discernable, especially if the animal is distant. Accordingly, humans have
evolved an ability to detect what Johansson (1973) terms biological motion. Based purely
on movement cues, we can quickly distinguish biological from non-biological motion.
Moreover, humans can readily distinguish between different types of biological motion
based solely on movement cues, for example, running versus jogging versus walking
versus jumping, and so on. Each gait represents a gestalt constructed from a sequence
of pendulum-like motions, specific to each activity type.
Evidence for this ability comes from the work of the visual psychologist Gunnar
Johansson. He videotaped actors in complete darkness. The actors had point-light dis-
plays (points of light) fixed at ten main body joints which served as the only illumination.
This eliminated all non-movement cues, such as the body contours of the actors. Subjects
were then asked to identify biological motion and the motor activities engaged in by
the actors. Johansson found that in the absence of motion subjects failed to recognise
the point light displays as representing a human form. However, with movement the
subjects vividly perceived human motion. In other words, subjects related moving lights
in order to perceive human movement, and moreover, were able to identify the pattern
of movement, that is, the kind of movement being engaged in.

3.4 The perception of three-dimensional space

In this section I briefly review how the brain constructs (three dimensional) space, that
is, depth, when the retina is a two-dimensional surface. In other words, where does the
third dimension come from? I consider below a number of cues that the brain extracts
from the visual stimuli in order to construct our experience of (three-dimensional)
space.
While depth and distance can be constructed on the basis of a range of visual (and
other) stimuli, including auditory cues, and the optic flow patterns described above,
an important means of obtaining depth information comes from binocular cues. This
relates to the spatial stimuli provided by virtue of having two eyes.
The eyes are separated by about 6.5 cm (Gregory 1998). The consequence of this
is that each eye sees a different view. As Gregory observes, ‘[t]his can be seen clearly if
each eye is closed alternately. Any near object will appear to shift sideways in relation
to more distant objects and to rotate slightly when each eye receives its view.’ (Ibid.: 60).
The difference between the two retinal images is known as binocular disparity, and gives
rise to the perception of depth or stereoscopic vision. However, stereoscopic vision only
applies to objects which are quite near. This follows as binocular disparity reduces the
further away an object is. As Gregory notes, ‘[w]e are effectively one-eyed for objects
 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 39

further than about 100 metres.’ (Ibid.: 60). In other words, depth is a consequence of
binocular rather than monocular (one-eyed) vision.

4 Cognitive maps

In this section I review in more detail the sorts of spatial representations that the brain
constructs from the sensory systems and perceptual stimuli described in previous sec-
tions. While I have examined distinct sensory systems, in practice perceptual informa-
tion from a range of modalities is integrated in order to form spatial or cognitive maps.
These are complex mental representations which facilitate navigation and moreover,
are necessary for the emergence of the concepts of place and location. The concepts of
place and location are independent of the entities and objects which occupy specific
places or locations. That is, without a cognitive mapping ability which allows us to
perceive places and locations independent of the objects which occupy them we would
have no means of understanding these concepts. Accordingly, the concepts place and
location are a consequence not of such notions being an inherent aspect of an objective
reality, but rather derive from innate cognitive mapping abilities, and particularly our
ability to construct spatial maps independently of our egocentric spatial location, as
discussed below.

4.1 Egocentric versus allocentric representations

There are two main sorts of spatial cognitive reference frames manifested by humans and
many other species. These are egocentric representations and allocentric representations.
In this section I briefly introduce cognitive reference frames of both these sorts.
There is good neurobiological evidence that humans, along with other mammals,
maintain multimodal cognitive spatial ‘maps’ in the parietal cortex (recall Figure 2). The
distinguishing feature of egocentric ‘maps’ is that they represent objects in space with
respect to the organism, or part of the organism, such as the organism’s hand, body or
head. This follows as cognitive ‘maps’ of this kind represent space in topographic fashion.
That is, neighbouring areas of neural space represent neighbouring regions of space
in the world of the perceiving organism, with respect to the organism which serves as
reference point or deictic centre for organising the location of the represented objects
and regions of space. As regions of space are organised with respect to the organism,
spatial maps of this kind are termed egocentric representations.
In addition, there is a second kind of spatial representation which is allocentric (or
other-focused) in nature. These representations, which are more appropriately thought
of in terms of maps (for reasons I shall discuss below), integrate information derived
from the egocentric spatial representations. Crucially, however, the allocentric mapping
ability represents space, and spatial regions independently of the momentary location
of the organism. That is, entities and objects, and the locations of objects are related to
one another independently of the ego. This system, which is located in the hippocampal
40 LANGUAGE, COGNITION AND SPACE

region of the brain (O’Keefe and Nadel 1978) represents place, direction and distance
information, rather than object details.

4.2 The hippocampus and the human cognitive mapping ability

In now classic work, neurobiologists John O’Keefe and Lynn Nadel (1978) show not only
that i) humans have an objective or absolutive spatial framework in which the entities
of our experience are located, but also that, ii) this ability is innate, and along with
other mammals is associated with the brain region often implicated in motor function:
the hippocampus. According to O’Keefe and Nadel, this allocentric mapping system
provides ‘the basis for an integrated model of the environment. This system underlies
the notion of absolute, unitary space, which is a non-centred stationary framework
through which the organism and its egocentric spaces move.’ (Ibid.: 2). This hippocampal
mapping system consists of two major subsystems, a place system and a misplace system.
The place subsystem is a memory system that allows the organism to represent
places in its environment and crucially to relate different locations with respect to each
other. That is, the place system allows the organism to represent relationships between
different locations without having to physically experience the spatial relations hold-
ing between distinct places. In other words, humans, like many other organisms, can
compute distances, and other spatial relations between distinct places such as directions,
without having to physically experience the spatial relationships in question. Such a
cognitive mapping ability is a consequence of the allocentric place subsystem.
The second subsystem to make up the allocentric cognitive mapping ability, the
misplace system, facilitates and responds to exploration. That is, it allows new informa-
tion experienced as a consequence of exploration to be incorporated into the allocentric
map of the organism’s environment. It thereby allows the organism to relate specific
objects and entities to specific locations, and to update the cognitive map held in the
place system based on particular inputs (cues) and outputs (responses). Thus, O’Keefe
and Nadel demonstrate two things. Firstly, three-dimensional Euclidean space is, in a
non-trivial sense, imposed on perceptual experience by the human mind. Secondly,
the notion of all-embracing continuous space, ‘out there’, which ‘contains’ objects and
other entities, as maintained by the misplace system, is in fact a consequence of first
being able to represent locations in an allocentric (i.e., a non-egocentric) fashion, as
captured by the place subsystem. In other words, our innate ability to form absolutive
cognitive maps of our spatial environment is a prerequisite to experiencing objects and
the motions they undergo.

4.3 Maps versus routes

In order to illustrate the distinction between egocentric and allocentric spatial mapping
abilities, O’Keefe and Nadel provide an analogy which I briefly discuss here. The analogy
relates to the geographic distinction between routes versus maps. In geographic terms,
 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 41

a route constitutes a set of instructions which directs attention to particular objects in

egocentric space. That is, routes are inflexible, identifying landmarks in order to guide
the traveller, and thus do not allow the traveller freedom of choice. Put another way,
routes are guide-post based. Moreover, routes are goal-oriented, focused on facilitating
travel from a specific, pre-specified location to another. In this, routes correspond to
egocentric cognitive representations.
In contrast, maps are, in geographic terms, representations of part of space. A map
is constituted of places, and the places which the map represents are systematically
connected and thus related to each other. Moreover, and crucially, the places captured by
the map are not defined in terms of the objects which may occupy a particular location.
That is, and unlike routes, maps are not guide-post based. Thus, maps capture space that
is held to exist independently of the objects which may be located at particular points
in space. Crucially, a map is a flexible representation, which can be used for a range
of purposes. In related fashion, this notion of a map is presented as an analogy of the
allocentric cognitive mapping ability that many organisms, including humans, possess.
While then map-like representations of the environment are constructed by humans,
as well as by other species, it is far from clear, in neurological terms, what the nature
of these representations are. Nevertheless, it is by now well established that humans do
possess complex information structures which can be used to generate highly-detailed
map-like representations, which can be used for a range of behaviours. Indeed, an impor-
tant finding to have emerged is that place memory has a high information capacity, and
can be permanently modified by a single experience. Moreover, experiments reported
on by O’Keefe and Nadel reveal that this mapping ability can be used to construct maps
in a highly flexible and efficient manner.
Finally, I reiterate that the ability to represent space in an allocentric fashion, i.e.,
map-like representations, is a trait common to a wide variety of organisms. As O’Keefe
and Nadel observe, ‘The ability of many animals to find their way back to their nests
over large distances would appear to be based on some type of mapping system’ (Ibid.:
63). Obvious examples include the migratory and homing behaviour exhibited by many
kinds of birds. Indeed, a robust finding from studies on homing pigeons is that they are
able to find their way ‘home’ using novel routes from new release sites. Such abilities
would appear to require a cognitive mapping ability.

5 Primitive spatial concepts

In this section I turn to an examination of spatial concepts and the way in which spatial
concepts are derived (or redescribed) from spatial experience. I focus here on the
notion of the image schema. Image schemas were first proposed by cognitive linguists
(e.g., Johnson 1987, 2007; Lakoff 1987; see Evans and Green 2006 for a review), and
represent a rudimentary conceptual building block derived from embodied experience
(discussed further below). This notion has been subsequently adopted by a range of
other cognitive scientists in their work (see papers and references in Hampe 2005). In
particular, the notion of the image schema has been developed in the influential work
42 LANGUAGE, COGNITION AND SPACE

of developmental psychologist Jean Mandler (e.g., 2004) in her work on how conceptual
development takes place.

5.1 Embodiment and experience

I begin this brief overview of the image schema by first introducing the role of embodi-
ment in the formation of concepts. Due to the nature of our bodies, including our
neuro-anatomical architecture, we have a species-specific view of the world. In other
words, our construal of ‘reality’ is mediated, in large measure, by the nature of our
embodiment. One obvious example of the way in which embodiment affects the nature
of experience relates to biological morphology (i.e., body parts). This, together with the
nature of the physical environment with which we interact, determines other aspects
of our experience. For instance, while gravity is an objective feature of the world, our
experience of gravity is determined by our bodies and by the ecological niche we have
adapted to. For instance, hummingbirds – which can flap their wings up to fifty times
per second – respond to gravity in a very different way from humans. They are able to
rise directly into the air without pushing off from the ground, due to the rapid movement
of their wings.
The fact that our experience is embodied – that is, structured in part by the nature
of the bodies we have and by our neurological organisation – has consequences for
cognition. In other words, the concepts we have access to and the nature of the ‘reality’
we think and talk about are a function of our embodiment – the phenomenon of variable
embodiment. That is, we can only talk about what we can perceive and conceive, and the
things that we can perceive and conceive derive from embodied experience. From this
point of view, the human mind must bear the imprint of embodied experience. This
thesis is known as the thesis of embodied cognition. This position holds that conceptual
structure – the nature of human concepts – is a consequence of the nature of our
embodiment and thus is embodied.

5.2 Image schemas

The theoretical construct of the image schema was developed by Mark Johnson in his
now classic 1987 book, The Body in the Mind. Johnson proposed that one way in which
embodied experience manifests itself at the cognitive level is in terms of image schemas.
These are rudimentary concepts like contact, container and balance, which are
meaningful because they derive from and are linked to human pre-conceptual experience.
This is experience of the world directly mediated and structured by the human body.
The term ‘image’ in ‘image schema’ is equivalent to the use of this term in psy-
chology, where imagistic experience relates to and derives from our experience of the
external world. Another term for this type of experience is sensory experience, because
it comes from sensory-perceptual mechanisms that include, but are not restricted to,
the visual system.
 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 43

According to Johnson (1987) there are a number of properties associated with image
schemas which I briefly review below.

Image schemas are pre-conceptual in origin

Image schemas such as the container schema are directly grounded in embodied
experience. This means that they are pre-conceptual in origin. Mandler (2004) argues,
discussed further in the next section, that they arise from sensory experiences in the
early stages of human development that precede the formation of concepts. However,
once the recurrent patterns of sensory information have been extracted and stored as
an image schema, sensory experience gives rise to a conceptual representation. This
means that image schemas are concepts, but of a special kind: they are the foundations
of the conceptual system, because they are the first concepts to emerge in the human
mind, and precisely because they relate to sensory-perceptual experience, they are
particularly schematic. Johnson argues that image schemas are so fundamental to our
way of thinking that we are not consciously aware of them: we take our awareness of
what it means to be a physical being in a physical world very much for granted because
we acquire this knowledge so early in life, certainly before the emergence of language.

Image schemas form the basis of word senses

Concepts lexicalised by words such as prepositions, for instance, in, into, out, out of and
out from are all thought to relate to the container schema: an abstract image schematic
concept that underlies all these much more specific senses – the semantic pole associated
with lexical forms (see Tyler and Evans 2003).
The container image schema is diagrammed in Figure 17. This image schema
consists of the structural elements interior, boundary and exterior: these are the mini-
mum requirements for a container (Lakoff 1987). The landmark (LM), represented
by the circle, consists of two structural elements, the interior – the area within the
boundary – and the boundary itself. The exterior is the area outside the landmark,
contained within the square. The container is represented as the landmark because the
boundary and the exterior together possess sufficient Gestalt properties (e.g., closure and
continuity) to make it the figure, while the exterior is the ground (recall my discussion
of Gestalt principles above).

LM

Figure 17. CONTAINER image schema

44 LANGUAGE, COGNITION AND SPACE

Although Figure 17 represents the basic container schema, there are a number of
other image schemas that are related to this schema, which give rise to distinct concepts
related to containment. For instance, let’s consider one variant of the container schema
lexicalised by out. This image schema is diagrammed in Figure 18 and is illustrated
with a linguistic example. The diagram in Figure 18 corresponds to example (1). The
trajector (TR) Fred, which is the entity that undergoes motion, moves from a position
inside the LM to occupy a location outside the LM. The terms ‘TR’ and ‘LM’ derive
from the work of Langacker (e.g., 1987), and relate to the Gestalt notions of figure and
ground respectively.

(1) Fred went out of the room

LM
TR

Figure 18: Image schema for OUT

The image schema shown in Figure18 represents a concept that is more specific and
detailed than the image schema diagrammed in Figure 17, because it involves motion
as well as containment. This shows that image schemas can possess varying degrees
of schematicity, where more specific image schemas arise from more fundamental or
schematic ones.

Image schemas derive from interaction

As image schemas derive from embodied experience, they derive from the way in
which we interact with the world. To illustrate this idea, consider the image schema for
force. This image schema arises from our experience of acting upon other entities, or
being acted upon by other entities, resulting in the transfer of motion energy. Johnson
illustrates the interactional derivation of this image schema – how it arises from experi-
ence – as follows:

[F]orce is always experienced through interaction. We become aware of force as it

affects us or some object in our perceptual field. When you enter an unfamiliar dark
room and bump into the edge of the table, you are experiencing the interactional
character of force. When you eat too much the ingested food presses outwards on
your taughtly stretched stomach. There is no schema for force that does not involve
interaction or potential interaction. (Johnson 1987: 43)
 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 45

Image schemas are inherently meaningful

As image schemas derive from interaction with the world, they are inherently meaning-
ful. Embodied experience is inherently meaningful in the sense that embodied experi-
ences have predictable consequences. To illustrate, imagine a cup of coffee in your hand.
If you move the cup slowly up and down, or from side to side, you expect the coffee
to move with it. This is because a consequence of containment, given that it is defined
by boundaries, is that it constrains the location of any entity within these boundaries.
In other words, the cup exerts force-dynamic control over the coffee. This kind of
knowledge, which we take for granted, is acquired as a consequence of our interaction
with our physical environment. For example, walking across a room holding a cup of
coffee without spilling it actually involves highly sophisticated motor control that we also
acquire from experience. This experience gives rise to knowledge structures that enable
us to make predictions: if we tip the coffee cup upside-down, the coffee will pour out.

Image schemas are analogue representations

Image schemas are analogue representations deriving from experience. The term ‘ana-
logue’ means image schemas take a form in the conceptual system that mirrors the
sensory experience being represented. Because image schemas derive from sensory
experience, they are represented as summaries of perceptual states, which are recorded
in memory. However, what makes them conceptual rather than purely perceptual in
nature is that they give rise to concepts that are consciously accessible (Mandler 2004).
In other words, image schemas structure (more complex) lexical concepts.

Image schemas can be internally complex

Image schemas are often, perhaps typically, comprised of more complex aspects that can
be analysed separately. For example, the container schema is a concept that consists of
interior, boundary and exterior elements. Another example of a complex image schema
is the source-path-goal or simply path schema. Because a path is a means of moving
from one location to another, it consists of a starting point or source, a destination or
goal and a series of contiguous locations in between, which relate the source and goal.
Like all complex image schemas, the path schema constitutes an experiential gestalt: it
has internal structure, but emerges as a coherent whole.
One consequence of internal complexity is that different components of the path
schema can be referred to. This is illustrated in example (2), where the relevant linguistic
units are bracketed. In each of these examples, different components of the path are
profiled by the use of different lexical items.

(2) a. SOURCE
John left [England]
b. GOAL
John travelled [to France]
46 LANGUAGE, COGNITION AND SPACE

c. SOURCE-GOAL
John travelled [from England] [to France]
d. PATH-GOAL
John travelled [through the Chunnel] [to France]
e. SOURCE-PATH-GOAL
John travelled [from England] [through the Chunnel] [to France]

Image schemas are not mental images

If you close your eyes and imagine the face of your mother or father, partner or lover,
what results is a mental image. Image schemas are not the same as mental images. Mental
images are detailed, and result from an effortful and partly conscious cognitive process
that involves recalling visual memory. Image schemas are schematic, and therefore
more abstract in nature, emerging from ongoing embodied experience. This means that
you can’t close your eyes and ‘think up’ an image schema in the same way that you can
‘think up’ the sight of someone’s face or the feeling of a particular object in your hand.

Image schemas are multi-modal

Image schemas derive from experiences across different modalities (different types of
sensory experience), and hence are not specific to a particular sense. In other words,
image schemas are abstract patterns arising from a range of perceptual experiences, and
as such are not available to conscious introspection. For instance, blind people have
access to image schemas for containers, paths, and so on, precisely because the kinds
of experiences that give rise to these image schemas rely on a range of sensory-perceptual
experiences in addition to vision, including hearing, touch, and our experience of
movement and balance.

Image schemas form the basis for abstract thought

Lakoff (1987, 1990, 1993) and Johnson (1987) have argued that rudimentary embodied
concepts of this kind provide the conceptual building blocks for more complex concepts,
and can be systematically extended to provide more abstract concepts and conceptual
domains with structure. According to this view, the reason we can talk about being in
states like love or trouble (3) is because abstract concepts like love are structured and
therefore understood by virtue of the fundamental concept container. In this way,
image schematic concepts serve to structure more complex concepts and ideas.

(3) a. John is in love.

b. Jane is in trouble.
c. The government is in a deep crisis.
 THE PERCEPTUAL BASIS OF SPATIAL REPRESENTATION 47

According to Johnson, it is precisely because containers constrain activity that it makes

sense to conceptualise power and all-encompassing states like love or crisis in terms
of containment.

5.3 Perceptual meaning analysis

The developmental psychologist Jean Mandler (e.g. 1992, 1996, 2004) has made a number
of proposals concerning how image schemas might arise from embodied experience.
Starting at an early age infants attend to objects and spatial displays in their environment.
Mandler suggests that by attending closely to such spatial experiences, children are able
to abstract across similar kinds of experiences, finding meaningful patterns in the proc-
ess. For instance, the container image schema is more than simply a spatio-geometric
representation. It is a ‘theory’ about a particular kind of configuration in which one
entity is supported by another entity that contains it. In other words, the CONTAINER
schema is meaningful because containers are meaningful in our everyday experience.
Mandler (2004) describes the process of forming image schemas in terms of a
redescription of spatial experience via a process she labels perceptual meaning analysis
(Mandler 2004). This process results from children associating functional consequences
with spatial displays. That is, image schemas emerge by virtue of analysing spatial
displays of various sorts as relating to the functional consequences with which they are
correlated. For example, we saw above that a consequence of coffee being located in a
coffee cup is that the coffee moves with the cup. That is, containment has functional
consequences in terms of containing, supporting and constraining the location of the
entity contained. Thus, the distinction between percepts and concepts such as image
schemas is that image schemas encode functional information, that is meaning. As
Mandler observes, ‘[O]ne of the foundations of the conceptualizing capacity is the image
schema, in which spatial structure is mapped into conceptual structure’ (Mandler 1992:
591). She further suggests that ‘Basic, recurrent experiences with the world form the
bedrock of the child’s semantic architecture, which is already established well before the
child begins producing language’ (Mandler 1992: 597). In other words, it is experience,
meaningful to us by virtue of our embodiment, that forms the basis of many of our
most fundamental concepts.

References
Biederman, Irving. (1987) Recognition-by-components: A theory of human image
understanding. Psychological Review 94: 115–147.
Bregman, Albert. (1990) Auditory Scene Analysis. Cambridge, MA: MIT Press.
Evans, Vyvyan and Melanie Green. (2006) Cognitive Linguistics: An Introduction.
Edinburgh: Edinburgh University Press.
Gibson, James. (1986) The Ecological Approach to Visual Perception. Mahwah, NJ:
Lawrence Erlbaum.
Milner, David and Melvyn Goodale. (1995) The Visual Brain in Action. Oxford:
Oxford University Press.
48 LANGUAGE, COGNITION AND SPACE

Gregory, Richard. (1998) Eye and Brain. (Fifth edition.) Oxford University Press.
Hampe, Beate. (ed.) (2005) From Perception to Meaning: Image Schemas in Cognitive
Linguistics. Berlin: Mouton de Gruyter.
Hurford, James. (2003) The neural basis of predicate-argument structure. Behavioral
and Brain Sciences 26: 261–283.
Johansson, Gunnar. (1973) Visual perception of biological motion and a model for its
analysis. Perception and Psychophysics 14: 201–211.
Johnson, Mark. (1987) The Body in the Mind. Chicago: University of Chicago Press.
Johnson, Mark. (2007) The Meaning of the Body. Chicago: University of Chicago Press.
Julesz, Bela. (1981) Textons: The elements of texture perception and their interac-
tions. Nature 290: 91–97.
Kennedy, John. (1983) What can we learn about pictures from the blind? American
Scientist 71: 19–26.
Kennedy, John and Ramona Domander. (1984) Pictorial foreground-background
reversal reduces tactual recognition by the blind. Journal of Visual Impairment
and Blindness 78: 215–216.
Lakoff, George. (1987) Women, Fire and Dangerous Things. Chicago: University of
Chicago Press.
Lakoff, George. (1990) The invariance hypothesis: Is abstract reason based on image
schemas? Cognitive Linguistics 1(1): 39–74.
Lakoff, George. (1993) The contemporary theory of metaphor. In A. Ortony (ed.)
Metaphor and Thought 202–251. (Second edition.) Cambridge: Cambridge
University Press.
Landau, Barbara and Ray Jackendoff. (1993) ‘What’ and ‘where’ in spatial language
and spatial cognition. Behavioral and Brain Sciences 16: 217–265.
Langacker, Ronald. (1987) Foundations of Cognitive Grammar. (Vol. I) Stanford, CA:
Stanford University Press.
Mandler, Jean. (1992) How to build a baby II. Conceptual primitives. Psychological
Review 99: 567–604.
Mandler, Jean. (1996) Preverbal representation and language. In P. Bloom,
M. Peterson, L. Nadel and M. Garrett (eds) Language and Space 365–384.
Cambridge, MA: MIT Press.
Mandler, Jean (2004) The Foundations of Mind: Origins of Conceptual Thought.
Oxford: Oxford University Press.
O’Keefe, John and Lynn Nadel. (1978) The Hippocampus as a Cognitive Map. Oxford:
Oxford University Press. [Available on-line at www.cognitivemap.net/]
Tyler, Andrea and Vyvyan Evans. (2003) The Semantics of English Prepositions: Spatial
Scenes, Embodied Experience and Cognition. Cambridge: Cambridge University
Press.
Ungerleider, Leslie and Mortimer Mishkin. (1982) Two cortical visual systems. In D.
Ingle, M. Goodale, and R. Mansfield (eds) Analysis of Visual Behavior 549–586.
Cambridge, MA: MIT Press.
Wertheimer, Max. (1923 [1938]) Untersuchungen zur Lehre von der Gestalt II.
Psycologische Forschung 4: 301–350. Laws of organization in perceptual forms. In
W. Ellis (ed.) (1938) A Source Book of Gestalt Psychology 71–88. London: Routledge.
 Part II
The interaction between language and
spatial cognition

49
2 Language and space: momentary
interactions
Barbara Landau, Banchiamlack Dessalegn and
Ariel Micah Goldberg

Knowledge of language and space constitute two of our most fundamental ways of
knowing the world, and the idea that these systems of knowledge interact is not new.
It has inspired work from diverse intellectual circles, including formal approaches to
language (Fillmore, 1997; Gruber, 1976; Jackendoff, 1983; Talmy, 1983; Langacker,
1986; Lakoff, 1987); theoretical and empirical studies of language learning (Bowerman,
1973; Brown, 1973; E. Clark, 1973; H. Clark, 1973; Mandler, 1992); studies of the
relationship between language and thought (Whorf, 1956; Levinson, 1996; Gleitman
and Papafragou, 2005; Munnich, Landau and Dosher, 2001; Hermer and Spelke, 1996),
and even theories of the way in which evolution could have built on non-linguistic
structures to build a human language (O’Keefe and Nadel, 1978; Hauser, Chomsky
and Fitch, 2002). The diversity of interest in the interaction between spatial cognition
and language represents the widespread assumption that, at some level, language must
map onto our visual and spatial representations of the world. How else would we be
able to talk about what we see?
At least two important consequences follow from this mapping. The first is that
pre-linguistic representations of space could provide a crucial developmental link for
language learning; children might be able to bootstrap their way into the linguistic
system by capitalizing on homologous structures in spatial cognition and language.
Students of language learning have long assumed that the infant’s spatial representations
play an important role in allowing him or her to break into the language system, and
indeed, there is growing evidence that skeletal spatial representations bear a formal
similarity to skeletal linguistic representations (see, e.g. Fisher, 2000; Lakusta, Wagner,
O’Hearn and Landau, 2007; Lakusta and Landau, 2005). The second consequence is
that language, once acquired, might come to modulate our spatial representations. In
this chapter, we focus on the latter effects.
The classical approach to this issue is best known through the views of Benjamin
Whorf (1956), who proposed that language shapes thought. Whorf ’s original observa-
tions focused on the coding of time among the Hopi, but was quickly taken up by
anthropologists and linguists examining other areas of perception and cognition, notably
color (e.g. Berlin and Kay, 1969; Kay and Kempton, 1984). Although experimental
studies during the 1960’s and 1970’s seemed to have settled the Whorfian question
of whether language shapes thought (in favor of a resounding ‘no’; see Brown, 1976
for review), the same question has recently re-emerged, with a flurry of new research
by scientists claiming victory on both sides (see Gentner and Goldin-Meadow, 2003;
Majid, Bowerman, Kita, Haun and Levinson, 2004; Munnich et al., 2001; Gleitman and

51
52 LANGUAGE, COGNITION AND SPACE

Papafragou 2005, for recent surveys). Some have strongly argued that the structure of
one’s spatial lexicon, morphology, and syntax have profound repercussions on non-
linguistic representations– i.e. spatial ‘thought’ (Levinson, 1996; Gentner, 2001). Others
have argued quite persuasively that cross-linguistic variation in spatial terminology
causes no permanent or substantive effects on one’s non-linguistic spatial understand-
ing (Munnich et al., 2001; Malt et al., 2003; Gleitman and Papafragou, 2005; Li and
Gleitman, 2002).
The main purpose of our review is to lay out evidence suggesting a new solution
to this impasse. In particular, we will suggest that a straightforward ‘yes’ or ‘no’ to the
question of whether language changes spatial thought is too simplistic. Rather, we
will present a different twist to the issue, suggested by some newer developments in
thinking about the interaction between language and spatial cognition. Specifically,
we will review evidence that language – once acquired – can strongly modulate our
non-linguistic spatial representations, but that much of this is done in the moment of
carrying out a specific task, and does not result in permanent organizational change
to spatial representation. The functions of language that we will discuss are in a sense
more ‘shallow’– more closely related to the immediate on-line time course within which
our acquisition, comprehension and production take place. The effects occur in a brief
time window, and therefore might be viewed by some as ‘mere’ temporary mechanisms
that operate as we speak and hear sentences, and as we process visual information.
However, we will argue that these temporally brief interactions can play a powerful
role by engaging language to modulate and enhance what is done by the visual system.
Our chapter is organized as follows. First, we provide a brief review of some of the
differences between our representations of language and space, considering some of the
ways in which the systems differ from each other. Where the systems overlap, we can
ask whether and how language modulates our spatial representations. We focus on two
possibilities. One is that language modulates attention because it is inherently selective:
Languages choose to encode certain spatial properties and not others, directing visual
attention accordingly. The second possibility is that language enriches visual-spatial
representations. The idea here is that language permits us to go beyond what is robustly
available to our spatial system of representation, expanding representational power.
Both of these possibilities have been used by researchers to test the Whorfian
hypothesis. The cross-linguistic differences in what languages choose to encode and
the possibility that language can enrich visual representations have been discussed in
the context of permanent changes to our capacity to carry out spatial problems. Despite
stronger claims that have been made about the permanent organizing effects of language,
we will argue that both of these effects (selectivity and enrichment) take place in a
limited time frame. We will review evidence that there are immediate dynamic effects
of language on visual-spatial representations, raising the possibility that the powerful
effects of language are more time-limited than proponents of the Whorfian hypothesis
would suggest.
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 53

1 Specialization of language and spatial representation

Several observations suggest that our linguistic and spatial systems are not redundant
but are complementary. To start, the primitives in each system are unique, as are
their combinatorial rules. Language traffics in primitive symbolic units such as noun
and verb, and configurations of these give rise to semantic and syntactic functions
such as agent, patient, subject, object. In contrast, the spatial system traffics in
primitives such as shapes, objects, locations, landmarks, geometric layouts, angles
and directions, all represented within different spatial reference systems. The rules
of combination for spatial systems are unique as well. For example, objects are
represented as sets of parts that are arranged in terms of hierarchical relationships
and layouts are represented in terms of elements and their geometric arrangements
(e.g. Marr, 1982; Gallistel, 1990).
The differences in formal properties are accompanied by differences in function.
Jackendoff (1987) provides a nice example of the differential power of language and
visual-spatial representations. He considers the case of ducks, geese, and swans (see
Figure 1). Exemplars of these species clearly differ in some set of geometric properties
that are naturally represented by the visual system: Swans have longer necks than geese,
and their overall proportions are somewhat different. These differences in overall shape
of the two animals, including differences in the length of their necks, are well-captured
in visual-spatial representations of objects (e.g. Marr and Nishihara, 1992). But they
are not well-captured in the basic lexicons of languages. To the extent that the overall
shapes of objects are captured in the lexicon or in morphology, the geometric properties
tend to be coarse, such as ‘long thin’ or ‘round’ (commonly encoded by classifiers).

Figure 1. Swans and Geese. An example of the differential power of language vs. visual-spatial
representations. The visual differences between swans and geese are easily represented by the visual
system but are not well-captured by the basic lexicons of languages (Jackendoff, 1987). On the other
hand, language (but not visual representations) naturally captures distinctions such as the difference
between types and tokens (e.g. ‘a swan’, ‘that swan’), which are not readily captured by visual repre-
sentations (see text for discussion).
54 LANGUAGE, COGNITION AND SPACE

Faces provide another example. Humans are experts at recognizing faces that differ
by subtle attributes such as individual features, spacing, and overall arrangement. We
easily encode the differences among enormous numbers of faces, and we are experts at
recognizing even faces we have seen rarely. Language provides proper names to encode
different individuals, but does not encode in any simple and straightforward way the
unique organization of different faces. While we can visually recognize a face quite
easily, we are remarkably ineffective in verbally communicating what the face looks like.
Finally, some aspects of more complex layouts show that visual-spatial representa-
tions can often capture essential relationships that are difficult or impossible to convey
efficiently in language. As Morton (2004) points out, although medical students can
learn the structure of the human skeletal system given linguistic descriptions (e.g. ‘the
hip bone is connected to the femur…’), a diagram of the skeletal system is a more natural
way to represent all of the spatial relationships at once. In general, global spatial layouts
are naturally captured by visual representations but can be quite underspecified when
we use language to describe them.
On the other side, many distinctions are uniquely captured in language but are
not, in any obvious way, a part of our visual-spatial representations. For example, the
very same object may be named as a unique individual (‘that swan’), a member of the
particular species (‘the swan’), or a member of a superordinate class (‘that animal’).
The distinction between a ‘type’ and ‘token’ representation of the same swan, or the
different hierarchical levels to which the same swan can belong cannot be differentiated
by anything in the visual-spatial representation, but these are clearly distinguished in
language. Other distinctions, such as the difference between ‘my book’ and ‘your book’
(given two books that look exactly the same) are naturally made by language, whereas
they are not distinguishable in visual-spatial representations.
These examples show that language and visual-spatial representations are best suited
to conveying distinct sorts of information. But while each system may most naturally
convey certain types of information and not others, it is not the case that their functions
are exclusive. Language can also encode information about spatial relationships and it is
here that we can ask whether and how language modulates our spatial representations.
We turn to the two mechanisms of interest: selectivity and enrichment.

1.1 Selectivity of language

Although language encodes many aspects of our spatial representations, it does not
encode everything. Selectivity is pervasive in language, and the particular elements
that languages select (and the differences over languages) have been central to debates
on whether language causes changes in spatial cognition. We consider two examples:
Selecting components of motion events, and selecting among reference systems (which
are necessarily engaged in order to represent object location).
Across languages, the structure of simple motion events is typically formalized in
terms of several major components, including Figure, Ground (or Reference object),
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 55

Motion, Manner, and Path (see, e.g. Talmy, 1985). In English, motion verbs tend to
encode the motion itself plus the manner, for example, run, skip, hop, dance, swim,
fly, etc. English also encodes the Path separately, often as a prepositional phrase that
includes two elements: the Path function itself (which describes the geometry of the
path) and the reference object(s) (in terms of which the path function is defined). In a
simple example, the sentence ‘Mary ran to the house’ includes a Figure (Mary), Motion
(+ Manner, ran), the Path-function (to) and its Reference object (house). Paths are
further subdivided into TO paths (which focus on the endpoint), FROM paths (which
focus on the starting point), and VIA paths (which focus on the intervening segment)
(Jackendoff, 1983).
This general pattern of encoding is not universal, however. Although English
(sometimes called a ‘Manner’ language) encodes the path in a prepositional component
separate from the verb, other languages (‘Path’ languages, e.g. Spanish, Greek) tend to
encode the path in the main verb, with the manner portion often encoded in a separate
phrase or clause. The difference in tendency to encode manner or path in the main verb
is a hallmark of a major typological division across languages (Talmy, 1985), and has
been used by scientists to examine whether a language’s predominant coding tendencies
have effects on their speakers’ non-linguistic coding of events.
A number of theorists have speculated that the path/manner typological distinction
could have major ramifications for the way that people represent events. That is, this
linguistic difference could affect non-linguistic functions such as memory for visually
perceived events. Simply, if a person’s language tends to encode the path in its main verb
(rather than a separate phrase), then people should represent and remember the path
component of the event more robustly than people whose language tends to encode the
manner of motion in the main verb. The reverse should hold for the manner component.
Despite the appealing simplicity of this proposal, it has been difficult to find such
non-linguistic effects of different languages (Gennari et al., 2002; Papafragou et al., 2002;
see also Munnich et al., 2001). A much more modest proposal has been advanced by
Slobin (1996), who suggests that people’s focus on manner vs. path when they observe
events might be explained as a consequence of ‘thinking for speaking’. That is, individuals
may differentially attend to the various aspects of an event strictly in preparation for
talking about an event– an effect that would not be surprising, since it is a necessary
consequence of preparing to linguistically encode manner that one focus on manner
(and similarly for path). We return to this issue in Section 2, where we discuss evidence
for attentional modulation while people prepare to describe an event but not when
they are preparing to simply remember it. This evidence is consistent with the idea of
temporary, on-line effects, rather than permanent organizational ones.
The second case of selection concerns the use of reference systems in encoding spatial
relationships. Consider Figure 2. A spatial (geometric) representation of this layout
includes the metric locations of each item relative to some reference system. A natural
way of representing the locations would be in terms of a reference system centered on
the box itself (i.e. with the origin at the center of the box). From this center, we can
derive the exact locations of each object relative to the origin, and relative to each other.
56 LANGUAGE, COGNITION AND SPACE

Figure 2. Frames of reference. The locations of A and B can be defined in terms of different reference
systems. For example, A and B can both be located relative to a reference system centered on the box
itself (i.e. with the origin at the center of the box), in which case, both A and B have specific coordi-
nates in this reference system (or, more coarsely, are ‘inside’ the box). Or A can be located relative to
a reference system centered on B (A is above and left of B), or vice versa (B is below and right of A).
Languages can engage several different reference systems (see text for discussion).

But languages do not capture metric relationships in a simple way, a point first elaborated
by Talmy (1983). Rather, languages typically have a stock of basic terms that encode
what Talmy calls ‘topological’ relationships – such as whether one object is inside or
outside of another, whether it is near or far from the other, etc. Even terms that engage
axes (non-topological properties) such as above/below and right/left discount exact
metric distance, and rather, encode categories of relationships relative to the axes of a
particular reference system.
Like the visual system, languages engage a number of different possible reference
systems, with different sets of closed class terms specifying which reference system is being
used. The same layout can be mentally represented in terms of different reference systems,
thereby changing the spatial interpretation of the layout and hence the particular spatial
terms one would choose to describe the inter-object relationships. For example, in Figure
2, we can say that A is near to or to the left of B; or that A is far from or to the right of the
box’s edge. In the first case, we are representing A in terms of a reference system centered
on B; in the second case the reference system is centered on the box’s left edge. Terms
such as ‘right’ and ‘left’ in English are typically used for spatial relationships encoding the
two directions along the horizontal axis, with the origin at one’s own body (the egocentric
system), or at some object or layout in the world (the ‘allocentric’ system). The same holds
for terms above and below, except that they encode the two directions along the vertical
axis (again, with the location of the origin defining the reference system). In contrast to
these sets of words, terms and phrases such as ‘the top/bottom/front/back/side’ typically
engage an object-centered reference system, that is, a reference system whose origin is
centered on an object. Terms ‘north/south/east/west’ engage a geocentric (earth-centered)
reference system, and are usually used (in English) for larger, environmental layouts.
Crucially, selection of a particular term assumes that the speaker has selected a particular
frame of reference; the hearer will have to adopt the same frame of reference (or be able
to translate from the speaker’s frame of reference into one the hearer chooses) in order
to understand what the speaker has said. The particular word or phrase that the speaker
chooses provides information to the hearer about which reference system he has in mind.
Additional specification can be made by including the name of the object that serves as
the center of the reference frame (e.g. ‘above me’ vs. ‘above the table’).
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 57

Although languages generally have the same range of reference systems available
for encoding spatial relationships, there are apparently some cross-cultural differences
in people’s tendency to adopt a given reference frame when they describe locations.
Pederson et al. (1998) argued that the speakers of Tzeltal tend to use geocentric frames of
reference (or ‘absolute’ frames in Pederson et al.’s terminology) rather than egocentric or
allocentric frames of reference (or ‘relative’ frames) when describing spatial relationships
in relatively small arrays. This is quite different from the general tendency of English
speakers: For a tabletop array, English speakers will likely (though not always) use terms
‘right/left’ (e.g. to my right/left or to the right/left of some object) rather than ‘east/west’.
In this work, Pederson et al. (1998) claimed that the habitual use of geocentric reference
frames by the Tzeltal leads to permanent changes in spatial thought, i.e. non-linguistic
problem solving. Pederson’s findings and interpretations have been disputed, however,
on the basis of logic as well as empirical data (Li and Gleitman, 2002). We will review
this dispute and will suggest that the range of empirical findings on reference frame use
is best explained within our framework of on-line, temporary modulation – rather than
permanent reorganization. We will return to this issue in the next section.

1.2 Enrichment by language

The idea of enrichment is quite different from that of selectivity. Where selectivity
emphasizes the power of language in directing attention to a previously available
aspect of spatial representation, enrichment suggests that language can add to spatial
representations. A recent proposal by Spelke and colleagues (Hermer-Vasquez et al.,
1999; Hermer-Vazquez, Moffet and Munkholm, 2001; Spelke et al., 2001) suggests that
language can increase the representational power by which we can carry out certain
spatial tasks. The particular case they discuss concerns a well-known pattern of error
seen in reorientation tasks, in which people are disoriented and then must regain their
bearings in space in order to find a hidden object.
The experiments are patterned after research on rats’ ability to reorient (Cheng
and Gallistel, 1984). In human experiments, a person is typically brought into a small
rectangular room (e.g. 4 x 6 feet) that is uniform in color, e.g. with all black walls and
ceiling (see Figure 3A). The person is shown an object being hidden in one corner, and
is then disoriented (by turning him or her around repeatedly in the center of the room).
When asked to search for the object, people ranging from 18 months through adulthood
tend to divide their search between the correct corner and the one that is geometrically
equivalent, e.g., the long wall to the left of the short wall, as one is facing a corner (see
Figure 3). The pattern is quite robust (in these test circumstances), and has been found
in species as diverse as rats, chickens, and fish, under some circumstances (see Cheng
and Newcombe, 2005, for review). Crucially, even when one of the walls is clearly and
uniquely distinguished by color (e.g. one blue, three black walls; see Figure 3B), toddlers
and other species do not make use of this information, still producing the geometric
error pattern (but see Cheng et al. for some counter examples). The explanation for
this pattern, originally proposed by Cheng and Gallistel (1984) is that the reorientation
58 LANGUAGE, COGNITION AND SPACE

system is modular and encapsulated, operating only on geometric layout information

(i.e. relative lengths and sense of walls) and not admitting other information that is not
relevant to layouts (such as surface color).

FC C FC C
X

GE NC GE NC
(A) (B)
Figure 3. Reorientation by toddlers and adults. (A) When toddlers or human adults are disoriented
in an all-black rectangular room, they then reorient themselves and will search for a hidden target
equally at the correct location (C) and the geometrically equivalent location (GE). (B) When the room
has a single colored wall (shown by thick black bar in 3B), toddlers continue to search at C and GE,
whereas older children and adults can use this added information to narrow their search to the cor-
rect corner only. Redrawn from Hermer and Spelke (1996).

Hermer and Spelke (1996) found that, unlike toddlers and rats, adults and children from
5 years onward can make use of the additional information, and do not commit the
geometric error. Rather, they correctly identify the corner where the object was hidden,
distinguishing between the geometrically equivalent corners on the basis of wall color.
Spelke and colleagues (1996, 1999, 2001) proposed that language plays a crucial role
in creating the uniquely human capacity to solve this problem. Specifically, they argue
that language has the formal power to combine the outputs of different modules– in this
case, the reorientation module (which computes only geometric information) and other
systems (such as the one processing surface color of object). If the reorientation module
produces two possible solutions, language can then combine the geometric description
(e.g. long wall to the left) with non-geometric information (i.e. blue), resulting in the
more powerful expression, ‘the corner that is left of the blue wall’. The larger idea is that
language is the only format that allows for combination of properties that are naturally
part of separate computational domains.
This hypothesis suggests a powerful role for language: Permanently changing spatial
cognition through a new capacity that allows the combination of information from
two different systems. In several studies, Hermer, Spelke and colleagues asked whether
language could play such a role. One experiment examined adults’ ability to solve
the reorientation task when carrying out a secondary task, either spatial (shadowing
a rhythm) or linguistic (verbal shadowing). If language is the mechanism by which
people combine geometric and non-geometric information, then verbal shadowing
should cause impairment, and perhaps reversion to the pattern shown by rats and
non-verbal humans (i.e. toddlers). This was the pattern found by Hermer-Vasquez et
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 59

al. (1999). A second study examined the correlation between children’s performance
in the reorientation task (with blue wall) and their production and comprehension of
terms ‘left/right’ (Hermer-Vazquez, Moffet and Munkholm, 2001). If children’s ability
to solve the reorientation problem depends on language (specifically, the children’s
knowledge of ‘left’ and ‘right’), then there should be a strong positive correlation between
this linguistic knowledge and performance on the reorientation task. Hermer-Vasquez
et al. found that there was a positive correlation between children’s production accuracy
for these terms and their success in the reorientation task. They suggested that this was
consistent with the hypothesis that language is crucial to the solution of the task.
There are several problems with this interpretation of the result (see Cheng and
Newcombe, 2005, for review). For example, there was no reliable correlation between
children’s comprehension of left/right and success on the reorientation task. This suggests
that the role of language may be more limited than was proposed by Hermer-Vasquez
et al. In Section 2, we return to this issue, and report evidence that is consistent with a
different hypothesis about the role of language: That it may support a temporary binding
function that permits combination of color and direction.

1.3 Summary

Language and spatial representation are qualitatively different, and hence are function-
ally specialized for different tasks. Where language and spatial cognition overlap, we can
ask whether and how language modulates our spatial understanding. Recently, several
different lines of research have proposed a strong role for language in permanently
modulating and changing our spatial representations. However, we have hinted that the
existing findings might be better understood as temporary (on-line) effects of language
in modulating spatial cognition. We now turn to research that illustrates some such
temporary effects, and propose that this research casts doubt on stronger Whorfian
claims about the effects of language on spatial cognition.

2 How language temporarily modulates visual-spatial

representations

2.1 Selectivity and the modulation of attention

In this section, we will discuss two examples showing that language can temporarily
modulate attention. These examples concern options for encoding motion events, and
options for encoding locations using different frames of reference. Before we turn to
these examples, however, it is important to address a key assumption underlying our
proposal that language may temporarily influence spatial cognition. For such temporary
interactions to occur, language and spatial cognition must be capable of interacting in an
online fashion. If they could only influence each other offline (that is, over time, rather
60 LANGUAGE, COGNITION AND SPACE

than in the moment of computation) temporary interactions would not be possible.

To this end, we now briefly present a recent line of research illustrating that language
can in fact modulate spatial attention in an on-line fashion, as an integrated part of a
basic cognitive task. This research shows that a basic mechanism in visual search can
be dynamically modulated by language, and lays the groundwork for our subsequent
arguments.
A classical finding in the study of visual search is that the characteristics of search
depend on the number of features that must be used to identify the target item (and
distinguish it from the other items in the array). If the target is distinguished by only
one feature such as color, visual search is very fast and response times do not generally
depend on the number of items in the display (e.g., Treisman and Gelade, 1980; Nagy
and Sanchez, 1990). For example, when one searches for a vertical bar embedded in
an array of horizontal bars, the vertical bar appears to ‘pop out’ of the display, making
search very easy and fast, with little difference due to the number of distracters. However,
when the target is distinguished from the distracters by the combination of two or more
features (e.g., vertical plus horizontal bars that distinguish a target L from a distracter
T) response times typically increase linearly with the size of the display.
Vision scientists have hypothesized that these differences in processing result from
the fact that the visual system encodes different visual features such as vertical and
horizontal orientation independently of one another. Briefly, search for a single feature
is thought to involve parallel pre-attentive mechanisms simultaneously operating over
all the stimuli, causing the target item to appear to ‘pop out’ from the distracters. Search
by multiple features (conjunction search) is thought to engage visual attention, bringing
it to bear on each object individually to determine whether or not the two features occur
together (Treisman and Gelade, 1980; though see Wolfe, 1998 for a review of arguments
against this view). Because each object must be individually attended, the reaction time
for this kind of search increases linearly with the number of items in the display.
Crucially for our argument, researchers have discovered another important prop-
erty of visual search: It is capable of using incremental information while searching.
Rather than being a rigid procedure that always functions the same way given the
visual properties of the stimuli, visual search turns out to be a dynamic process that
is capable of making use of information as it becomes available to it. Thus, reaction
times for conjunction search are significantly reduced if half of the distracter items are
previewed for 1 second before the display is presented (Watson and Humphreys, 1997).
The preview allows people to ignore these (distracter) items, narrowing the set of items
to which visual attention must be directed (see also Olds et al., 2000).
Recently, Spivey et al. (2001) demonstrated that language can have the same effect
on visual search. The general framework for Spivey et al.’s experiments was initiated by
Tanenhaus et al. (1995) who showed that people process language incrementally (i.e.,
they do not wait until the end of a sentence to begin parsing it) and are able to make
use of the linguistic information as soon as it becomes available to guide visual proc-
esses. Spivey and colleagues showed that this same close time-dependent interaction
is engaged during standard visual search tasks. In all of the experiments, subjects were
informed of the target item via recorded speech (e.g., ‘Is there a red vertical?’), and all
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 61

of the visual displays were the same over the different conditions. In the ‘Auditory First’
condition, subjects heard the sentence before the visual display was presented and the
display appeared immediately following the offset of the audio, as is standard in visual
search experiments. In the ‘A/V Concurrent’ condition, however, the display appeared
immediately before the words ‘red vertical’ were heard. Subjects thus heard the target
item and saw the display at the same time (Figure 4).

Auditory First Control Condition

2.5 sec.

Onset

Display
“Is there a red vertical?” Offset
Speech

A/V Concurrent Condition

1 sec.

Onset

Display
Offset
Speech “Is there a red vertical?”

Figure 4. The timeline of presentation. In the ‘Auditory First’ condition, the visual display was pre-
sented following the offset of the audio. In the ‘A/V Concurrent’ condition, the display was presented
as the search target was described auditorily. Adapted from Spivey et al. (2001), Experiment 1.

Spivey et al. replicated the effects of standard visual search tasks in the ‘Auditory First’
condition (which is the standard method used in the vision science community).
Compared to the Auditory First condition, response times in the ‘A/V Concurrent’
condition increased at a significantly slower rate as the number of distracters increased.
Spivey and colleagues hypothesized that subjects were able to make use of the incremental
nature of speech to search the display in two phases. Specifically, as soon as subjects
processed the word ‘red’, they were able to narrow their attention to only the red items.
This allowed search for the vertical figure to proceed much more quickly when ‘vertical’
was subsequently heard (Figure 5). This suggests that even basic visual search tasks can
be modulated by ‘instruction’ from language. Crucially, this modulation occurred while
the sentence was being processed. Language thus can cause significant modulation of
attention in a highly time-bound fashion; language can have an online influence on visual-
spatial computations. We now turn to the role of language in directing attention when
there are options for encoding motion events and locations (using frames of reference).
62 LANGUAGE, COGNITION AND SPACE

1800

1700 y = 7.7x + 1539

1600

1500
A/V Concurrent
Reaction Time (ms)

Auditory First
1400

1300

1200

1100
y = 19.8x + 830

1000

900

800
0 5 10 15 20 25
Set Size

Figure 5. Response time by set size for the ‘A/V Concurrent’ and ‘Auditory First’ conditions. The slope
in the A/V Concurrent condition is shallower than the Auditory First condition, suggesting that the in-
cremental nature of the auditory stimulus facilitated search. Figure adapted from Spivey et al. (2001),
Experiment 1.  Data used with permission.

2.1.1 Language directs attention to different components of motion events

a. Manner vs. Path: Attending to different elements

The different typological patterns for motion events described by Talmy (1985) have been
used to test the possibility that language causes permanent effects on people’s representa-
tion of motion events. As discussed earlier, these studies have generally resulted in weak
findings, with little convincing evidence that, e.g. speaking a Path language permanently
and completely alters our event perception. More importantly, recent findings suggest
that the effects are short-lived, and highly task dependent. They suggest that language
is a powerful modulator of visual attention in the moment of a task.
Papafragou, Trueswell and Hulbert (2006) examined attentional allocation among
Greek and English speakers, asking whether the different typological patterns would
alter visual search. Greek is predominantly a Path language (in Talmy’s typology)
whereas English is a Manner language; the prediction is that the former should show
heightened allocation of attention to Path and the latter to Manner. Papafragou et al.
showed that the different tendencies in Greek vs. English do indeed have consequences
for attentional allocation, but these consequences hold principally for the purposes of
linguistic encoding.
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 63

In the study, native speakers of English and of Greek were shown a sequence of
brief animated events depicting events that were either bounded (e.g. a person skating
around/to a snowman) or unbounded (e.g. a person skating, with no goal in sight). All
participants were told that they would view each event, which would be followed by a
beep. People in the ‘Linguistic’ condition were told that they would be asked to verbally
describe the events after the beep; people in the ‘Non-linguistic’ condition were told
that they should watch the video and then continue inspecting the event after the beep,
because they would be tested afterwards for recognition of the seen events.
Crucially, people in the study were eye-tracked as they carried out the task. Eye
movements can be tracked very precisely, and are an excellent way to track people’s
changing focus of attention throughout a task. Generally, an eye movement to a location
is preceded by attention (e.g., Hoffman and Subrahmanian, 1995; Henderson, 1993; see
Irwin, 2004, for a review), so the pattern of eye movements provides insight into the
allocation of attention. Moreover, eye movements generally occur without conscious
awareness and are not usually subject to explicit voluntary modulation, so they reveal
how attention is allocated as a consequence of the goals driving different cognitive tasks.
Papafragou et al. reported several notable results. First, people in the ‘Linguistic’
condition showed language-specific differences in allocation of attention as soon as
each event started and as it began to unfold over time: English speakers looked to
the region closely related to the manner of motion (e.g. the skates in a skating event)
whereas Greek speakers looked to the region closely related to the endpoint of the path
(e.g. the snowman in the bounded event). These differences map onto the kind of main
verb most likely to be used by English vs. Greek speakers (manner of motion verbs and
path verbs, respectively) and suggest that as the speakers were preparing to describe the
events linguistically, they attended to those properties of the event that would be most
relevant to choosing the proper verb. These cross-linguistic differences occurred only
in the bounded events, which are events that would naturally engage different linguistic
patterns (see Papafragou et al., for discussion).
In contrast to these clear effects of preparing to describe the events, people in the
‘Non-linguistic’ condition showed no differences in eye movements as the bounded
events unfolded over time. That is, both English and Greek speakers who were inspecting
the event for later recall distributed their attention in similar ways. The only differences
in this condition appeared after each event had finished, while people continued to
inspect the still frames of the finished video in order to prepare for the memory task.
Now the English speakers focused attention on the path endpoint (snowman), where as
the Greek speakers focused attention on the manner-relevant element (e.g. the skates).
Papafragou et al. interpret this as evidence that, in this last phase when they were getting
ready for the recall test, people attempted to encode the events verbally. When they did
so, they attempted to focus attention on those elements not well encoded by the main
verbs in their language (e.g. path endpoint for English speakers, manner for Greek
speakers), perhaps as a means to clearly remember those ‘secondary’ elements (i.e. the
ones not encoded by their main verbs).
The overall pattern of results shows that the allocation of attention differs depending
on the cognitive task, specifically, by the kind of cognitive goal that a person has. When
64 LANGUAGE, COGNITION AND SPACE

the goal is to tell ‘what happened’, speakers of different languages will immediately focus
attention on the element(s) that is most likely to be encoded in the main elements of
their language, in this case, the main verb. When the goal is to merely ‘inspect’ the
event, speakers of different languages simply focus on elements that are highly salient,
independent of the language they speak. When the goal is to inspect in order to recall,
speakers may attempt to encode the event linguistically, but then they will also focus
attention on aspects of the event that are not the highest ranked within their linguistic
system (i.e. the aspect not encoded by the main verb). This would seem to be a smart
strategy for improving the overall quality of the event representation, perhaps by creating
a hybrid representation that includes a compact linguistic representation (i.e. of the
main aspect of the event) and a visual-spatial representation that encodes the rest. In
Section 2.2, we will report a related example, in which the combination of language
and spatial representations provides a powerful combination, perhaps richer than either
one or the other alone.

b. TO-Paths vs. FROM-Paths: Reversing natural attentional biases

Although the visual-spatial description of a single path includes the entire geometry
(with no particular bias towards the beginning or end), languages allow us to move
our ‘attentional zoom lens’ (Fisher, Hall, Rakowicz and Gleitman, 1994) over any or
all conceptual portions. For the same observed event, we can say ‘The racehorse ran
from the starting gate at top speed’ or ‘The racehorse ran towards the finish line’ or ‘The
racehorse ran past the viewing stand when he stumbled’. Or we can describe all portions
of the event by combining these pieces. Most linguistic analyses put these components
on equal footing, without marking any one as more primitive than any other.
But when people describe events, they are not so even-handed. Lakusta and Landau
(2005) found that there is a strong tendency for children and adults to encode the goal
of spatial motion events in preference to the source. At the same time, we also found
that language can provide a powerful means to alter this preference, leading people
to focus either on the source or goal, depending on the particular lexical item used in
instruction (see also, Fisher et al., 1994).
In our experiments, we asked 3 and 4 year-old children and adults to describe simple
videotaped manner of motion events. Each of these events showed a person moving
along a specific path from one object to another by a variety of manners, e.g. hopping,
walking, running, crawling, etc. After each event was shown, people were asked to tell
the experimenter ‘what happened’.
In English, events such as these are readily encoded by manner of motion verbs
(e.g. hop, walk, run), and these verbs freely and grammatically take either FROM paths,
TO paths, both or neither. For example, a person could aptly describe a given event as
‘The girl hopped from the mailbox’ or ‘The girl hopped to the lamppost’ or ‘The girl
hopped from the mailbox to the lamppost’ or simply ‘The girl hopped’. Although these
are all – in principal – equally possible, children and adults showed a strong tendency
to explicitly encode the goal path in preference to the source path, saying, e.g. ‘The girl
hopped to the lamppost’ in preference to ‘The girl hopped from the mailbox’ or ‘The girl
hopped from the mailbox to the lamppost’. The tendency was slight among adults, who
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 65

clearly understood that the experimenters wanted a ‘nice complete’ description, but it
was pronounced in young children. This suggests that, in the relatively neutral case of
manner of motion events – where the verb does not ‘care’ whether one or the other, or
any path is encoded – people tend to include the goal path but omit the source path.
In follow-up experiments, we asked whether this pattern – which we called the ‘Goal
bias’ – extended to other kinds of events. Here, we built on a linguistic theory devel-
oped by Gruber (1976) and extended by Jackendoff (1983) as the Thematic Relations
Hypothesis. This hypothesis starts with the observation that there are significant parallels
between the way that paths are encoded linguistically in inherently spatial events –
such as manner of motion events – and in non-spatial events. For example, the same
prepositions ‘to’ and ‘from’ (and their relatives) are used freely in the domain of transfer
of possession (e.g. verbs give/get, throw/catch, buy/sell). As Jackendoff points out, this
kind of transfer is analogous to motion of an object through space from one person to
another, and the verbs that encode these transfers show expressions that are parallel to
those encoding motion of objects in non-transfer contexts. Thus, we can say ‘Mary gave
the watch TO Bob’ or ‘Bob got the watch FROM Mary’, focusing on either the path from
giver to recipient, or vice versa. The parallel also extends to other domains, for example,
change of state (verbs such as turn and grow) and attachment/detachment (verbs such
as attach or hook, detach or remove, etc.)
In order to see whether the same source/goal asymmetry applied to these domains,
we videotaped events that could readily be encoded using verbs appropriate for transfer
of possession, change of state, and attachment/detachment. For example, one set of
events showed ‘giving/getting’, ‘throwing/catching’ and ‘selling/buying’. These events
could be encoded with ‘goal verbs’ (give, throw, sell) or equally well, with ‘source verbs’
(get, catch, buy)– each of which focuses on a distinctly different ‘viewpoint’ for the event.
A second set of events showed changes of state, e.g., an animal whose ears changed
colors and a person whose expression changed from happy to sad. A third set of events
showed events in which a person either attached or detached one object to/from another.
When children and adults were asked to freely describe what happened in these
events, we found the same goal-source asymmetry (i.e. goal bias), with people choosing
‘goal-oriented’ verbs (e.g. give, throw, sell) rather than source-oriented verbs (get, catch,
buy), and specifying the goal paths (e.g. ‘give/throw to X’) rather than source paths (e.g.
‘got/caught from Y’). Thus the goal bias is very robust and appears to represent a bias
for people to construe events in terms of their goal states and/or endpoints rather than
their source states and/or starting points (see Lakusta and Landau, 2005 for discussion).
Given this strong bias (especially among children), one might wonder when and
how people come to flexibly reorient their attentional lens to choose descriptions in
terms of starting points. In a separate experiment, we asked whether we could modulate
interpretation of the event (and full description) by providing children with a verb that
has a source/starting point bias. Using the same videotaped events, we asked a separate
group of children to tell us what happened, but we also told them that we would give
them a ‘hint’. The hint was the target verb, and it was either a goal-oriented verb (e.g.
give) or a source-oriented verb (e.g. get). For example, an event in which an object is
transferred from one person to another could equally well be described with the verb
66 LANGUAGE, COGNITION AND SPACE

‘give’ or ‘get’. When children were told to describe the event using their hint verb ‘give’,
they all complied, saying e.g. ‘The girl gave a present to the boy’. But when they were
told to describe the event using their hint verb ‘get’, they also complied, saying, e.g. ‘The
boy got a present from the girl’ (or ‘The boy got a present’). This shows that 3 year-old
children were adept at framing their description of the event in terms of the goal or
source, even though their spontaneous tendency (found in the previous experiment)
was to frame it in terms of the goal-oriented verb (give).
Children’s facility in this task shows that reframing the construal of the event in
terms of source or goal is relatively easy, if one has the powerful ‘hint’ from language, i.e.
the particular verb that will force the reinterpretation. Although we did not gather eye
movement data to monitor the children’s changing attentional focus, we expect that, as
in Papafragou et al.’s study, eye movements would differ depending on which hint verb
is given. The larger point is that modulation of the mental construal of the event– either
as a ‘source-oriented’ or ‘goal-oriented’ event– can be done quickly, efficiently, and easily
through use of different lexical items that focus attention on different possible construals
of the very same event. This would appear to be a clear case of the immediate modulation
of attention using language as the mental pointer to a new construal.

2.1.2 Language directs attention to different available reference systems

A reference system is a geometric system with an origin and at least two orthogonal axes;
locations of an object can be specified in terms of the locations (possibly coordinates)
on each axis. The same physical location in space can be represented using many dif-
ferent reference systems, each of which is defined by the location of the origin. Thus,
for example, the location of a point (x) in space might be represented relative to a
reference system centered on the retina, the head, the torso, the entire body. It can be
represented relative to a reference system centered on another object (point x is left
of object y), any aspect of the environment (a room, a building, a city), or even larger
spaces, e.g. the earth.
The idea of reference systems is crucial to understanding how we (and other mobile
species) represent location and how we do this for different purposes– e.g. reaching,
pointing, looking, searching, or talking about locations. Because of the importance of
reference systems across all of these domains of inquiry, the literature on use of reference
frames is vast, ranging from studies of how the visual system programs saccades (e.g.
Colby et al., 1999) to how we reach and grasp objects (Milner and Goodale, 2005) to
how we deploy attention (Carlson-Radvansky and Logan, 1997) to the acquisition and
use of spatial language (Carlson-Radvansky and Irwin, 1993; Landau and Hoffman,
2005; Li and Gleitman, 2002).
What is absolutely clear from the wealth of information available on reference
frames is that humans represent location in terms of a variety of reference frames and
that they are flexible in engaging these. This flexibility has recently been examined
within the attention literature, and a striking fact has emerged: When carrying out
tasks requiring that people locate one object relative to another (in order to verify a
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 67

sentence that describes the location), people activate more than one reference system,
and then select one over the other (the latter of which is inhibited; Carlson-Radvansky
and Logan, 1997; Carlson-Radvansky and Jiang, 1998).
The evidence in these studies shows that the engagement of any particular reference
frame is subject to the goals of the perceiver, but that multiple reference frames are likely
to be available for any given task (see also Gallistel, 2002). Still, the mechanisms by
which people select reference frames are not well understood. This makes it surprising
that the literature on language and thought has recently been dominated by a strong
Whorfian hypothesis: Levinson (2003) has proposed that the reference system most
frequently used in one’s native language will lead to permanent organizational changes
in one’s non-linguistic spatial cognition.
This hypothesis was spurred by findings from Pederson et al. (1998), who investi-
gated the linguistic choices of reference frames among speakers of a variety of languages
including Tzeltal, Mopan, Longgu, Dutch and Japanese. The task involved a Director
and a Matcher, who were seated next to each other with a screen placed between them.
They viewed a set of photos and as the Director described a picture, the Matcher was
supposed to select the corresponding one in his own array. Individual pictures were set
up so that a person could use one of several possible frames of reference to describe the
picture. The question was whether speakers of different languages would use different
frames of reference.
For example, one picture displayed a man and a tree; the picture could equally well
be described using ego-centered frame of reference (e.g. ‘the man is on the left’, relative
to the viewer), a geocentric one (or what Levinson and others have called ‘absolute’,
e.g. ‘the man is to the north of the tree’), or an object-centered frame of reference1 (e.g.
‘the man is facing the tree’). All of these frames were used to some extent, but language
groups differed in their tendencies. Speakers of some languages mainly used one of the
frames of reference (e.g. object-centered in Mopan) while speakers of other language
groups used a combination of egocentric and object-centered (Dutch, Japanese) or
object-centered and geocentric (Tzeltal).
Given this variation, Pederson et al. then tested the hypothesis that ‘users of different
language systems (in a given context) should correspondingly vary in their choice of
nonlinguistic spatial problem-solving strategies (in analogous contexts)’ (p. 574). In a
new task, subjects viewed three animals placed facing a particular direction on a table.
They were asked to ‘remember the objects just as they are’ (see Figure 6A). Subjects
were then rotated 180 degrees and after a 30 second delay, walked over to a new table.
At this table, they were given the same animals and were asked to arrange the animals
in the same sequence they had just seen. Notice that the task is ambiguous – given the
setup on the stimulus table, subjects could reproduce the pattern on the recall table
using an egocentric (relative) frame of reference or a geocentric (absolute) frame of
reference (Figure 6B).
68 LANGUAGE, COGNITION AND SPACE

A: Stimulus Table B: Response Table

Relative Absolute

Figure 6. The Animals in a Row task. (A) Subjects are shown the arrangement of animals on the
Stimulus Table. They are then rotated 180 degrees to the Response Table (B), and are asked to recreate
the arrangement of animals. People may replicate the pattern using an egocentric frame of reference
(‘relative’) or a geocentric frame of reference (‘absolute’), depending on the conditions of test and the
person’s native culture (adapted from Pederson et al., 1998 and Li and Gleitman, 2002).

Pederson et al. reported that people from language groups in which geocentric system
is often used gave geocentric (absolute) responses, while other language groups (Dutch
and Japanese) gave egocentric (relative) responses. Thus, taken together the results ‘…
indicate that the frame of reference identified in the linguistic elicitation task correlates
well with the conceptual frame of reference used in this recall task.’ (pg. 580). These
correlations were taken to suggest that ‘…we must represent our spatial memories in
a manner specific to the socially normal means of expression.’ (pg. 586). That is, the
language one uses determines the choice of frame of reference one uses in nonlinguistic
representations.
Li and Gleitman (2002) argued on both empirical and theoretical grounds that
Pederson et al.’s findings did not reflect the effects of language on spatial thought, but
rather, could be viewed as the reverse causal chain: A group’s choice of a particular
frame of reference for encoding in language could easily be the result of culture, ter-
rain, or other variables such as environmental factors; and this non-linguistic choice
could then bias the tendency of language to use one or the other frame of reference.
Li and Gleitman tested this possibility by manipulating the conditions under which
monolingual speakers of English solve the rotation problem. If such speakers (who do
not naturally tend to describe the small scale layout using geocentric terms) also tend
to change their choice of reference frame (without accompanying changes in their
primary language, obviously), then it would follow that the choice of reference frames
is not caused by language.
Li and Gleitman began by administering Pederson et al.’s linguistic and nonlinguistic
tasks to native English speakers. In the linguistic task, with a Director describing photos
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 69

to a Matcher, they found that English speakers overwhelmingly used terms that engage an
egocentric frame of reference, i.e. left and right relative to their body. In the nonlinguistic
task (reconstructing the order of toys they had seen), choice of reference frame was
strongly affected by the surrounding spatial context. When minimal landmark cues were
present (i.e. the experiment was conducted in a lab room with window blinds down)
subjects responded using an egocentric frame of reference, e.g. ‘to my left’ before and
after rotation. This pattern was consistent with that of Dutch and Japanese speakers
in Pederson et al.’s study. But when there were salient landmarks (e.g. the experiment
was conducted outside or in a lab room with the blinds up), people varied quite a
bit, responding either on the basis of an egocentric or geocentric frame of reference.
Moreover, people changed their responses as significant landmarks were introduced:
If a small landmark was placed on both tables such that it was in the same geocentric
location (e.g. north), then people responded using a geocentric frame of reference. If
the landmark was placed in the same egocentric location (e.g. on the left end of the
pre-rotation table but the right end of the post-rotation table), then people adopted an
egocentric frame of reference.
These findings show that, contrary to Pederson et al.’s claim, the language one speaks
has no permanent organizing effect on one’s choice of reference frame for a non-linguistic
task. Rather, as shown by many experiments on spatial cognition on humans and non-
humans, there is great flexibility in which reference system will be adopted for what
task (for review, see Landau 2002; Gallistel, 2002). Moreover, as Carlson-Radvansky and
colleagues have shown, multiple reference frames are likely to be activated in parallel
(Carlson-Radvansky and Logan, 1997). It is the selection of one, and the inhibition of
others, that causes a particular response (Carlson-Radvansky and Jiang, 1998). This
selection occurs in a limited time frame (in less than a second, in Carlson’s studies),
and is unlikely to persist.
So why are there tendencies for speakers of one language to choose one reference
system rather than another in Pederson’s studies? We propose that their findings can be
easily explained as the tendency to solve the ‘non-linguistic’ problem using language– in
which case, of course, one’s linguistic coding would automatically select one or another
reference frame. If one uses the dominant coding of one’s native language, it is not at all
surprising that one would then recreate the test array in accordance with that linguistic
coding. What people would have effectively done in this case is to activate all reference
systems, choose one (if linguistically coding the location, choosing the dominant coding
of their language) and then recreate the array using that coding. Li and Gleitman’s results
show that people’s selection of reference frames can be easily changed depending on
external task conditions, which surely interact with the perceiver/actor’s goal.
The bottom line is that people can freely choose to represent a particular location
within many different reference systems; language may be a mechanism that ramps up
attention to certain reference systems over others, without forcing any permanent change
in the availability of multiple reference frames. And language surely has a powerful
function in providing the means by which a speaker informs the hearer which of the
multiple reference system he or she has in mind. Without such power, it would be hard
to imagine how any of us could ever understand directions.
70 LANGUAGE, COGNITION AND SPACE

2.2 Enrichment: language binds together elements, modulating and

enhancing our visual representations

Our second focus in this chapter is on the role of language in enriching spatial rep-
resentations. Earlier, we discussed a hypothesis put forth by Spelke and colleagues,
suggesting that language provides the computational power to combine outputs of
different systems that are otherwise modular. The case that they offered concerned
reorientation by humans, and they proposed that the solution to this task might require
that language be used to combine geometry and color. We now discuss a related case
from our own research which shows that such combinations might be the product of
on-line, temporary computations in which language and spatial representation enrich
each other.
The general question we have been investigating is the extent to which language
can play a role in enriching visual representations. To do this, we have examined a case
involving a well-known problem in the visual system. The visual system is thought
to process different visual features independently of each other (at some level) and
abundant research has shown that the visual system sometimes fails to bind together
visual features that co-occur in a single object or multiple objects. A classic example was
reported by Treisman et al. (e.g., 1982): If people are very briefly presented with a display
containing a red O adjacent to a green L, they will often mistakenly report that they have
seen either a red L or a green O. This phenomenon is called illusory conjunction and
is thought to reflect difficulty in binding together the two kinds of features (color and
shape). Theories of visual attention have suggested that binding requires active allocation
of focused attention at the location of the target object, with the location serving as the
‘glue’ that binds the features together. Although the bulk of work has been carried out
with normal adults, there are also reports of brain-damaged individuals who experience
both attentional difficulties and a pronounced occurrence of illusory conjunctions. This
combination suggests that attention is necessary for the process of binding together
individual visual features (Arguin, Cavanagh and Joanette, 1994).
We took the case of failure to bind as a possible arena within which to test the
effects of language. Previous findings had shown that young children (around 6 years of
age) might have binding problems when the features are color and location. Hoffman,
Landau and Pagani (2003) showed children geometric blocks that were split in half either
horizontally, vertically, or diagonally and were filled with two different colors in each
half (see Figure 7A). Children were shown a target block, and were then asked to match
it to one of a set of 8 blocks below the target. Children tended to choose the correct split
(e.g. if the target was a vertical split, they chose a block with a vertical split), but they
also tended to err in assignment of color: They might choose a vertically split block with
red on the left/green on the right, or the mirror image. Even a very short (1 second)
delay between viewing the target and selecting the match, there were significant errors.
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 71

A B

Target

Reflection Distracter Target

Figure 7. Geometric blocks split in half horizontally, vertically, or diagonally. (A) The eight blocks
used by Hoffman et al. (2003). When subjects about 6 years old were given a block split, e.g.,
vertically (black right, white left) and were asked to match the target after a 1 second delay,
subjects confused the target with its reflection (black left, white right). (B) Dessalegn and Landau
(2008) used a task where a target block is shown followed by three test items after a 1 second
delay. Subjects confused the target and its reflection in all conditions except when the position
of one of the colors was labeled with a directional phrase: ‘The black is on the right’ (see text for
discussion of the conditions). Note that in the text and actual experiments, blocks split by red
and green were used.  The figure displays black and white splits for ease of printing.

Dessalegn and Landau (2005; 2008) asked if such apparent failures to bind visual features
could be modulated – or even overcome – by language. In a first experiment, we used a
task similar to that used by Hoffman et al. (2003). Four year-old children were shown
target squares that were split vertically, horizontally, or diagonally by color (see Figure
7B). One square was presented on each trial, in the top center of a computer screen.
Children were instructed to look at the target very carefully, so they could find exactly
the same one out of a set of alternatives. When they had finished inspecting the target,
the experimenter clicked the mouse, and the target disappeared; after a 1 second delay,
three options appeared on the bottom of the screen. Options included the target replica,
its reflection (i.e. colors and locations switched), and a distracter square having a differ-
ent split from the target (see Figure 7B). The key question was whether children would
be able to retain the exact identity of the target without confusing it with its reflection.
Doing so would require that each color (e.g. red, green) be bound with its location (e.g.
left, right, top, bottom).
Results showed that children performed above chance, selecting the correct target
block on about 60% of the trials. However, 88% of the errors were target reflection
confusions, e.g. selecting a red-left/green-right block instead of a red-right/green-left
block. This pattern of performance suggests that the visual-spatial representation of
the target did not include a stable representation in which colors were bound to their
respective locations.
72 LANGUAGE, COGNITION AND SPACE

In a second experiment, Dessalegn and Landau asked whether such failures in

forming a stable representation could be overcome by using linguistic instructions that
explicitly provide the location of each color. The same targets were presented using
the same method, except that the child was told, e.g. ‘See this? The red is on the left’
(or right, top, bottom). Children’s performance improved by about 20%, with accurate
performance now hovering around 80%.
In some ways, it may seem trivial that the linguistic instruction ‘The red is on the
left’ helped children to select the correct target after a one second delay. One interpre-
tation is that language can enhance attention (as shown by the phenomena discussed
in the previous sections), drawing the child’s attention to the figures and enhancing
the process of binding color and location. Several other experiments explored the
mechanisms that might accomplish this. One experiment tested whether simply
labeling the target with a novel whole object label would do the trick. Some have
argued that labeling objects has a powerful attentional pull for children, resulting in
heightened attention to certain properties over others (Smith et al., 1996; Smith et al.,
2002). However, when the targets were labeled as whole objects (e.g. ‘See this? This is
a dax.’), children’s performance dropped back to the level observed with no specific
linguistic instruction (i.e. around 60%). The same pattern of performance occurred
when we substituted neutral spatial terms for the directional ones (e.g. ‘See this? The
red is touching the green.’). Moreover, other attempts to directly manipulate the child’s
attention failed. For example, the child was shown the target and told ‘Let’s see where
the red part is’, after which, the red section of the block flashed briefly on and off for
several seconds. This did not result in better performance. In another condition,
the red section grew and shrunk, which presumably should have drawn the child’s
attention. But it did not result in better performance. In yet another condition, the
child was asked to point to the red part, but this did not help.
So how did the ‘left/right’ sentences help? The most obvious interpretation of this
entire pattern of results is that the children had a full and accurate long-term repre-
sentation of the terms top, bottom, left, and right, and that they used this knowledge
to distinguish between the target and the reflection. That is, they stored the linguistic
representation (e.g., ‘the red is on the left’) and were able to make use of it in the
matching task without using their visual-spatial representation of the target at all.
We had anticipated this possibility, and had therefore carried out a production and
comprehension task after the main task, testing children’s long term knowledge of the
terms top, bottom, left, and right.
Children had very accurate representations for terms top and bottom. For left and
right, however, they were near chance at distinguishing the two directions. When asked
to place a dot ‘to the left of ’ a square, they correctly placed the dot along the horizontal
axis, but often erred on the direction, showing that they did not know which end of the
horizontal axis was left and which was right. Crucially, there was no significant correla-
tion between accuracy in the production and comprehension tasks and accuracy on the
main matching task, suggesting that they were not using their long-term understanding
of left/right to carry out the matching task.
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 73

Dessalegn and Landau (2008) interpreted this pattern to suggest that children were
using language in this task as follows: When they heard the sentence ‘The red is on the
left’, the children temporarily represented the term ‘left’ accurately, by just noting the
location (i.e. direction) of the red part relative to the green. This temporary representa-
tion could then be held over the span of the one second delay, and brought to bear on
the task of choosing the target. That is, when the test items appeared, children could
use their temporary representation to select the test-item that had the red on the left
side of the object, successfully distinguishing it from its mirror image. But ten minutes
later, when given the production and comprehension tasks, this representation was
gone, resulting in failure to distinguish between left and right.
As a whole, these findings suggest that language did not have its powerful effect
because of stable, long-term representations of words like left/right. Instead, the findings
point to a powerful but temporary enhancement of the representation of the target’s
direction, which was used on-line, for the purposes of matching, but which rapidly
disappeared. This enhancement augmented the visual-spatial representation of the
target in the context of the task, working to bind together the color and location in the
moment of test.

3.0 Summary and conclusions

As we noted in the beginning of this chapter, the idea that language and visual-spatial
cognition interact is not new. What remains unclear, however, is exactly how these
two quite different systems of representation affect each other– whether the effects are
temporary or permanent, task-dependent or quite general, and the degree to which
the interactions confer greater increased representational power to human cognition.
In this chapter, we have proposed two specific mechanisms of interaction– selectivity
and enrichment. Selectivity occurs because language is inherently selective, encoding
certain distinctions and not others; and because language can serve as a mental pointer,
indicating which of many possible representations we have in mind. Surprisingly, these
effects occur incrementally, as we speak and hear, providing a continually changing
pointer to our different mental construal of the world. Enrichment occurs because
language has the representational power to robustly encode certain properties that are
only encoded in fragile form in the visual-spatial system. We have provided examples
of how each of these mechanisms operates in a time-bound fashion, as people carry
out particular cognitive tasks. The evidence that language can play a time-bound role
in modulating spatial cognition makes us question whether any effects of language on
spatial cognition can be considered permanent, as envisioned by strong versions of the
Whorfian hypothesis. But giving up a strong version of Whorf ’s hypothesis does not
mean relinquishing the idea that language is a powerful modulator of human thinking.
Indeed, the real power of language may be precisely in its time-bound effects, which
ultimately permit humans the flexibility to communicate to others, on a moment to
moment basis, the rich variety of mental construals of the world.
74 LANGUAGE, COGNITION AND SPACE

References
Arguin, M., Cavanagh, P. and Joanette, Y. (1994) Visual feature integration with an
attention deficit. Brain and Cognition 24(1): 44–56.
Berlin, B. and Kay, P. (1969) Basic color terms, their universality and evolution.
Berkeley: University of California Press.
Bowerman, M. (1973) Structural relationships in children’s utterances: Syntactic
or semantic? In T. E. Moore (ed.) Cognitive development and the acquisition of
language. New York: Academic Press.
Brown, R. (1973) Development of the first language in the human species. American
Psychologist 28(2): 97–106.
Brown, R. (1976) Reference in memorial tribute to Eric Lenneberg. Cognition 4(2):
125–183.
Carlson-Radvansky, L. A. and Irwin, D. E. (1993) Frames of reference in vision and
language: Where is above? Cognition 46(3): 223–244.
Carlson-Radvansky, L. A. and Jiang, Y. (1998) Inhibition accompanies reference-
frame selection. Psychological Science 9(5): 386–391.
Carlson-Radvansky, L. A. and Logan, G. D. (1997) The influence of reference frame
selection on spatial template construction. Journal of Memory and Language
37(3): 411–437.
Cheng, K. and Gallistel, C. R. (1984) Testing the geometric power a spatial repre-
sentation. In H. L. Roitblat, H. S. Terrace and T. G. Bever (eds) Animal cognition
409–423. Hillsdale, NJ: Erlbaum.
Cheng, K. and Newcombe, N. S. (2005) Is there a geometric module for spatial
orientation? Squaring theory and evidence. Psychonomic Bulletin & Review
12(1): 1–23.
Clark, E. (1973) What’s in a word? On the child’s acquisition of semantics in his
first language. In T. E. Moore (ed.) Cognitive development and the acquisition of
language 65–110. New York: Academic Press.
Clark, H. (1973) Space, time, semantics, and the child. In T. E. Moore (ed.) Cognitive
development and the acquisition of language 27–64. New York: Academic Press.
Colby, C. and Goldberg, M. (1999) Space and attention in parietal cortex. Annual
Review of Neuroscience 22: 319–349.
Dessalegn, B. and Landau, B. (2005) Relational language binds visual-spatial represen-
tations. Paper presented at the 4th Biennial meeting of Cognitive Development
Society, San Diego, CA.
Dessalegn, B. and Landau, B. (2008) More than meets the eye: The role of language
in binding and maintaining feature conjunctions. Psychological Science 19(2):
189–195.
Fillmore, C. J. (1997) Lectures on deixis. Stanford, CA: CSLI Publications.
Fisher, C., Hall, D. G., Rakowitz, S. and Gleitman, L. (1994) When it is better
to receive than to give: Syntactic and conceptual constraints on vocabulary
growth. In L. Gleitman and B. Landau (eds) Acquisition of the lexicon 333–375.
Cambridge, MA: MIT Press.
Gallistel, C. R. (1990) The organization of learning. Cambridge, MA: MIT Press.
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 75

Gallistel, C. R. (2002) Language and spatial frames of reference in mind and brain.
Trends in Cognitive Sciences 6(8): 321–322.
Gennari, S. P., Sloman, S. A., Malt, B. C. and Fitch, W. T. (2002) Motion events in
language and cognition. Cognition 83(1): 49–79.
Gentner, D. (2001) Spatial metaphors in temporal reasoning. In M. Gattis (ed.)
Spatial schemas and abstract thought 203–222. Cambridge, MA: MIT Press.
Gentner, D. and Goldin-Meadow, S. (2003) Language in mind: Advances in the study
of language and thought. Cambridge, MA: MIT Press.
Gleitman, L. and Papafragou, A. (2005) Language and thought. In K. J. Holyoak and
R. G. Morrison (eds) The Cambridge handbook of thinking and reasoning 633-
661. Cambridge: Cambridge University Press.
Gruber, J. S. (1976) Lexical structures in syntax and semantics. New York: North-
Holland Publishing Company.
Hauser, M. D., Chomsky, N. and Fitch, W. T. (2002) The faculty of language: What is
it, who has it, and how did it evolve? Science 298(5598): 1569–1579.
Henderson, J. M. (1993) Visual attention and saccadic eye movements. In G.
d’Ydewalle and J. Van Rensbergen (eds) Perception and cognition: Advances in
eye-movement research 37–50. Amsterdam: North-Holland.
Hermer-Vazquez, L., Moffet, A. and Munkholm, P. (2001) Language, space, and the
development of cognitive flexibility in humans: The case of two spatial memory
tasks. Cognition 79: 263–299.
Hermer-Vazquez, L., Spelke, E. S. and Katsnelson, A. S. (1999) Sources of flex-
ibility in human cognition: Dual-task studies of space and language. Cognitive
Psychology 39(1): 3–36.
Hermer, L. and Spelke, E. (1996) Modularity and development: The case of spatial
reorientation. Cognition 61(3): 195–232.
Hoffman, J., Landau, B. and Pagani, B. (2003) Spatial breakdown in spatial construc-
tion: Evidence from eye fixations in children with Williams syndrome. Cognitive
Psychology 46(3): 260–301.
Hoffman, J. and Subramaniam, B. (1995) The role of visual attention in saccadic eye
movements. Perception & Psychophysics 57(6): 787–795.
Irwin, D. E. (2004) Fixation location and fixation duration as indices of cognitive
processing. In J. M. Henderson and F. Ferrreira (eds) The interface of language,
vision, and action: Eye movements and the visual world 105–133. New York:
Psychology Press.
Jackendoff, R. (1983) Semantics and cognition. Cambridge, MA: MIT Press.
Jackendoff, R. (1987) On Beyond Zebra: The relation of linguistic and visual informa-
tion. Cognition 26(2): 89–114.
Kay, P. and Kempton, W. (1984) What is the Sapir-Whorf hypothesis? American
Anthropologist 86(1): 65–79.
Lakoff, G. (1987) Women, fire, and dangerous things: What categories reveal about the
mind. Chicago: University of Chicago Press.
Lakusta, L. and Landau, B. (2005) Starting at the end: The importance of goals in
spatial language. Cognition 96(1): 1–33.
76 LANGUAGE, COGNITION AND SPACE

Lakusta, L., Wagner, L., O’Hearn, K. and Landau, B. (2007) Conceptual foundations
of spatial language: Evidence for a goal bias in infants. Language Learning and
Development 3(3): 179–197.
Landau, B. (2002) Spatial cognition. In V. Ramachandran (ed.) Encyclopedia of the
human brain 395–418. (Vol. 4) San Diego: Academic Press.
Landau, B. and Hoffman, J. E. (2005) Parallels between spatial cognition and spatial
language: Evidence from Williams syndrome. Journal of Memory and Language
53(2): 163–185.
Langacker, R. (1986) An introduction to cognitive grammar. Cognitive Science 10(1):
1–40.
Levinson, S. C. (1996) Language and space. Annual Review of Anthropology 25:
353–382.
Levinson, S. C. (2003) Language and mind: Let’s get the issues straight! In D. Gentner
and S. Goldin-Meadow (eds) Language in mind: Advances in the study of lan-
guage and thought 25–46. Cambridge, MA: MIT Press.
Li, P. and Gleitman, L. (2002) Turning the tables: Language and spatial reasoning.
Cognition 83(3): 265–294.
Malt, B., Sloman, S. and Gennari, S. (2003) Universality and language specificity in
object naming. Journal of Memory and Language 49(1): 20–42.
Majid, A., Bowerman, M., Kita, S., Haun, D. and Levinson, S. (2004) Can language
restructure cognition? The case for space. Trends in Cognitive Sciences 8(3):
108–114.
Marr, D. (1982) Vision. San Francisco: Freeman.
Marr, D. and Nishihara, H. (1992) Visual information processing: Artificial intel-
ligence and the sensorium of sight. Frontiers in cognitive neuroscience 165–186.
Cambridge, MA: MIT Press.
Mandler, J. (1992) How to build a baby: II. Conceptual primitives. Psychological
Review 99(4): 587–604.
Milner, A. and Goodale, M. (2005) The visual brain in action. New York: Oxford
University Press.
Morton, J. (2004) Understanding developmental disorders: A causal modeling
approach. Oxford: Blackwell Publishing.
Munnich, E., Landau, B. and Dosher, B. (2001) Spatial language and spatial represen-
tation: A cross-linguistic comparison. Cognition 81(3): 171–207.
Nagy, A. L. and Sanchez, R. R. (1990) Critical color differences determined with a
visual search task. Journal of the Optical Society of America, A, Optics, Image &
Science 7(7): 1209–1217.
O’Keefe, J. and Nadel, L. (1978) The hippocampus as a cognitive map. Oxford:
Clarendon.
Olds, E. S., Cowan, W. B. and Jolicoeur, P. (2000) The time-course of pop-out search.
Vision Research 40(8): 891–912.
Papafragou, A., Massey, C. and Gleitman, L. (2002) Shake, rattle, ‘n’ roll: The repre-
sentation of motion in language and cognition. Cognition 84(2): 189–219.
 LANGUAGE AND SPACE: MOMENTARY INTERACTIONS 77

Papafragou, A., Trueswell, J. and Hulbert, J. (2006) Mapping event perception onto

language: Evidence from eye movements. Paper presented at the 80th Annual LSA 
Meeting, Albuquerque.  
Pederson, E., Danziger, E., Wilkins, D., Levinson, S., Kita, S. and G. Senft (1998)
Semantic typology and spatial conceptualization. Language 74: 557–589.
Slobin, D. I. (1996) From ‘thought and language’ to ‘thinking for speaking’. In J.
J. Gumperz and S. C. Levinson (eds) Rethinking linguistic relativity 70–96.
Cambridge: Cambridge University Press.
Smith, L., Jones, S. and Landau, B. (1996) Naming in young children: A dumb
attentional mechanism? Cognition 60: 143–171.
Smith, L. B., Jones, S. S., Landau, B., Gershkoff-Stowe, L. and Samuelson, L. (2002)
Object name learning provides on-the-job training for attention. Psychological
Science 13(1): 13–19.
Spelke, E. S. and Tsivkin, S. (2001) Language and number: A bilingual training study.
Cognition 78(1): 45–88.
Spivey, M. J., Tyler, M. J., Eberhard, K. M. and Tanenhaus, M. K. (2001) Linguistically
mediated visual search. Psychological Science 12(4): 282–286.
Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M. and Sedivy, J. C. (1995)
Integration of visual and linguistic information in spoken language comprehen-
sion. Science 268: 1632–1634.
Talmy, L. (1983) How language structures space. In H. Pick and L. Acredolo (eds)
Spatial orientation: Theory, research, and application 225–282. New York: Plenum
Press.
Talmy, L. (1985) Lexicalization patterns: Semantic structure in lexical forms. In
T. Shopen (ed.) Language typology and syntactic description 57–149. (Vol. III)
Cambridge: Cambridge University Press.
Treisman, A. and Gelade, G. (1980) A feature integration theory of attention.
Cognitive Psychology 12: 97–136.
Treisman, A. and Schmidt, H. (1982) Illusory conjunctions in the perception of
objects. Cognitive Psychology 14(1): 107–141.
Watson, D. G. and Humphreys, G. W. (1997) Visual marking: Prioritizing selection
for new objects by top-down attentional inhibition of old objects. Psychological
Review 104: 90–122.
Whorf, B. L. (1998) Language, thought, and reality. Cambridge, MA: MIT Press.
Wolfe, J. M. (1998) Visual search. In H. Pashler (ed.) Attention 13–74. Hove, UK:
Psychology Press/Erlbaum.

Notes
1 This case is more complex than stated. A simple object-centered frame of reference
(origin on the tree) would not uniquely capture the fact that the man is facing the tree.
This requires an additional coordination with a frame of reference centered on the man,
including specification of front, back, etc. For purposes of exposition, we are staying
close to Pederson’s own analysis.
3 Language and inner space
Benjamin Bergen, Carl Polley and Kathryn Wheeler

1 Introduction: space in language and cognition

Much of language, including spatial prepositions and verbs of motion, is dedicated to

describing the physical space that our bodies operate in. But it isn’t just that humans
occupy space; in an Escherian twist, our conceptual systems also contain internal rep-
resentations of the world around them. These internal spatial representations become
activated when we reason about spatial events and relationships in the world, when we
recall such spatial events and relationships, and, the focus of the current paper, when
we understand language about space and related domains. The convergent findings we
survey in the following pages are nothing short of remarkable; they show that, in order
to understand language about space, human language users internally reconstruct
the spatial world and experience it through dynamic mental simulations. In short,
understanding language about space is cognitively akin to perceiving space.
For many years, spatial language has been a primary domain of typological and
cognitive linguistic research. All languages appear to dedicate resources to describing
spatial relationships (Majid et al. 2004), but they differ in exactly how they express
them. Different languages group spatial configurations in different ways (Bowerman and
Pederson 1992). For instance, English clusters together vertical and horizontal attach-
ment as on (a piece of paper can be on a wall or on a desk), while German distinguishes
between these relations using an and auf. Even more striking, across languages, different
frames of reference are used for spatial location descriptions. Languages like English
prefer a relative frame of reference for small objects and short distances (the pen is to
the left of the lamp), while languages like Guugu Yimithirr prefer an absolute frame of
reference (something like the pen is North of the lamp) in such cases (Majid et al. 2004).
Despite cross-linguistic variation, however, systematic typological research has sug-
gested at least two interesting ways in which spatial language is similar across languages.
The first is that in all languages, words describing space, including closed-class sets
of function morphemes such as spatial adpositions, appear to be organized in terms
of conceptually constrained semantic primitives. Sometimes termed image schemas
(Johnson 1987, Lakoff 1987), these conceptual primitives capture only schematic infor-
mation, such as contact or containment relations between objects, but not the absolute
size, color, or position of such objects or their containers (Talmy 2000). For instance,
English in encodes the topological relation between an object and its container but not
their Euclidean properties.
A second apparent cross-linguistic spatial universal is that many classes of abstract
concepts are described spatially (Lakoff and Johnson 1980, Lakoff 1993). For instance, in
doesn’t only have a spatial meaning; it can also be used to relate an entity to a state, as in

79
80 LANGUAGE, COGNITION AND SPACE

We’re in trouble, or This research project is in its final throes. Such use of concrete spatial
language for abstract domains is analyzed in the literature as being driven by conceptual
metaphor (Lakoff and Johnson 1980). Across languages, spatial terms come over time
to acquire abstract meanings (Sweetser 1990), which results synchronically in words
that are polysemous, with both spatial and abstract meanings. It has been suggested
that this relation between abstract and concrete goes beyond mere language, such that
understanding of abstract concepts is grounded in experiences with their concrete spatial
counterparts, such as states being understood as containers (Lakoff 1993).
Until recently, there was limited evidence for the psychological reality of the spatial
primitives that purportedly underlie the semantics of spatial and abstract language. The
evidence that abstract concepts are understood in terms of concrete spatial ones was
predominantly linguistic.
During the last decade, however, a number of lines of research have converged
upon a set of common findings, suggesting that understanding language about space, as
well as other abstract domains that are figuratively described in terms of space, results
in the activation of the same cognitive mechanisms that are responsible for perceiving
actual space. This is an instance of a larger movement within cognitive science variably
called simulation semantics (Bergen 2007), motor resonance (Zwaan and Taylor 2006),
or perceptual simulation (Barsalou 1999). Evidence collected in support of these views
indicates that deep language understanding is accomplished by the internal creation
or recreation of experiences of the world, which are triggered by descriptions encoded
in language. These recreated mental experiences – known as mental imagery or mental
simulation – make use of the same neurocognitive resources (the motor or perceptual
systems, for instance) that are typically used for acting on or perceiving aspects of the
world. This paper surveys evidence showing that spatial language does indeed engage
those neurocognitive systems.

2 Spatial language

All normal humans, like most other animals, are endowed with brain systems that serve
the functions of perceiving and acting in space. In humans, these systems are highly
specialized, but at the same time they are so complex that the study of their localization
and mechanics remains in its infancy. One thing we do know is that a variety of higher
cognitive processes recruit systems dedicated to spatial cognition in order to bootstrap
off the existing functionality of these systems.
Foremost among these parasitic cognitive capacities are memory and imagery.
Behavioral and brain imaging evidence over the past century convergently indicates that
recalling or imagining aspects of space involves activating a set of neural circuits that
overlap with those used to perceive or act in space. One of the earliest demonstrations
of this reuse of spatial circuits for imagery employed early image projection technology.
Perky (1910) asked one set of subjects to imagine seeing an object (such as a banana or
a leaf) while they were looking at a blank screen, while the other group was just asked
to look at the screen. At the same time, unbeknownst to them, an actual image of the
 LANGUAGE AND INNER SPACE 81

same object was projected on the screen, starting below the threshold for conscious
perception, but with progressively greater and greater definiteness. Perky found that
subjects who were imagining a banana or a leaf failed to recognize that there was
actually a real, projected image, even at levels where the projected image was perfectly
perceptible to those subjects who were not performing simultaneous imagery. The
interference between imagining and perceiving shown in this early study and scores
of subsequent experiments demonstrates that the system for perceiving objects is also
used for imagining objects.
More recently, work on the Perky effect has shown that interference can also arise
from shared location of a real and imagined object. Craver-Lemley and Arterberry
(2001) presented subjects with visual stimuli in the upper or lower half of their visual
field, while they were (i) imagining objects in the same region where the visual stimulus
appeared, (ii) imagining objects in a different region, or (iii) performing no imagery
at all. They were asked to say whether they saw the visual image or not, and were
significantly less accurate at doing so when they were imagining an object (of whatever
sort) in the same region than when they were performing imagery in another region
or performing no imagery.
Behavioral measures like the Perky effect show that systems dedicated to spatial
cognition are recruited during mental imagery. These same behavioral tools have also
been used to investigate the processing of language about space; in order to test the
hypothesis that understanding language about space, like performing imagery and
recalling aspects of space, makes use of perceptual and motor systems.
A first such line of research investigated whether language denoting action along
the horizontal versus the vertical axis produced Perky effects, similar to those described
above. When concrete spatial language denotes motion, that motion is often likely to
occur along a particular axis. For example, actions like springing and shoving typically
involve vertical or horizontal motion, respectively. Simulation-based theories of lan-
guage understanding predict that the processing of such language involving motion will
automatically drive the activation of dynamic and analog mental simulations, capturing
the embodied experience of motion. Preliminary work by Richardson et al. (2001)
showed that naïve subjects systematically associated gradient axes of motion with action
verbs like spring and shove. This finding was interpreted as suggesting that language
understanders can access image schemas implicating direction of spatial movement
during semantic judgment tasks. A subsequent experiment took these same verbs
and placed them in the context of a Perky-like task. Subjects first listened to sentences
denoting horizontal (1a) or vertical (1b) motion:

(1) a. The miner pushes the cart. (horizontal)

b. The plane bombs the city. (vertical)

Following this, they then saw a shape – either a circle or a square – flash in either the
vertical or the horizontal axis of a computer screen and were asked to press a button as
soon as possible to indicate whether the shape was a circle or a square. Naturally, subjects
were kept unaware of the experimenters’ hypothesis: that the previously presented
82 LANGUAGE, COGNITION AND SPACE

sentence would interfere with processing the shapes if it denoted motion in the same
axis. What Richardson et al. found in this spatial language-induced replication of the
Perky effect was that, as predicted, reaction times to the shapes were longer when the
implied sentential axis matched that of the picture presentation axis.
Reversing the order of presentation, Lindsay (2003) performed a replication in
which he showed subjects an object moving along the horizontal or vertical axis before
presenting a sentence involving movement. Reading times once again showed a sig-
nificant interaction between the direction of the perceived movement of the object and
that of the implied movement in the sentence.
Naturally, the question arises how detailed these activated spatial representations
are. The visual field can in principle be divided more finely than just into axes, and
indeed it would be surprising if internal representations of described events did not
distinguish at least between left and right, let alone up and down. To investigate the
level of detail displayed by language-driven spatial representations, Bergen et al. (2007)
presented subjects with sentences describing actions that would by default occur in
either the upper (2a) or lower (2b) part of the visual field:

(2) a. The mule climbed. (upward movement)

b. The pipe dropped. (downward movement)

Bergen et al. measured how long it would take subjects – after hearing these sentences
– to categorize shapes that appeared in the upper or lower parts of a computer screen.
They found the same Perky-like interference effect described above, with visual objects
presented in the same part of the visual field being processed more slowly.
In order to investigate what sentence parts drive spatial imagery, Bergen et al.
performed an additional manipulation, in which language stimuli had spatial connota-
tions only by dint of their subject nouns and not their verbs (3).

(3) a. The rainbow faded. (up-related noun)

b. The ground shook. (down-related noun)

The results of this study came back the same as those of the previous one; when verbs
are neutral for spatial orientation, a sentence with just an up- or down-connoted noun
can drive location-specific spatial imagery. While a full discussion of the ramifications
of these finding are beyond the scope of the current paper, they suggest that spatial
meanings, perhaps in the form of conceptual primitives like image schemas, appear to
be at work not only in function words, as has often been reported, but also in content
words like nouns and verbs.
Spatial representations of actions are not limited to static relationships like location
along an axis or in a quadrant but can also include direction of movement within the
visual spatial field. Zwaan et al. (2004) had language understanders perform a task
similar to those described above, but with moving target object stimuli. Subjects first
heard a sentence denoting motion that, if perceived, would move away from (4a) or
towards (4b) the body of the listener.
 LANGUAGE AND INNER SPACE 83

(4) a. The shortstop hurled the softball at you. (motion towards the body)
b. You hurled the softball at the shortstop. (motion away from the body)

Subjects then saw two slightly differently sized images of the same object in quick
succession, which subtly yielded the appearance of motion away from or towards the
subject. They were asked to decide if the two objects were the same or different. Just as
in previous studies, subjects’ response times to decide if the objects were the same or not
was affected by the direction of the sentence they had just heard. But interestingly, in this
study, subjects were faster to say the objects were the same if the implied movement was
in the same direction as that of the sentence they had heard. The reasons why studies like
this one yield compatibility effects, whereas other work has shown interference effects,
is still hotly debated (Lindsay 2003, Kaschak et al. 2005, Bergen 2007).
One further finding of note pertains to fictive motion, that is, the use of language
about motion through space to describe static scenes (5).

(5) a. The road runs across the desert.

b. The road winds through a rocky ravine.

In a series of experiments, Matlock (2004) has shown that even fictive motion language
yields dynamic spatial imagery. Subjects reading fictive motion sentences like those in
(5) take significantly less time when the sentence describes fictive motion across short
distances, on smooth terrain or by fast means of travel, as contrasted with descriptions
of long, impeded or slow fictive travel. This effect suggests that language understanders
build up a mental image of the path of described motion through space as they process
spatial language.
The behavioral experiments described above support the idea that, when under-
standing language involving spatial information, people activate spatial simulation
and imagery in a dynamic, analog and modal fashion. In the next section, we examine
evidence for the use of these same systems during the processing of spatial language
used to describe abstract concepts.

3 Metaphorical language

Metaphorical language uses space as a source domain for a number of basic conceptual
target domains. Chief among these are quantity (tax rates are rising again), quality (their
newest film is top-notch) and time (let’s move the meeting forward an hour). The linguistic
facts are unambiguous: spatial language can progressively acquire new conventionalized
non-spatial meanings, and it can also be used in novel ways to describe non-spatial
scenarios with figurative expressions (the price of corn has cannonballed). Nonetheless,
this evidence from language leaves open the question of whether, when processing
a word with a spatial meaning (like rising) to describe a non-spatial event (like an
increase in price), language users actually engage their systems for spatial cognition
in the same way that the behavioral evidence above suggests they do for literal spatial
84 LANGUAGE, COGNITION AND SPACE

language understanding. The empirical results we discuss in this section suggest that
the processing of abstract target domains, such as time, does indeed involve activation
of spatial systems.
There are a number of different ways in which time is linguistically and conceptu-
ally cast in terms of space, and any given language is likely to employ a combination of
these. Time is commonly viewed as a landscape across which the speaker moves, often
labeled the Time Passing is Motion over a Landscape metaphor (6a). Alternatively,
it may be viewed as a row of objects that move in relation to a stationary speaker (6b),
as in the Time Passing is Motion of an Object metaphor (Lakoff 1993, Boroditsky
2000, see also McTaggart 1908).

(6) a. We’re coming up quickly on Easter. (TIME PASSING IS MOTION OVER A LANDSCAPE)
b. Easter flew by. (TIME PASSING IS MOTION OF AN OBJECT)

English employs both of these metaphorical construals of time and, in some cases,
ordinary expressions can be ambiguous as to the underlying spatial metaphor. For
instance, when told that Wednesday’s meeting has been moved forward two days, the
metaphorical forward motion may be interpreted in terms of the motion over a land-
scape metaphor, in which case forward is defined in terms of the experiencer’s direction
of motion – into the future so that moving the meeting forward makes it later. It can
alternatively be interpreted in terms of the motion of an object metaphor, in which
case a line of times move along in a queue with the earliest times first, making forward
motion temporally earlier.
Do these two ways of interpreting language about time in terms of space also rely
on thinking about space, using neurocognitive resources dedicated to spatial reasoning?
Logically, if reasoning about time depends on spatial structures, then inducing language
understanders to think either about themselves moving through space, or contrarily
about objects moving through space, should lead them to interpret ambiguous temporal
language according to the primed spatial schema. In a series of innovative studies, this is
precisely what Boroditsky and colleagues have shown. Boroditsky and Ramscar (2002)
demonstrated that, when standing at the end of a line or waiting for someone to arrive,
a speaker is more likely to adopt the Time Passing is Motion of an Object metaphor
when interpreting ambiguous descriptions of temporal events. In contrast, when first
entering a vehicle or preparing to disembark during the course of a long journey, a
speaker is more likely to employ the Time Passing is Motion over a Landscape
metaphor (Boroditsky and Ramscar 2002). In other words, interpreting language about
time seems to depend upon contextually modulated activation of spatial knowledge.
While all languages cast time as space, the dimensions of space involved in meta-
phorical time language can vary across languages. Chinese, for example, employs not
only the front-back axis to describe past and future events, as in English, but also the
vertical dimension. The Chinese character shang (‘up’) is used in compound words that
refer to past events, while xia (‘down’) denotes future events (Yu 1998). The psychological
reality of this up/down metaphorical mapping is supported by experiments showing
that native Chinese speakers do indeed conceive of time as abstractly laid out along the
 LANGUAGE AND INNER SPACE 85

vertical axis (Boroditsky 2001). In these studies, subjects were primed by pictures to
think about vertical or horizontal motion of objects and then asked to answer a temporal
question (Is April before May?). Even when performing the task in English, native speak-
ers of Mandarin showed better performance on the time question when they had just
performed a vertical spatial reasoning task, as compared with native English speakers,
whose correct time responses were faster following a horizontal spatial reasoning task.
Another way that languages vary in their use of spatial terms for time is the orienta-
tion of the speaker within the Time Passing is Motion over a Landscape metaphor.
English and many other languages describe future events as being in front of a speaker
(We’re just coming up on midterms now) and past events as behind the speaker (I’m so glad
we’re past the rainy season). However, in Aymara, a language spoken by indigenous people
of Bolivia, Peru and Chile, as well as other languages, time is conceived of and described
as though the future were behind and the past ahead. Aymara speakers say quipa pacha
(literally ‘behind time’) to refer to the future and nayra pacha (literally ‘sight time’ or
‘front time’) to refer to the past. At first blush, this arrangment is jarring to speakers of
languages like English that use the reverse orientation. But it is quite well motivated.
The past is known, thus seeable, thus in front, while the future is unknown and, as such,
still hidden or unseen and behind. The backwards-motion-through-time perspective
that underlies metaphorical Aymara expressions can also be seen in the gestures that
monolingual Aymara speakers use when referring to temporal events. When describing
events occurring in the past, they gesture only toward the front, but when referring to
the future they gesture exclusively toward the back (Núñez and Sweetser 2006).
Languages can also differ in the number of dimensions they use to measure time.
English and Indonesian, among many others, commonly describe the duration of an
event with linear spatial descriptors: a short wait. In contrast, Greek and Spanish speak-
ers tend to describe event durations in terms of volume rather than distance, with
expressions equivalent to it took much time. To what extent, though, do these language
differences result in differences in cognitive processing of space and time independ-
ently of language? Casasanto et al. (2004) addressed this question through a series of
psychophysical experiments.
In their first experiment, Casasanto et al. requested native English, Indonesian,
Greek and Spanish speakers to state the most natural phrases in their languages describ-
ing a large period or a long period of time. As predicted, English and Indonesian speak-
ers used expressions corresponding to long time, while Greek and Spanish responses
predominantly described much time. To determine whether there were relativistic
effects of these metaphors on speakers’ cognition, Casasanto et al. presented English
and Indonesian speakers (who tend to quantify time linearly) with a video of a grow-
ing line and asked them to estimate the period of time for which it was presented on
a screen. As predicted, the length of the line interfered with subjects’ judgments of
temporal length: the longer the line was spatially, the more time subjects thought it had
remained on the screen. However, the reverse was not found: duration of display did
not affect subjects’ judgments of spatial length. Showing moving images of an abstract
‘container’ gradually filling also interfered with English and Indonesian speakers’ tem-
poral judgments, but the dynamic container displays did so to a much lesser extent than
86 LANGUAGE, COGNITION AND SPACE

the linear motion displays. In contrast, the temporal reasoning of Greek and Spanish
speakers was modulated to a greater degree by the filling-container animation than by
the growing-line animation (Casasanto et al., 2004, Casasanto and Boroditsky 2003).
In other words, cross-linguistic differences in mappings from space to time correlated
with non-linguistic differences in the extent to which speakers’ temporal judgments
were influenced by their spatial perception.
We started this section with the well-worn observation that time can be described
using spatial terms, in language after language, even among those with no historical
genetic relationship. Cross-linguistic variations in the directions that are mapped (front/
back in English and up/down in Chinese), in the orientation of a speaker in relation to
temporal events (future-facing English speakers versus past-facing Aymara speakers),
and in the image schemas appropriated for temporal terms (long periods of time for
English and Indonesian speakers versus much duration of time for Greek and Spanish
speakers) correlate with cross-linguistic differences in the behavior of speakers of those
languages in reasoning and psychophysical tasks. All of this goes to show that metaphori-
cal language about time is grounded in spatial processes, and additionally that the ways
in which a language construes time in terms of space modulate its speakers’ conceptual
representations of time.

4 The spatial brain

If processing space and processing language about space really do use a shared biological
substrate, then this should be corroborated by imaging studies of the living brain. The
brain exhibits a good deal of localization according to function, with certain regions,
like the well-known Broca’s and Wernicke’s areas of the left hemisphere, often selectively
active during language behavior. Other areas, such as the visual cortex and the so-called
parietal where pathway, are active predominantly in the right hemisphere during the
processing of spatial scenes. Despite evidence that language and space are localized in
discrete neuroanatomical regions, however, recent neurophysiological research indicates
that there is overlap between the structures associated with attending to spatial relations
and processing language about such relations (Kemmerer 2006). This result corroborates
the behavioral evidence described in section 2 above.
The parietal cortex houses neural regions involved in attending to and process-
ing spatial relationships. These same areas become active during retrieval of words
identifying spatial relationships. Using positron emission tomography (PET), Damasio
et al. (2001) imaged the brains of subjects performing naming and spatial relation
judgments. In the first, they were presented with static pictures involving two objects
in a spatial relation (X is on Y) and were asked to name the item (X), and in the second,
they had to name the spatial relationship between them (on). Results showed that the
regions dedicated to perceiving spatial relationships – left parietal and frontal cortices,
and in particular the supramarginal gyrus (SMG) – were also active during spatial
language processing. In other words, the neural structures necessary for perceiving
and understanding spatial relationships appear to get selectively activated for retrieval
 LANGUAGE AND INNER SPACE 87

of language describing spatial relations. Emmorey et al. (2002) also showed left SMG
activation in bilingual speakers of English and American Sign Language (ASL) in a
naming task similar to Damasio et al. (2001). The critical region was more active when
the subjects were naming spatial relations than when they named objects, implicating
this area in semantic processing.
The significance of the SMG in spatial language processing has also been subse-
quently reinforced by a series of studies by Kemmerer (2005) with subjects who had
incurred damage to the SMG. Kemmerer asked subjects to perform a task that used
prepositions in a spatial and then temporal test (at the store; at nine o’clock), and
subjects were required to fill in the appropriate preposition in given contexts. Subjects
with damage to the left SMG performed well on the temporal test but poorly on the
spatial test, while the subject with no damage to this region was able to competently
produce spatial prepositions. These results not only underscore the importance of the
SMG – a region dedicated to spatial perception – in processing spatial language, but
also raise important considerations for studying the neurological basis for metaphorical
uses of spatial terms. The results showed a double dissociation between temporal and
spatial prepositions, suggesting that independent neural substrates can process the same
prepositions when used spatially versus temporally.
These studies show that language users employ neural structures initially devoted
to concrete spatial reasoning when processing and producing language about spatial
relations. This overlap of neural regions is critical to a theory of spatial language
processing that is grounded in spatial cognition. At the same time, the evidence
that spatial and temporal uses of prepositions like at do not require identical neural
substrates implies that metaphorical language using space as a source domain is not
processed in an identical fashion as are literal, spatial uses. This is hardly surprising,
since it could not in principle be the case that spatial and temporal uses of preposi-
tions for example have precisely the same neural substrates. Without different neural
patterns, there would be no behavioral or subjective differences between these differ-
ent uses. However, it does open up important questions about the relation between
spatial and metaphorical uses of spatial language. How is the apparent use of spatial
cognitive mechanisms, evidenced by the behavioral studies cited above, realized in
neural terms? What effect might the conventionality of the metaphorical sense of
a word have on the likelihood that its use will engage the spatial cognition system?
Regardless of the answers, which we can only hope that future work will give hints to,
the neural basis of spatial language processing appears to overlap significantly with
that of spatial cognition.

5 Computational models of language and spatial cognition

Due to the complexity of human linguistic cognition, computational models of language

learning and use are also valuable tools for testing the relative viability of competing
views. The most successful models of spatial language bootstrap linguistic behavior
off of representations of the spatial cognition system. Models of the human semantic
88 LANGUAGE, COGNITION AND SPACE

potential for learning static (e.g. at) and dynamic (e.g. into) spatial language (Regier
1996) and the evolution and acquisition of language for spatial perspectives (Steels et
al. In Prep) have succeeded when spatial language is learned through computational
mechanisms responsible for aspects of spatial cognition. The success of these models
supports the view that the human language faculty is not a discrete component within
the mind, but rather the product of many interconnected units, including, in the case
of spatial language learning and use, some dedicated principally to spatial cognition.
In developing a computational model of the acquisition of spatial terms like in, out,
into, and out of, Regier (1996) drew inspiration from the architecture of the neural cortex,
with discrete computational subunits for (i) creation and comparison of perceptual
maps; (ii) orientation and directional awareness on the basis of perceptual maps; (iii)
motion detection and analysis; and (iv) association of signals from the above three
subunits with an array of locative prepositions. The first three structures within this
architecture process stimuli in a manner globally similar to that of the human visual
cortex, while the fourth serves as an interface between perceptual representations and
the lexicon.
Regier’s model incorporates several explicit constraints, or principles, that guide
the classification of spatial relationships according to sets of primitive features. For
example, an ‘endpoint configuration constraint’ allows the model to recognize the static
perceptual feature of inclusion with a series of images showing movement of a trajectory
from the exterior to the interior of a landmark, which can then be associated with an
appropriate word such as into. This endpoint configuration constraint mirrors findings
of behavioral studies in developmental psychology indicating that children categorize
events more often on the basis of their results than by event-interior relationships
(Behrend 1989, 1990, Smiley and Huttenlocher 1994) and provides a computational
mechanism for linguistic processing according to a Source-Path-Goal image schema
(Lakoff 1987).
Using this architecture, Regier’s model can learn the correct classifications of spatial
percepts according to sets of spatial terms from English, German, Japanese, Mixtec or
Russian. Since each of these languages groups spatial features differently in its encoding
of spatial relationships, Regier’s model supports the idea that spatial language learning
and use is grounded in a primitive and universal set of computational mechanisms
arrayed in the human perceptual cognitive system.
This model, however insightful, is based on the implausible assumption that spatial
language describes scenes viewed from an invariant perspective. Of course, in the real
world, this is rarely true: two speech participants tossing a ball back and forth will
have dramatically different views of the scene, where what is on the right for one will
be on the left for the other, and what is close for one will be far for the other. Successful
communication about spatial scenes thus requires language that responds to these
differences in perspective. One key tool that languages provide for this end is the use
of perspective encoding, namely, language indicating the perspective from which a
particular description holds. In English, possessive nouns can identify the orientation
from which an object description holds (my right, your right, John’s right). On the basis
of the evidence described in the preceding sections, we might hypothesize that learning
and using language describing different perspectives relies on language users engaging
 LANGUAGE AND INNER SPACE 89

those components of the spatial cognition system responsible for adopting alternative
spatial perspectives. In other words, to calculate whether the ball is on John’s right,
a language user might have to perform a mental rotation of the scene as they see it
(Shepard and Metzler 1971) to envision what it would look like from John’s perspective.
Steels et al. (In Prep) conducted a series of simulations where robots were pro-
grammed to describe for each other scenes in which objects moved in one direction
or another, but where speakers did not necessarily share the same perspective as their
interlocutors. The aim was to determine whether endowing these communicating agents
with the ability to mentally rotate a perceived scene would facilitate their communicative
success. The agents for this study were a community of autonomous robots endowed
with various processing modules including (i) a real-time image processing system; (ii)
probabilistic modeling of a three-dimensionally perceived visual world; (iii) active vision
systems to track moving objects; (iv) motor control and obstacle avoidance mechanisms;
and (v) behavioral mechanisms for exploration of the immediate environment, establish-
ment of joint attention and communication of motion events.
During a given trial of the experiment, two robots explored a room until finding a
colored ball and establishing joint attention to it. The human experimenter would then
move the ball and, after perceiving this movement, the robot agents verbally described
the movement. In each trial, if the communication was successful (i.e., if the first robot
was able to describe the movement in terms that the second robot could recognize as
matching the perceived event), the cognitive and linguistic mechanisms used for the
communication task were positively reinforced. If the communication was unsuccess-
ful, however, the cognitive and linguistic mechanisms involved were incrementally
inhibited for future trials. Over the course of a large number of trials (usually on the
order of several thousand), the population collaboratively evolved linguistic terms for
movement events.
By manipulating the robots’ cognitive mechanisms, Steels et al. (In Prep) discovered
that the agents were much more successful at evolving adequate spatial language when
endowed with a cognitive mechanism allowing them to adopt their interlocutor’s per-
spective before formulating or interpreting an utterance, in comparison to those that
were not endowed with such an ability. Moreover, when allowed to invent new words
to indicate the perspective from which a spatial description held, they consistently did
so. This study thus suggests that perspective reversal through mental rotation of an
egocentric view to simulate that of the hearer allows for more efficient development
of language. In light of this finding, it comes as no surprise that human languages
universally encode perspective in spatial language (my left, in front of you).
These experiments assume that certain neurocognitive resources used for human
language, such as the mechanisms required to perform mental rotation or to calculate ori-
entation and direction, ontogenetically precede linguistic capacities. If these computational
models are any indication, human language learning and use seems to draw from other
existing specialized cognitive systems. For the autonomous agents used in the last study,
language development and use were most successful when mechanisms for movement
recognition and mental rotation systems were accessible for recruitment. Just like the
behavioral and imaging evidence above, the computational models described here indicate
a critical role for cognitive mechanisms dedicated to space in the use of spatial language.
90 LANGUAGE, COGNITION AND SPACE

6 Conclusions

The behavioral, neural, and computational evidence surveyed above suggests that lan-
guage users recruit cognitive systems dedicated to spatial cognition when processing
language with spatial content, and that such systems also appear to be used for processing
metaphorical spatial language about abstract domains. These findings coalesce with
those of related work on language processing, which similarly shows that language about
action is understood through the activation of motor circuitry (Glenberg and Kaschak
2002, Pulvermueller et al. 2001), and that language about visually perceptible scenes is
understood through the activation of visual imagery of the implied shape and orienta-
tion of described objects (Stanfield and Zwaan 2001, Zwaan et al. 2002). At the same
time, evidence suggesting that spatial language used for metaphorical purposes recruits
spatial cognition is in line with other work showing that metaphorical language using
other source domains like containment, searching, and possession are also activated
when used to metaphorically structure abstract domains like emotional states (Tseng
et al. 2005, Sato et al. 2006).
The picture of the human capacity for language that emerges from these convergent
results is one where linguistic capacities recruit and make consistent use of existing
cognitive systems – in the case of the studies described in this paper, systems dedicated
to spatial cognition. The use of spatial language, whether literal or metaphorical, appears
to involve a process whereby the spatial configurations described by language trigger
the internal reconstruction by the language user of experiences that are akin to actually
perceiving motion through or relations in space. In processing language about space, an
understander recreates the described spatial experience. This mechanism may serve to
explain large expanses of what it means to deeply understand spatial language.

References
Barsalou, L. W. (1999) Perceptual symbol systems. Behavioral and Brain Sciences 22:
577–660.
Behrend, D. (1989) Default values in verb frames: Cognitive biases for learning verb
meanings. In Proceedings of the Eleventh Annual Conference of the Cognitive
Science Society. Hillsdale, NJ: Lawrence Erlbaum.
Behrend, D. (1990) The development of verb concepts: Children’s use of verbs to label
familiar and novel events. Child Development 61: 681–696.
Bergen, B. (2007) Experimental methods for simulation semantics. In Gonzalez-
Marquez, M., Mittelberg, I., Coulson, S. and Spivey, M. J. (eds) Methods in
cognitive linguistics. Amsterdam: John Benjamins.
Bergen, B., Lindsay, S., Matlock, T. and S. Narayanan. (2007) Spatial and linguistic
aspects of visual imagery in sentence comprehension. Cognitive Science 31:
733–764.
Boroditsky, L. (2000) Metaphoric structuring: Understanding time through spatial
metaphors. Cognition 75(1): 1–28.
Boroditsky, L. (2001) Does language shape thought? English and Mandarin speakers’
conceptions of time. Cognitive Psychology 43(1): 1–22.
 LANGUAGE AND INNER SPACE 91

Boroditsky, L. and Ramscar, M. (2002) The roles of body and mind in abstract
thought. Psychological Science 13(2): 185–188.
Bowerman, M. and Pederson, E. (1992) Crosslinguistic perspectives on topologi-
cal spatial relationships. Paper presented at the 87th annual meeting of the
American Anthropological Association, San Francisco, CA.
Casasanto, D. and Boroditsky, L. (2003) Do we think about time in terms of space? In
Proceedings of the Twenty-fifth Annual Conference of the Cognitive Science Society.
Hillsdale, NJ: Lawrence Erlbaum.
Casasanto, D., Boroditsky, L., Phillips, W., Greene, J., Goswami, S., Bocanegra-Thiel,
S., Santiago-Diaz, I., Fotokopoulu, O., Pita, R. and Gil, D. (2004) How deep
are effects of language on thought? Time estimation in speakers of English,
Indonesian, Greek and Spanish. In Proceedings of the Twenty-Sixth Annual
Conference of the Cognitive Science Society. Hillsdale, NJ: Lawrence Erlbaum.
Craver-Lemley, C. and Arterberry, M. (2001) Imagery-induced interference on a
visual detection task. Spatial Vision 14: 101–119.
Damasio, H., Grabowski, T. J., Tranel, D. Ponto, L. L. B., Hichwa, R. D. and Damasio,
A. R. (2001) Neural correlates of naming actions and naming spatial relations.
NeuroImage 13: 1053–1064.
Emmorey, K., Damasio, H., McCullough, S., Grabowski, T., Ponto, L. L. B., Hichwa,
R. D. and Bellugi, U. (2002) Neural systems underlying spatial language in
American Sign Language. NeuroImage 17: 812–824.
Glenberg, A. M. and Kaschak, M. P. (2002) Grounding language in action.
Psychonomic Bulletin & Review 9: 558–565.
Johnson, M. (1987) The body in the mind: The bodily basis of meaning, imagination,
and reason. Chicago: The University of Chicago Press.
Kaschak, M. P., Madden, C. J., Therriault, D. J., Yaxley, R. H., Aveyard, M., Blanchard,
A. and Zwaan, R. A. (2005) Perception of motion affects language processing.
Cognition 94(3): B79–B89.
Kemmerer, D., (2005) The spatial and temporal meanings of English prepositions can
be independently impaired. Neuropsychologia, 43, 797–806.
Kemmerer, D. (2006) The semantics of space: Integrating linguistic typology and
cognitive neuroscience. Neuropsychologia 44: 1607–1621.
Lakoff, G. (1987) Women, fire and dangerous things. Chicago: The University of
Chicago Press.
Lakoff, G. (1993) The contemporary theory of metaphor. In Ortony, A. (ed.)
Metaphor and thought 202–251. Cambridge: Cambridge University Press.
Lakoff, G. and Johnson, M. (1980) Metaphors we live by. Chicago: University of
Chicago.
Lindsay, S. (2003) Visual priming of language comprehension. Unpublished University
of Sussex Master’s Thesis.
Majid, A., Bowerman, M., Kita, S., Haun, D. and Levinson, S. (2004) Can language
restructure cognition? The case for space. Trends in Cognitive Sciences 8(3):
108–114.
Matlock, T. (2004) Fictive motion as cognitive simulation. Memory & Cognition 32:
1389–1400.
92 LANGUAGE, COGNITION AND SPACE

McTaggart, J. (1908) The unreality of time. Mind 17: 457–474.

Núñez, R. and Sweetser, E. (2006) With the future behind them: Convergent evidence
From Aymara language and gesture in the crosslinguistic comparison of spatial
construals of time. Cognitive Science 30(3): 401–450.
Perky, C. W. (1910) An experimental study of imagination. American Journal of
Psychology 21: 422–452.
Pulvermüller, F., Haerle, M. and Hummel, F. (2001) Walking or talking?: Behavioral
and neurophysiological correlates of action verb processing. Brain and Language
78: 143–168.
Regier, T. (1996) The human semantic potential. Cambridge, MA: The MIT Press.
Richardson, D., Spivey, M., Edelman, S. and Naples, A. (2001) Language is spatial:
Experimental evidence for image schemas of concrete and abstract spatial
representations of verbs. Proceedings of the Twenty-third Annual Meeting of the
Cognitive Science Society 873–878. Mawhah, NJ: Erlbaum.
Sato, M., Schafer A. and Bergen, B. (2006) Effects of picture perception on the
expression of abstract concepts in sentence production. Paper presented at the
19th annual CUNY Conference on Human Sentence Processing, New York, NY.
Shepard, R. N. and Metzler, J. (1971) Mental rotation of three-dimensional objects.
Science 171: 701–703.
Smiley, P. and Huttenlocher, J. (1994) Conceptual development and the child’s early
words for events, objects and persons. In Tomasello, M. and Edward, W. (eds)
Beyond names for things: Young children’s acquisition of verbs. Hillsdale, NJ:
Lawrence Erlbaum.
Steels, L. and Loetzsch, M. (2009) Perspective alignment in spatial language. In
Coventry, K. R., Tenbrink, T. and Bateman, J. A. (eds) Spatial language and
dialogue. Oxford: Oxford University Press.
Stanfield, R. A. and Zwaan, R. A. (2001) The effect of implied orientation derived
from verbal context on picture recognition. Psychological Science 12: 153–156.
Sweetser, E. (1990) From etymology to pragmatics. Cambridge: Cambridge
University Press.
Talmy, L. (2000) Toward a cognitive semantics. (2 volumes) Cambridge, MA: MIT
Press.
Tseng, M., Hu, Y., Han, W. and Bergen, B. (2005) ‘Searching for happiness’ or ‘Full of
joy’? Source domain activation matters. In Proceedings of the 31st Annual Meeting
of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society.
Yu, N. (1998) The contemporary theory of metaphor: A perspective from Chinese.
Amsterdam: John Benjamins.
Zwaan, R. A., Madden, C. J., Yaxley, R. H. and Aveyard, M. E. (2004) Moving words:
Dynamic mental representations in language comprehension. Cognitive Science
28: 611–619.
Zwaan, R. A., Stanfield, R. A. and Yaxley, R. H. (2002) Do language comprehenders
routinely represent the shapes of objects? Psychological Science 13: 168–171.
Zwaan, R. A. and Taylor, L. J. (2006) Seeing, acting, understanding: Motor resonance in
language comprehension. Journal of Experimental Psychology: General 135(1): 1–11.
 Part III
Typological, psycholinguistic and
neurolinguistic approaches to spatial
representation

93
4 Inside in and on: typological and
psycholinguistic perspectives
Michele I. Feist

1 Introduction

Spatial language offers us many windows onto the landscape of human spatial cognition.
But how can we best understand the insights offered by spatial language? What do we
pay attention to when we talk about space? Researchers investigating these questions
have suggested a variety of factors, often individually. How then to make sense of this
complex landscape?
In this chapter, I will sketch the view through two windows onto the landscape of
spatial cognition: one being that of a semantic typologist; the other, that of a psycholin-
guist. The evidence gathered by looking through these two windows will suggest that
despite surface differences in how we talk about space, all humans are attuned to the
same three abstract families of factors – geometric, functional, and qualitative physi-
cal – which together influence the ways in which we talk about relations in space. I will
examine each of these families of factors in turn, along with limitations on proposed
meanings based on a single type of factor.
The importance of geometry to the meanings of spatial relational terms has long
been noted (Bennett, 1975; Feist, 2000; Feist and Gentner, 2003; Herskovits, 1986;
Landau, 1996; Lindkvist, 1950; Miller and Johnson-Laird, 1976; Talmy, 1983; Tyler
and Evans, 2003). Geometry includes information such as the relative vertical and
horizontal positions of the Figure and Ground,1 their proximity to one another (with
inclusion being the closest possibility and contact the next closest), their shapes, and
their relative sizes. Such information forms the basis of many proposed meanings of
topological spatial prepositions, exemplified by the following two researchers’ proposed
meanings for in:

(1) A[locative[interior of B]]

(Bennett, 1975, p. 71)

(2) inclusion of a geometric construct in a one-, two-, or three-dimensional geometric

construct
(Herskovits, 1986, p. 48)

Consistent with geometric approaches to spatial meaning, it has been found that
simply changing the geometric relations in a spatial scene can shift speakers’ intui-
tions regarding the most appropriate preposition to describe the scene (Coventry

95
96 LANGUAGE, COGNITION AND SPACE

and Prat-Sala, 2001; Coventry, Prat-Sala and Richards, 2001; Feist, 2000, 2002; Feist
and Gentner, 1998, 2003). For example, Coventry and Prat-Sala (2001) showed
participants piles of objects placed in containers. They varied the heights of the piles,
placing the Figure at the very top, then asked participants to rate the appropriateness
of in, over, and above to the resultant scenes. They found that this manipulation
resulted in higher ratings for in when the piles were low, and for over and above
when the piles were high.
Although they are intuitively appealing, there are a variety of problems with
representations of the semantics of spatial relational terms based solely on geometry.
First, and most importantly, there are many static spatial uses that cannot be accounted
for by a purely geometric meaning. A simple example will suffice. Consider the two
proposed meanings for in cited above. In both cases, in is described as applicable to
situations in which the Figure is located at the interior of, or included in, the Ground,
as in the case of the pear in the bowl in Figure 1a. However, many spatial terms used
to describe situations of full inclusion, like English in, can also be used for partial
inclusion (Figure 1b; cf. Levinson, Meira and the Language and Cognition Group,
2003) or, in some cases, situations in which the Figure is not geometrically included
in the Ground at all (Figure 1c). It is difficult for a geometric approach to account
for such uses.

(a) (b) (c)

Figure 1. Three pears in three bowls

A second problem faced by geometric accounts of spatial relational meaning is the exist-
ence of multiple possible descriptions for a single scene, as demonstrated in example (3).
Although one can argue that there are distinct shades of meaning, or conceptualizations
(Tyler and Evans, 2003), corresponding to the two sentences, the fact remains that there
is but one geometric relation being described. In addition to failing to motivate alternate
conceptualizations, purely geometric approaches are unable to provide a principled
means of explaining why a speaker might choose one over the other for a particular
situation.

(3) (a) The players are on the field.

(b) The players are in the field.

More recently, researchers have begun to argue that the meanings of spatial relational
terms rely crucially on functional attributes of spatial scenes (Coventry, Carmichael and
Garrod, 1994; Coventry and Garrod, 2004; Coventry and Prat-Sala, 2001; Feist, 2000,
2005b; Feist and Gentner, 2003; Vandeloise, 1991, 1994), as in the proposed meanings
 INSIDE IN AND ON: TYPOLOGICAL AND PSYCHOLINGUISTIC PERSPECTIVES 97

in (4) and (5). Functional attributes include knowledge about the normal uses (if any)
of the objects (particularly the Ground), with special attention to the purpose for which
they were created (Coventry et al., 1994; Feist, 2000, 2002; Feist and Gentner, 1998, 2003;
Vandeloise, 1991), knowledge about whether or not the Figure and Ground normally
interact (Coventry and Prat-Sala, 2001), and knowledge of the manner in which they
are interacting in the current scene.

(4) D/H: a est [=is] dans/hors de b if the landmark and the target are/are no longer the first
and second elements in the container/contained relation.
(Vandeloise, 1991, p. 222)

(5) in: functional containment – in is appropriate if the [G]round is conceived of as fulfilling

its containment function.
(Coventry et al., 1994)

Consistent with such analyses, Coventry and his colleagues (Coventry et al., 1994)
found that the typical function of the Ground object influenced participants’ judg-
ments about the applicability of spatial relational terms: solid objects were judged
more in bowls, which typically hold solids, than jugs, which more typically hold
liquids. Similarly, Feist (2000; Feist and Gentner, 1998, 2003; see below) found that
participants were reliably more likely to use in than on if a pictured Ground was
labeled as a bowl rather than a plate, despite the fact that all participants saw the
same picture.
The functional approach provides a superior explanation for the range of pictures in
Figure 1, as the bowl in each case is fulfilling its usual function as a container, motivating
the use of in. The approach meets up with problems, however, when the Ground object
does not have a normal function (as, for example, in the case of natural kinds), or when
it is filling a qualitative physical role different from its normal function (see below). In
such situations, it is unclear how a functional approach might predict speakers’ uses of
spatial relational terms.
Finally, it has been suggested that the meanings of spatial relational terms are
influenced by the qualitative physics of the spatial scene per se (Bowerman and Choi,
2001; Bowerman and Pederson, 1992, 1996; Feist, 2000, 2005a, 2005b; Feist and Gentner,
2003; Forbus, 1983, 1984; Talmy, 1988; Vandeloise, 2003). Although considerably less
attention has been paid to the independent role of qualitative physical attributes (such
attributes, in fact, do not form the basis for any proposed spatial prepositional mean-
ings), these may prove to be equal to geometry and function in their importance. By
qualitative physics, I am referring to information about the physics of the configuration,
including the presence or absence of a support relation and the ability of one entity to
control the movement of itself or another (cf. Coventry and Garrod’s 2004 discussion of
location control). Often, qualitative physical aspects of the scene result from functional
features, as when a canonical container fulfills its typical function by constraining the
location of another entity. However, this is not always the case. As a case in point, the
typical function of an umbrella is to protect the bearer from falling rain. In the scene
98 LANGUAGE, COGNITION AND SPACE

in Figure 2, however, the umbrella is constraining the location of the apple, motivating
the appropriate use of in. As this example shows, it is important to carefully separate
qualitative physical and functional features, despite their normal co-occurrence.

Figure 2. An apple in an umbrella

Although much theoretical work has suggested important roles for geometry, function,
and qualitative physics in the semantics of spatial relational terms, there remain large
gaps in our knowledge. First, most proposed meanings of spatial relational terms,
such as those cited above, have their basis in a single feature, noting other aspects
only as they support the prominent feature (as, for example, geometric inclusion is a
characteristic of the functional containment relation (Vandeloise, 1991)). Such a view of
spatial meaning, however, leaves many uses of spatial relational terms – even static spatial
uses – unexplained (Feist, 2000), as outlined above. Second, the majority of the work to
date has considered a single language (most commonly English). Yet because linguistic
typology helps to separate out the motivated and explainable from the arbitrary (Croft,
1999), a deep understanding of the semantics of spatial terms may benefit from a wider
crosslinguistic perspective. Third, while the roles of geometry, function, and qualitative
physics have been suggested, their importance awaits detailed empirical verification
(although there have been some efforts in this area, as noted above). To address these
gaps, I will describe two studies. The first, a crosslinguistic survey, addresses the question
of which, if any, of the identified factors recur in the spatial vocabularies of a variety
of languages. The second, a psycholinguistic experiment, addresses the question of
whether small changes in the geometric, functional, and qualitative physical attributes
of a depicted spatial relationship will lead to concomitant changes in speakers’ use of
English spatial prepositions, thus providing empirical evidence for the importance
of these factors to English prepositional meaning. As such, I will be presenting two
complementary views onto the landscape of factors that combine to make up spatial
relational meanings – one typological and one psycholinguistic. What we seek are the
organizing principles around which spatial vocabularies are built.
 INSIDE IN AND ON: TYPOLOGICAL AND PSYCHOLINGUISTIC PERSPECTIVES 99

2 A view through the window of typology

If there is any domain where we might expect universals, it is surely space, due in part
to the universality of our early experience with space (Clark, 1973). It is perhaps this
assumption that has led researchers to examine the semantics of spatial terms largely in
single languages, as the simple topological notions into which spatial terms have been
decomposed (Bennett, 1975; Herskovits, 1986; Miller and Johnson-Laird, 1976) are
largely considered universal, with neurocognitive correlates (Landau and Jackendoff,
1993). In contrast to this intuition, however, the variation in spatial descriptions that
has been uncovered in crosslinguistic studies is astonishing (Bowerman and Choi, 2001;
Bowerman and Pederson, 1992, 1996; Brown, 1994; Feist, 2000, 2004, 2008; Gentner
and Bowerman, 1996, 2009; Levinson et al., 2003; Majid, Bowerman, Kita, Haun and
Levinson, 2004; Pederson, Danziger, Wilkins, Levinson, Kita and Senft, 1998; Sinha and
Thorseng, 1995; Sinha, Thorseng, Hayashi and Plunkett, 1994). Careful examination of
the extensional range of spatial terms in multiple languages further suggests that the
very dimensions of variation may differ across languages, as in the oft-cited difference
between English and Korean spatial terms (Bowerman and Choi, 2001). A simple
example will illustrate this difference. Imagine two scenes: a cassette in its case, and an
apple in a bowl. In English, the two scenes would be described using the same word,
as both are instances of inclusion. In Korean, however, it would be inappropriate to
describe them alike, as one (the cassette in its case) is an instance of tight fit, while the
other (the apple in the bowl) is an instance of loose fit. In describing these two scenes,
the dimensions of contrast that are important in English and Korean are in fact quite
different (but see Vandeloise, 2003, this volume, for an alternate view of this distinction).
Does this mean that the sets of attributes of spatial scenes that languages encode
in their spatial relational vocabularies are incommensurable? Perhaps not. Consider
again the English-Korean distinction. English in communicates inclusion, which is both
geometric, and (due to our three-dimensional gravitational world) physical. Korean, on
the other hand, distinguishes tight and loose fit – a qualitative physical (Vandeloise, 2003,
this volume) and geometric distinction. Thus, despite surface differences in the ways
in which words map to scenes, there are similarities at the abstract level of attention to
geometry and qualitative physics. This explanation echoes the findings of Levinson and
his colleagues (2003), who suggested that there may be universal ‘attractors’, or abstract
relations which languages will tend to recognize. This is also in line with Croft and
Poole’s (2008) suggestion that what is universal across languages may be the constraints
on variation, rather than the specifics of how languages work (see also Feist, 2008).
In addition to uncovering abstract similarities in the semantics of spatial relational
terms – and verifying them across a wide range of languages – there is yet another
reason to examine the typology of spatial semantics. By including more languages in
a sample, we increase the chances that potentially important factors will be identified,
as in the identification of tight vs. loose fit as a result of studying Korean. In addition
to shedding light on human spatial cognition in their own right, some of these factors
may prove relevant even in languages where they were previously discounted. As a
case in point, attributes of the Figure object have largely been considered unimportant
100 LANGUAGE, COGNITION AND SPACE

to the uses of English spatial prepositions (Landau and Stecker, 1990; Talmy, 1983).
Looking across languages, this is by no means a universal fact about spatial relational
terms. For instance, in Mayan languages such as Tzeltal, the nature of the Figure seems
to carry particular importance in the selection of a spatial relational term to describe a
scene (Brown, 1994; Levinson, 1996). Upon reexamination of the role of the Figure in
the use of the English prepositions in and on, Feist (2000; Feist and Gentner 2003; see
below) found a small but reliable effect, suggesting that the role of the Figure had been
mistakenly discounted in previous accounts of English spatial meanings.
Although the field of semantic typology is still in its infancy, seminal work has
already laid the foundations for important advances in our understanding of the ways in
which languages categorize spatial relations (Bowerman and Choi, 2001; Bowerman and
Pederson, 1992, 1996; Feist, 2008; Levinson et al., 2003). I will here describe one further
contribution to this growing area (for complete details of this study, see Feist 2000,
2004, 2008), based on the pioneering work of Bowerman and Pederson (1992; 1996).
Bowerman and Pederson elicited descriptions of a range of topological spatial
relations from speakers of thirty-four languages (see also Levinson et al., 2003), using
a single set of pictures to elicit the terms from all of the languages in a uniform manner.
Their findings illustrated a number of facts about the extensions of spatial terms across
a range of languages. First, none of the languages in their sample used separate terms
for each of the relations exemplified by pictures in their set. Rather, spatial terms in
each of the languages grouped together multiple spatial relations for the purpose of
communication. This finding is important, as it validates the study of the extensions
of spatial relational terms as a means of examining those factors of spatial scenes that
humans deem important. By examining the ways in which the elicited spatial terms
grouped the pictures in their set, Bowerman and Pederson were able to infer the kinds of
semantic distinctions that tend to appear in spatial language. They found that, along with
prodigious cross-linguistic variation, there was a striking commonality. The pictures in
their set could be arranged in a semantic map (Haspelmath, 2003), or ‘similarity gradient’
(Bowerman and Choi, 2001), over which the range of application of each of the elicited
terms could be mapped. Further, in keeping with Croft’s Semantic Map Connectivity
Hypothesis (Croft, 2001, 2003; Croft and Poole, 2008), Bowerman and Pederson found
that none of the terms which they had elicited grouped together discontinuous portions
of their similarity gradient. This systematicity suggests that significant variation co-exists
with deep commonality.
By presenting a single set of pictures to speakers of a wide variety of languages,
Bowerman and Pederson were able to directly compare the extensions of the languages’
spatial terms. Inspired by this, my study borrows Bowerman and Pederson’s methodol-
ogy in order to elicit a data set from which the crosslinguistic importance of particular
attributes to the semantics of spatial relational terms may be inferred. If geometry,
function, and qualitative physics are important structuring elements for human spa-
tial cognition, we can expect to see their influence in the spatial terms of a variety of
unrelated languages.
 INSIDE IN AND ON: TYPOLOGICAL AND PSYCHOLINGUISTIC PERSPECTIVES 101

Twenty-nine simple line drawings, each depicting two objects in a topological spatial
relation, were used in this study. In each picture, one object (the Figure) was colored
in yellow; the second object (the Ground) was left in black and white. Twenty-seven of
the drawings were borrowed from Melissa Bowerman and Eric Pederson’s Topological
Picture Series (Bowerman and Pederson, 1992, 1996; Gentner and Bowerman, 1996,
2009; Levinson et al., 2003), one of the remaining two was modified from the Topological
Picture Series, and the final one was borrowed from an example in Coventry (1998).
Participants were asked to describe the locations of the yellow objects with respect to
the other objects in the most natural manner. Twenty-seven speakers volunteered to
describe the picture series, providing terms from sixteen languages and nine language
families. The languages are listed, along with their genetic affiliations2 and the number
of speakers participating, in Table 1.
Table 1. Languages surveyed in the crosslinguistic study

Language Language Family Number of speakers

in sample
Polish Indo-European, Slavic, West, Lechitic 3
Russian Indo-European, Slavic, East 2
Croatian Indo-European, Slavic, South, Western 1
German Indo-European, Germanic, West, Continental, High 3
Swedish Indo-European, Germanic, North, East Scandinavian 1
Italian Indo-European, Italic, Romance, Italo-Western, Italo-Romance 1
French Indo-European, Italic, Romance, Italo-Western, Western, Gallo- 2
Romance, North
Hindi Indo-European, Indo-Iranian, Indo-Aryan, Central zone, Western Hindi, 2
Hindustani
Hebrew Afro-Asiatic, Semitic, Central, South, Canaanite 3
Hungarian Uralic, Finno-Ugric, Ugric, Hungarian 2
Cantonese Sino-Tibetan, Chinese 1
Telegu Dravidian, South-Central, Telugu 1
Turkish Altaic, Turkic, Southern, Turkish 1
Tagalog Austronesian, Malayo-Polynesian, Western Malayo-Polynesian, Meso 2
Philippine, Central Philippine, Tagalog
Japanese Japanese, Japanese 1
Korean Language Isolate3 1

In order to understand the ways in which a small set of attributes may influence the use
of spatial relational terms across the language sample, the pictures were first analyzed
separately from the elicited terms. Afterwards, the analysis of the pictures was combined
with an examination of the extensional maps of each of the elicited terms in order to
isolate attributes which may influence the uses of the terms.
102 LANGUAGE, COGNITION AND SPACE

First, each of the pictures was coded for whether it matched each of a small set
of geometric, functional, and qualitative physical attributes. The set of attributes was
chosen largely from characterizations of spatial terms in the literature. The geometric
attributes examined were:
a difference in vertical position – important to terms such as above, below, over,
and under (O’Keefe, 1996; Tyler and Evans, 2003)
contact – important to terms such as on (Cienki, 1989; Herskovits, 1986; Miller
and Johnson-Laird, 1976)
inclusion4 – important to terms such as in (Cienki, 1989; Herskovits, 1986; Miller
and Johnson-Laird, 1976; Tyler and Evans, 2003)
relative size – not cited in the literature, but chosen because a larger Ground
might facilitate other attributes, such as inclusion (above) and support (below).

One functional attribute – the presence of a functional relation based on the Ground’s
typical function (Coventry et al., 1994; Vandeloise, 1991, 1994) – was examined. To make
this concrete, coffee and a coffee cup are functionally related, as the typical function of
a cup is to contain a volume of liquid. As such, a functional relation would be coded as
present for a picture of coffee in a coffee cup. On the other hand, a cloud and a mountain
are not functionally related, and a functional relation would be coded as absent for a
picture of a cloud over a mountain.
Finally, the following three qualitative physical attributes were examined:
support – important to terms such as on (Bowerman and Pederson, 1992, 1996;
Herskovits, 1986; Miller and Johnson-Laird, 1976)
control by Ground – important to terms such as in (Coventry et al., 1994;
Coventry and Garrod, 2004)
animacy – important to terms such as in (Feist, 2000; Feist and Gentner, 2003)

Next, the range of application of each of the terms was examined as follows. For each
term, all of the pictures described by the term were grouped together for further
analysis. Each of the groups was then examined in order to isolate the attribute or
attributes that were common to the pictures in the group, based on the picture codings
just described.
Four of the coded attributes emerged as unifying factors in this analysis: a difference
in vertical position, contact, support, and inclusion. The influences of these attributes,
individually and in combination with one anther, are exemplified by the representative
terms in Table 2. For each of the terms listed in Table 2, a plus under an attribute indicates
that the attribute is present in all of the pictures described by the term; a minus indicates
that the attribute is absent from all pictures described by the term.
 INSIDE IN AND ON: TYPOLOGICAL AND PSYCHOLINGUISTIC PERSPECTIVES 103

Table 2. Representative terms

Term Figure higher Contact Ground Inclusion
than Ground supports Figure
[Un0] (Cantonese) +
taas (Tagalog) +
[nad] (Russian) + -
[O‹!CN] (Hebrew) + -
sotto (Italian) - +
sous (French) - +
na (Polish) + +
på (Swedish) + +
auf (German) +
an (German) +
u (Croatian) +
[la], [NQR‹NC] +
(Telegu)
iqinde (Turkish) +
[V5‹] (Cantonese) -

As further evidence of the unifying nature of these attributes, they together served to
categorize fifty-six of the sixty-three collected terms into the following seven classes
of meaning.5

Figure higher than Ground

Figure higher than Ground, no contact
Figure lower than Ground, with contact
Ground supports Figure with contact
Contact
Inclusion of Figure by Ground
Absence of inclusion of Figure by Ground

These four attributes together provide evidence for the importance of geometry, function,
and qualitative physics to the meanings of spatial terms across a variety of languages.
The first two, a difference in vertical position and contact, provide information about
the geometry of the Figure-Ground relationship. The third, support, provides qualita-
tive physical information about the Figure-Ground interaction and the forces acting
between the objects. Finally, inclusion provides information about geometry, function,
and qualitative physics. In a three-dimensional world, geometric inclusion of one entity
by another entails that the location of the included entity (the Figure) is constrained
by the including entity (the Ground): in order to be geometrically included, the Figure
must be located at the interior of the Ground. As such, the geometric attribute inclusion
validates inferences about the presence of the qualitative physical attribute location
control.6 Similarly, as control of the location of an object is a common human goal, many
artifacts have been created to fulfill this function, with the result that if the Ground is an
artifact, inclusion of the Figure likely results from the fact that the Ground was created
for this purpose (as was the case for the pictures in the current study).
104 LANGUAGE, COGNITION AND SPACE

The view through the window of semantic typology shows a landscape in which
significant variation coexists with abstract similarities. Although spatial relations are
grouped differently by the languages studied, attributes from all three families – geo-
metric, functional, and qualitative physical – recurred in the meanings of the collected
spatial terms. However, while typological studies such as the one presented here may
suggest factors that are important to the meanings of spatial relational terms, controlled
experimental investigation is necessary in order to test the roles of the factors in speakers’
decisions to use specific terms. It is to this issue that we will now turn.

3 A view through the window of psycholinguistics

The view through the window of typology provided support for the theoretical import of
geometry, function, and qualitative physics to the meanings of spatial relational terms.
In language after language, it was found that geometric, functional, and qualitative
physical properties united the disparate sets of scenes that could be described by a single
term. Yet to be sure that these attributes truly influence speakers’ choice of a word to
describe a situation, we must seek corroborating evidence. Will a change in one of these
factors lead to a concomitant change in a speaker’s likelihood to employ a given term?
In order to closely examine the influence of any given attribute, it is desirable
to hold as many other attributes constant as possible. This problem is nontrivial, as
many of the attributes of spatial scenes that participate in spatial relational meaning
co-occur in the real world. For example, support (a qualitative physical attribute)
seldom occurs without contact (a geometric attribute) in everyday interactions (Feist,
2000). Similarly, as discussed above, many artifacts are created for the purpose of
constraining the location of other objects, thus combining geometric, functional, and
qualitative physical attributes in relations resulting from their normal use. In an attempt
to tease apart a small set of attributes of scenes that influence the use of the English
spatial prepositions in and on, Feist (2000; Feist and Gentner, 1998; 2003) adapted a
method developed by Labov (1973) to study complex interacting factors in the use of
English nouns. The details of the experimental study are reported in Feist (2000; see
also Feist and Gentner, 2003). I present here an outline of the main experiment along
with reasonably complete results.
In his classic study of naming patterns for cuplike artifacts, Labov (1973) sys-
tematically varied the functional context (e.g., holding coffee or holding flowers) and
the relative height and width of a set of similarly shaped objects, which he then asked
participants to name. He found that the variation in these factors led to changes in
the nouns adults chose to name the objects. Similarly, I created a set of spatial scenes
which were systematically varied with respect to geometric, functional, and qualitative
physical factors in order to closely examine their influences on the use of the English
prepositions in and on. The extent to which the differences in the pictures correlate with
the changing rate of use of these English spatial prepositions is taken as indicative of
the roles of these factors in the meanings of the prepositions.
In approaches to the meanings of in and on based on geometry, it is apparent that,
while in requires that there be an interior of the Ground at which the Figure may be
 INSIDE IN AND ON: TYPOLOGICAL AND PSYCHOLINGUISTIC PERSPECTIVES 105

located, on requires merely a surface with which the Figure must be in contact (Bennett,
1975; Herskovits, 1986; Miller and Johnson-Laird, 1976). Consider a Figure in contact
with the upper surface of a Ground. By manipulating the concavity of the Ground,
without further change in the position of either object, it is possible to shift the relative
applicability of the prepositions in and on (Figure 3). The influence of geometry was thus
examined via changes in the curvature of the Ground. If geometry influences preposition
choice, greater curvature (and concomitantly deeper concavity) of the Ground should
correspond to a higher proportion of in responses.

(a) (b)
Figure 3. Two scenes differing only with respect to the concavity of the Ground

To vary the perceived function of the Ground, we took advantage of Labov’s (1973)
finding that the choice of a noun to label an object is influenced by the functional context
within which the object is presented. Thus, we introduced the inanimate Ground with
one of five labels, each communicating different information about the typical function of
the Ground. The labels chosen were dish, plate, bowl, rock, and slab. If function influences
preposition choice, we should see the greatest use of in when the inanimate Ground is
labeled as a bowl, which is a prototypical container. Use of in should be lower for plate,
which typically names objects that function as a supporting surface, and intermediate for
dish, which is a superordinate term for plate and bowl. Finally, use of in should be low for
rock, which is an afunctional solid, and for slab, which is an afunctional surface.
As information about qualitative physics is difficult to directly manipulate in static
scenes while holding geometry constant, we indirectly manipulated qualitative physical
properties by varying the animacy of the Figure and the Ground. An animate Figure, by
virtue of its ability to enter and exit a configuration under its own power, may be conceived
of as being less under the control of the Ground than would be an inanimate Figure.
Conversely, an animate Ground is able to exert volitional control over the location of the
Figure, while an inanimate Ground is not. If indirect effects of animacy on qualitative
physical attributes related to location control influence preposition use, we might expect
to see the greatest use of in for those situations that are physically most stable – situations
where the Ground is animate and situations where the Figure is not. Similarly, we might
expect to see the least use of in for those situations which are least stable – situations in
which the Figure is animate and situations where the Ground is not. Thus, we should see
greater use of in when the Ground is animate than when it is not. Likewise, we should see
that the use of in is more prevalent when the Figure is inanimate than when it is animate.
In all, there were a total of twelve pictures. The set included two Figure objects – one
animate and one inanimate. The Figures were each placed with respect to two Ground
objects – one animate and one inanimate – and the Grounds were depicted at three
levels of concavity, with the concavity of the two Grounds being equal at each level. The
complete design is sketched in Figure 4.
106 LANGUAGE, COGNITION AND SPACE

Geometry – concavity differences

Animate Ground

Low Medium High

Figure is
firefly or coin

Inanimate Ground
Low Medium High

Figure is
firefly or coin

Function - Labeling Condition for inanimate Ground

Figure 4. Design of the psycholinguistic study

The twelve pictures were presented individually on a computer screen in random order,
and participants were given answer sheets with sentences of the following form:
The Figure is IN/ON the Ground.

Figure was replaced with the noun referring to the pictured Figure (firefly or coin);
likewise Ground was replaced with hand when the pictured Ground was the animate,
and with the noun corresponding to the participant’s labeling condition (dish, plate,
bowl, rock, or slab) when the inanimate Ground was shown. The participant’s task was
to circle in or on to make each sentence describe the corresponding picture on the
computer screen.
As predicted, participants’ choices between in and on were found to be influenced
by geometric, functional, and qualitative physical factors, as confirmed by a 2 (Ground:
hand or inanimate) x 2 (Figure: firefly or coin) x 3 (concavity) x 5 (labeling condition)
repeated measures analysis of variance. I will discuss each of these factors in turn.
That geometry plays a role in the meanings of in and on can be seen from the effect
of changing the concavity of the Ground. As the concavity of the Ground increased,
so did the use of in, with the average proportion of in responses for scenes depicting
low concavity at .38, the average proportion for scenes depicting medium concavity at
.45, and the average proportion for scenes depicting high concavity at .54, F(2,172) =
28.34, p < .0001 (Figure 5).
 INSIDE IN AND ON: TYPOLOGICAL AND PSYCHOLINGUISTIC PERSPECTIVES 107

proportion in responses
0.8

0.6

0.4

0.2

0
low medium high
concavity level
Figure 5. Effect of concavity, averaged across both Figures, both Grounds, and all five labeling conditions

That functional information plays a role in the meanings of in and on can be seen from
the effect of varying the label provided for the inanimate Ground (F(4,86) = 10.77, p
< .0001). As expected from the fact that the label was only changed for the inanimate
Ground, there was also an interaction between the labeling condition and the animacy
of the Ground (F(4,86) = 5.43, p = .001) (Figure 6). When the inanimate Ground was
labeled as a bowl, a label normally applied to prototypical containers, the use of in
was most prevalent (mean proportion in responses = .65). When the inanimate was
labeled with plate, a noun normally used to label a functional surface, the propor-
tion in responses was much lower (mean proportion in responses = .09). When the
superordinate term dish was used, the proportion in responses was in between (mean
proportion in responses = .50). Finally, the use of in was quite rare when the Ground was
presented along with a label which suggested that it was not a functional artifact (mean
proportion in responses for rock = .07; mean proportion in responses for slab = .08).

0.8
proportion in responses

0.6

0.4

0.2

0
bowl dish plate slab rock

Figure 6. Effect of labeling condition for the inanimate Ground, averaged across all three concavities
and both Figures
108 LANGUAGE, COGNITION AND SPACE

The influence of qualitative physics on the meanings of in and on can be inferred from
the effects of the animacy of the Ground and the animacy of the Figure. When the
depicted Ground was a hand, which is able to exert volitional control over another
entity, the use of in was more prevalent than when the depicted Ground was inanimate
(mean proportion in responses, hand as Ground = .63; mean proportion in responses,
inanimate Ground = .28, F(1,86) = 65.59, p < .0001). Further, I found an interaction
between the animacy of the Ground and its concavity whereby the increase in the
proportion in responses as concavity increased was sharper for the hand than for the
inanimate Ground (F(2,172) = 5.50, p = .005) (Figure 7). This difference makes sense
in qualitative physical terms: because it can continue to close, a hand may be thought
of as having more control over the location of its contents as it becomes more concave
(more closed), while an inanimate object’s degree of control, like its ability to continue
closing, would remain constant across concavities.

1
jproportion in responses

0.8

0.6
hand
inanimate
0.4

0.2

0
low medium high
concavity of Ground

Figure 7. Interaction of animacy of the Ground and concavity, whereby the increase in in responses
with increased concavity is sharper for the hand than for the inanimate Ground

In support of this explanation of the effect of the animacy of the Ground, and consistent
with the predictions, when the depicted Figure was animate (a firefly), and thereby able
to exert control over its own location, the use of in was less prevalent than when the
depicted Figure was inanimate (mean proportion in responses, firefly as Figure = .43;
mean proportion in responses, coin as Figure = .49, F(1,86) = 9.69, p < .005). Further,
the influence of the animacy of the Figure interacted with the influence of functional
information about the Ground: the extent to which firefly received a lower proportion in
responses than did coin was greatest when the label for the inanimate Ground suggested
a containment function (bowl and dish), F(4,86) = 2.73, p < .05 (Figure 8). The function
of a container is, at its most basic, to fulfill the qualitative physical role of constraining
the location of another object. This function can best be fulfilled if the object is more
constrainable. As such, qualitative physics and function reinforce one another in scenes
depicting an inanimate Figure and a Ground labeled as a container, hence raising the
applicability of in.
 INSIDE IN AND ON: TYPOLOGICAL AND PSYCHOLINGUISTIC PERSPECTIVES 109

proportion in responses 0.8

0.6
coin
firefly
0.4

0.2

0
bowl dish plate slab rock

Figure 8. Interaction of labeling condition and animacy of the Figure, whereby the difference be-
tween responses to the coin and the firefly appear predominantly when the Ground is labeled as a
functional container

Taken together, this set of results demonstrates that geometric, functional, and qualita-
tive physical properties all influence speakers’ uses of the English spatial prepositions in
and on. Furthermore, although each exerts an independent influence on English prepo-
sitional usage, these three families of factors are not completely independent. Rather,
they influence one another in complex ways, often providing reinforcing information
that can raise the applicability of a preposition to a scene. Thus, the view through the
window of psycholinguistics echoes the view through the window of typology, providing
evidence that those factors which recur in the uses of spatial terms across languages also
individually influence speakers’ choices in a controlled communicative environment.

4 Conclusions

Multiple times each day, speakers choose from among a relatively small set of spatial
relational terms (Landau and Jackendoff, 1993) to describe one of infinitely many pos-
sible spatial configurations between two objects in the environment. Their decisions are
quick and sure, reflecting the automaticity of spatial relational terms. What attributes of
spatial configurations must speakers attend to in order to fluently use the set of spatial
relational terms available in their language?
While the semantics of spatial relational terms has received extensive attention, the
picture of spatial relational meaning that emerges from an examination of theoretical
treatments of spatial semantics is difficult to interpret. First, most characterizations
of the meanings of spatial relational terms rely on a single type of feature. As a result,
many common uses of spatial relational terms are left unexplained by the proposed
meaning. Further, there is disagreement about whether geometric or functional features
are criterial for spatial relational meaning. Second, the majority of the studies to date
have involved single languages. Although these studies have catalogued the uses of
110 LANGUAGE, COGNITION AND SPACE

the terms in the language under consideration, they are unable to provide a sense of
spatial language more generally. Such a sense can only be gotten by considering the
spatial vocabularies of many languages. It is precisely this sense of spatial language more
generally that may provide the insights necessary to arrive at a descriptively adequate
account of the meanings of individual spatial relational terms. Third, while theoretical
treatments of spatial relational terms have proposed hypotheses about the factors that
participate in the meanings of the terms, very few controlled experimental tests of the
hypotheses have appeared.
In recent years, all three of these open issues have begun to be addressed, leading to
a clearing picture of the factors participating in the semantics of spatial relational terms.
With regard to the first issue, meanings incorporating more than one type of factor have
been proposed (Coventry and Garrod, 2004; Feist, 2000; Herskovits, 1986), expanding
the range of uses that can easily be accounted for within the proposed meaning. On
the second count, researchers have begun to examine the spatial relational terms of
multiple languages within a single project (Bowerman and Choi, 2001; Bowerman and
Pederson, 1992, 1996; Feist, 2000, 2004, 2008; Levinson et al., 2003), concomitantly
expanding the range of distinctions of which they are aware. Finally, with regard to the
third open issue, researchers have begun to test the validity of the proposed factors in
controlled psycholinguistic experiments (Carlson, Regier, Lopez and Corrigan, 2006;
Coventry et al., 1994; Coventry and Prat-Sala, 2001; Coventry et al., 2001; Feist, 2000,
2002, 2005b; Feist and Gentner, 1998, 2003), allowing them to verify the role that each
one plays in the meanings of individual spatial relational terms.
In this chapter, I have provided an overview of two studies designed to address the
second and third of the identified gaps in our understanding of the semantics of space.
In doing so, these studies provide valuable data which can be used to further efforts to
address the first gap.
The first of the studies discussed compared the extensional ranges of sixty-three
spatial relational terms collected from sixteen languages, representing data from nine
language families. In order to be made maximally comparable, the terms were elicited by
having all of the participants describe the same set of simple line drawings. The results
showed that four attributes of spatial scenes, a difference in vertical position, contact,
support, and inclusion, together provided unifying explanations for the individual exten-
sional ranges of the fifty-six specific spatial terms collected (those encoding relatively
detailed information about the Figure’s location; see Feist (2004, 2008)). At a more
abstract level, these four attributes impart information about geometric, functional,
and qualitative physical aspects of the spatial scenes, providing evidence that these
three families of factors influence the uses of spatial relational terms across a range of
languages.
The second of the studies discussed in this chapter examined English speakers’ uses
of the prepositions in and on to describe a small set of scenes designed to vary along
geometric, functional, and qualitative physical parameters. The results suggest roles
for all three kinds of factors in the meanings of these two prepositions. The influence
of geometry was demonstrated by the rise in in responses as concavity of the Ground
increased (Figure 5). The influence of function was demonstrated by the observed
 INSIDE IN AND ON: TYPOLOGICAL AND PSYCHOLINGUISTIC PERSPECTIVES 111

effect of labeling condition: the use of in was most prevalent when the noun labeling
the inanimate Ground typically names a container (bowl), with concomitantly low rates
of use when the noun labeling the Ground typically names a functional surface (plate)
or a nonfunctional entity (rock or slab) (Figure 6). Finally, the influence of qualitative
physics was indirectly demonstrated via the effects of animacy of the Figure and Ground:
use of in was most prevalent when the Ground was animate, enabling it to exert control
over the location of the Figure, and when the Figure was inanimate, preventing it from
exerting control over its own location.
Taken together, these two studies sketch two complementary views onto the
landscape of human spatial cognition. The first view, that of the semantic typologist,
considers both the unity and diversity of spatial language in order to arrive at a com-
prehensive picture of the set of factors involved in spatial relational meaning. The
second view, that of the psycholinguist, considers the separable effects of a complex set
of interacting factors on the uses of spatial relational terms. Both views suggest roles
for three families of attributes of spatial scenes: geometric, functional, and qualitative
physical. In combination, these three types of attributes can form the basis for a new
representation of spatial relational meaning which, with one eye on typology and one
on psycholinguistics, may better account for the uses of spatial relational terms than
any one type of factor alone.

Notes
1 Following Talmy (1983), I will be referring to the located object, alternately called the
trajector, or TR (Langacker, 1987), as Figure, and the reference object, alternately called
the landmark, or LM, as Ground.
2 Data on genetic affi liations from Ethnologue, produced by the Summer Institute of
Linguistics: http://www.ethnologue.com.
3 There is a difference of opinion among scholars as to whether or not Korean is related to
Japanese. Further, Korean is possibly distantly related to Altaic.
4 Although inclusion is listed here as a geometric attribute, its presence bears on both
functional and qualitative physical inferences, as will be discussed below.
5 The remaining seven terms fall into an eighth class, general spatial terms, which do not
encode any specific attribute values. For details, see Feist (2000, 2004, 2008).
6 Note that this is not the case for the other geometric attributes. For example, although
contact tends to co-occur with support across a variety of situations, the two attributes
can easily be dissociated (e.g., in the case of two boxes side-by-side on the floor – they
are in contact, but neither supports the other (Feist, 2000)).

References
Bennett, D. C. (1975) Spatial and Temporal Uses of English Prepositions. London:
Longman.
Bowerman, M. and Choi, S. (2001) Shaping meanings for language: Universal
and language specific in the acquisition of spatial semantic categories. In M.
112 LANGUAGE, COGNITION AND SPACE

Bowerman and S. C. Levinson (eds) Language Acquisition and Conceptual

Deveolpment. Cambridge, UK: Cambridge University Press.
Bowerman, M. and Pederson, E. (1992) Cross-linguistic perspectives on topological
spatial relationships. Paper presented at the 91st Annual Meeting of the American
Anthrolopogical Association, San Francisco, CA.
Bowerman, M. and Pederson, E. (1996) Cross-linguistic perspectives on topological
spatial relationships. Unpublished manuscript.
Brown, P. (1994) The INs and ONs of Tzeltal locative expressions: The semantics of
static descriptions of location. Linguistics 32: 743–790.
Carlson, L. A., Regier, T., Lopez, B. and Corrigan, B. (2006) Attention unites form
and function in spatial language. Spatial Cognition and Computation 6: 295–308.
Cienki, A. J. (1989) Spatial Cognition and the Semantics of Prepositions in English,
Polish, and Russian. Munich, Germany: Verlag Otto Sagner.
Clark, H. H. (1973) Space, time, semantics, and the child. In T. E. Moore (ed.)
Cognitive Development and the Acquisition of Language 27–63. New York:
Academic Press.
Coventry, K. R. (1998) Spatial prepositions, functional relations, and lexical speci-
fication. In P. Olivier and K.-P. Gapp (eds) The Representation and Processing of
Spatial Expressions. Mahwah, NJ: Lawrence Erlbaum Associates.
Coventry, K. R., Carmichael, R. and Garrod, S. C. (1994) Spatial prepositions, object-
specific function, and task requirements. Journal of Semantics 11: 289–309.
Coventry, K. R. and Garrod, S. C. (2004) Saying, Seeing and Acting: The Psychological
Semantics of Spatial Prepositions. London: Psychology Press.
Coventry, K. R. and Prat-Sala, M. (2001) Object-specific function, geometry, and
the comprehension of in and on. European Journal of Cognitive Psychology 13:
509–528.
Coventry, K. R., Prat-Sala, M. and Richards, L. (2001) The interplay between geom-
etry and function in the comprehension of over, under, above, and below. Journal
of Memory and Language 44: 376–398.
Croft, W. (1999) Some contributions of typology to cognitive linguistics, and vice
versa. In T. Janssen and G. Redeker (eds) Foundations and Scope of Cognitive
Linguistics 61–94. Berlin: Mouton de Gruyter.
Croft, W. (2001) Radical Construction Grammar: Syntactic Theory in Typological
Perspective. Oxford: Oxford University Press.
Croft, W. (2003) Typology and Universals. (2nd ed.) Cambridge: Cambridge
University Press.
Croft, W. and Poole, K. T. (2008) Inferring universals from grammatical variation:
Multidimensional scaling for typological analysis. Theoretical Linguistics 34: 1–37.
Feist, M. I. (2000) On in and on: An investigation into the linguistic encoding of spatial
scenes. Northwestern University, Evanston, IL.
Feist, M. I. (2002) Geometry, function, and the use of in and on. Paper presented
at the Sixth Conference on Conceptual Structure, Discourse, and Language,
Houston, TX.
Feist, M. I. (2004) Talking about space: A cross-linguistic perspective. In K. D.
Forbus, D. Gentner and T. Regier (eds) Proceedings of the Twenty-sixth Annual
 INSIDE IN AND ON: TYPOLOGICAL AND PSYCHOLINGUISTIC PERSPECTIVES 113

Meeting of the Cognitive Science Society 375–380. Mahwah, NJ: Lawrence

Erlbaum Associates.
Feist, M. I. (2005a) In support of in and on. Paper presented at New Directions in
Cognitive Linguistics. Brighton, UK.
Feist, M. I. (2005b) Semantics and pragmatics of a general spatial term: The case of
Indonesian di. Paper presented at the 2005 Annual Meeting of the Linguistic
Society of America, Oakland, CA.
Feist, M.I. (2008) Space between languages. Cognitive Science 32: 1177–1199.
Feist, M. I. and Gentner, D. (1998) On plates, bowls, and dishes: Factors in the use of
English IN and ON. In M. A. Gernsbacker and S. J. Derry (eds) Proceedings of the
Twentieth Annual Conference of the Cognitive Science Society 345–349. Mahwah,
NJ: Lawrence Erlbaum Associates.
Feist, M. I. and Gentner, D. (2003) Factors involved in the use of in and on. In R.
Alterman and D. Kirsh (eds) Proceedings of the Twenty-fifth Annual Meeting
of the Cognitive Science Society 390–395. Mahwah, NJ: Lawrence Erlbaum
Associates.
Forbus, K. D. (1983) Qualitative reasoning about space and motion. In D. Gentner
and A. L. Stevens (eds) Mental Models. Hillsdale, NJ: Lawrence Erlbaum.
Forbus, K. D. (1984) Qualitative process theory. Journal of Artificial Intelligence 24:
85–168.
Gentner, D. and Bowerman, M. (1996) Crosslinguistic differences in the lexicalization
of spatial relations and effects on acquisition. Paper presented at the Seventh
International Congress for the Study of Child Language, Istanbul, Turkey.
Gentner, D. and Bowerman, M. (2009) Why some spatial semantic categories are
harder to learn than others: The Typological Prevalence hypothesis. In J. Guo,
E. Lieven, S. Ervin-Tripp, N. Budwig, S. Özçaliskan and K. Nakamura (eds)
Crosslinguistic approaches to the psychology of language: Research in the tradition
of Dan Isaac Slobin 465–480. New York, NY: Lawrence Erlbaum Associates.
Haspelmath, M. (2003) The geometry of grammatical meaning: Semantic maps
and cross-linguistic comparison. In M. Tomasello (ed.) The New Psychology of
Language Volume 2 211–242. Mahwah, NJ: Lawrence Erlbaum Associates.
Herskovits, A. (1986) Language and Spatial Cognition: An Interdisciplinary Study of
the Prepositions in English. Cambridge, UK: Cambridge University Press.
Labov, W. (1973) The boundaries of words and their meanings. In C-J. N. Bailey and
R. W. Shuy (eds) New ways of analyzing variation in English. Washington, DC:
Georgetown University Press.
Landau, B. (1996) Multiple geometric representations of objects in languages and
language learners. In P. Bloom, M. A. Peterson, L. Nadel and M. F. Garrett (eds)
Language and Space 317–363. Cambridge, MA: MIT Press.
Landau, B. and Jackendoff, R. (1993) ‘What’ and ‘where’ in spatial language and
spatial cognition. Behavioral and Brain Sciences 16: 217–265.
Landau, B. and Stecker, D. S. (1990) Objects and places: Geometric and syntactic
representations in early lexical learning. Cognitive Development 5: 287–312.
Langacker, R. W. (1987) Foundations of Cognitive Grammar: Theoretical Prerequisites.
Stanford: Stanford University Press.
114 LANGUAGE, COGNITION AND SPACE

Levinson, S. C. (1996) Relativity in spatial conception and description. In J. J.

Gumperz and S. C. Levinson (eds) Rethinking linguistic relativity 177–202.
Cambridge: Cambridge University Press.
Levinson, S. C., Meira, S. and The Language and Cognition Group. (2003) ‘Natural
concepts’ in the spatial topological domain – adpositional meanings in crosslin-
guistic perspective: An exercise in semantic typology. Language 79: 485–516.
Lindkvist, K.-G. (1950) Studies on the Local Sense of the Prespositions in, at, on, and to
in Modern English. Lund, Sweden: C.W.K. Gleerup.
Majid, A., Bowerman, M., Kita, S., Haun, D. B. M. and Levinson, S. C. (2004) Can
language restructure cognition? The case for space. Trends in Cognitive Science 8:
108–114.
Miller, G. A. and Johnson-Laird, P. N. (1976) Language and Perception. Cambridge,
MA: Belknap Press.
O’Keefe, J. (1996) The spatial preposition in English, vector grammar, and the cogni-
tive map theory. In P. Bloom, M. A. Peterson, L. Nadel and M. F. Garrett (eds)
Language and Space 277–316. Cambridge, MA: MIT Press.
Pederson, E., Danziger, E., Wilkins, D., Levinson, S. C., Kita, S. and Senft, G. (1998)
Semantic typology and spatial conceptualization. Language 74: 557–589.
Sinha, C. and Thorseng, L. A. (1995) A coding system for spatial relational reference.
Cognitive Linguistics 6: 261–309.
Sinha, C., Thorseng, L. A., Hayashi, M. and Plunkett, K. (1994) Comparative spa-
tial semantics and language acquisition: Evidence from Danish, English, and
Japanese. Journal of Semantics 11: 253–287.
Talmy, L. (1983) How language structures space. In H. Pick and L. Acredolo (eds)
Spatial Orientation: Theory, Research, and Application. New York: Plenum Press.
Talmy, L. (1988) Force dynamics in language and cognition. Cognitive Science 12:
49–100.
Tyler, A. and Evans, V. (2003) The Semantics of English Prepositions. Cambridge:
Cambridge University Press.
Vandeloise, C. (1991) Spatial Prepositions: A Case Study from French (A. R. K. Bosch,
Trans.) Chicago: University of Chicago Press.
Vandeloise, C. (1994) Methodology and analyses of the preposition in. Cognitive
Linguistics 5: 157–184.
Vandeloise, C. (2003) Containment, support, and linguistic relativity. In H. Cuyckens,
R. Dirven and J. Taylor (eds) Cognitive Approaches to Lexical Linguistics 393–425.
Berlin and New York: Mouton de Gruyter.
Vandeloise, C. (this volume) Genesis of spatial terms. In V. Evans and P. Chilton (eds)
Language, Cognition and Space: The State of the Art and New Directions. London:
Equinox.
5 Parsing space around objects
Laura Carlson

1.0 Introduction

Imagine you are planning to get together with a friend after a class, and that you arrange
to meet in front of the lecture hall. After class, you momentarily position yourself at the
front door, but because there are a lot of students milling in and out of the building, you
walkway halfway down the steps and sit down. While you wait, you become very warm
in the sun, and move down past the steps to sit on a bench off to the side under a tree
to await your friend. Although you are now quite a distance from the building, you feel
confident that your friend will find you. The research project described in the current
chapter investigates how all of these locations (at the door of the building, halfway
down the steps, and on the bench to the side of the building) are understood to be ‘in
front of’ the lecture hall. The chapter begins with an overview of the cognitive processes
and representations that assist in defining projective spatial terms such as front. This is
followed by a brief summary of the previous attempts at studying the regions denoted by
such terms. A new methodology is described that addresses some limitations with these
previous approaches. The utility of the new methodology is established by demonstrating
that various factors known to affect the interpretation of spatial terms also impact the
size and shape of the region denoted by front. These factors include: the identity of the
objects, the functional characteristics of the objects, the presence of additional objects
in the scene, and the reference frame used to define the spatial term.

2.0 Projective spatial terms

One way to describe the location of a target object is by spatially relating it to an object
whose location is known, as in ‘My friend is at the front of the lecture hall’. In this utter-
ance, ‘my friend’ is designated the located object (also known variously as figure, locatum
or trajector); finding this object is the presumed goal of the utterance. The ‘lecture hall’
is designated the reference object (or variously, relatum, ground, or landmark), an object
whose position is presumed known or easily found (see Miller and Johnson-Laird, 1976;
Levelt, 1996; Talmy, 1983; Tyler and Evans, 2003). Understanding of this statement
involves not only linking the objects in the world to the referents in the sentence, but also
mapping the spatial relational term (i.e., front) to the appropriate region of space around
the reference object. Terms such as front belong to the class of projective spatial terms
that convey direction information; these contrast with terms such as ‘near’ that belong
to the class of proximal terms that convey distance information (for an organizational
chart, see Coventry and Garrod, 2004).

115
116 LANGUAGE, COGNITION AND SPACE

Logan and Sadler (1996) present a computational framework that specifies the proc-
esses and representations involved in mapping spatial terms such as front onto regions
of space around the reference object. For the current chapter, two representations and
their constituent processes are of interest: reference frames and spatial templates. For
projective terms, a reference frame consists of a set of three axes that assign directions
onto space. Specifically, the reference frame is imposed on the reference object, and its
axes extend outward defining the space beyond. One set of axes corresponds to the verti-
cal dimension, and its endpoints delineate space above and below the reference object.
Two sets of axes correspond to the horizontal dimensions, with one set delineating
space ‘in front of’ and ‘in back of’ the reference object, and the other set delineating space
‘to the right of’ and ‘to the left of’ the reference object. (Garnham, 1989; Levelt, 1984;
Levinson, 1996; Miller and Johnson-Laird, 1976). Figure 1, Panel A shows a reference
frame imposed on a reference object (a chair) with its axes defining the space around
the reference object.
Reference frames are flexible representations that have a set of parameters that
define their use in a given context (Logan and Sadler, 1996; for a summary of evidence,
see Carlson, 2004). These parameters include the origin, orientation, direction and
distance. The origin parameter defines where on the reference object the reference
frame is imposed. For example, in Figure 1, Panel A, the reference frame is imposed
at the center of the chair. More generally, the origin may be defined on the basis of the
reference object’s geometry and/or on the basis of its functional properties (Carlson-
Radvansky, Covey and Lattanzi, 1999). The orientation parameter determines how to
orient the vertical and horizontal axes. For example, in Figure 1, Panel A, the vertical
axis is aligned with the top/bottom of the chair and with the top/bottom of the picture
plane. The direction parameter determines the endpoints of these axes (e.g., the above
and below endpoints of the vertical axis). In Figure 1, Panel A, the endpoint of the
vertical axis closest to the top of the chair is above; the endpoint closest to the bottom
of the chair is below. The orientation and direction parameters can be defined on
the basis of various sources of information, such as the environment, the object, or
interlocutor (speaker or addressee) (Logan and Sadler, 1996), with the source typically
taken to define the type of reference frame in use (i.e., absolute, intrinsic or relative,
respectively; see Levinson, 1996). For example, in Figure 1, Panel B, the chair is rotated
90 degrees to the right. It is now possible to define two different axes as vertical – one
corresponding to the top/bottom of the picture plane (absolute) and one corresponding
to the top/bottom of the chair (intrinsic). Finally, the distance parameter indicates the
spatial extent of the region, and is defined at least in part by properties of the objects
and the spatial term relating them. For example, Carlson and Covey (2005) asked
participants to imagine sentences such as ‘The squirrel is in front of the flowers’. The
main task was to provide an estimate for how far apart the objects were in their image.
Distance estimates varied systematically as a function of the spatial term relating the
objects, both for proximal terms such as ‘near’ and ‘far’ (as might be expected by the
semantics of these terms) and for projective terms such as front and back, with front
estimates consistently smaller than back estimates.
 PARSING SPACE AROUND OBJECTS 117

a Back Above

Left Right

Front
Below

Above
b

Below Above

Below

Above

Below

Figure 1. Panel A. A reference frame imposed on a reference object (chair), with its axes defining the
space around the reference object. Panel B. With the reference object rotated, above and below can
either be assigned to the vertical axis consistent with the top/bottom of the picture plane or the hori-
zontal axis consistent with the top/bottom of the chair. Panel C. An illustration of a possible spatial
template representing the region above with respect to the chair. Note that the size and shape of this
region are speculative.

A second representation that has been subsequently considered an additional parameter

of a reference frame (Carlson-Radvansky and Logan, 1997) is a spatial template. Spatial
templates can be thought of as an illustration of the spatial region around a reference
object for a given spatial term. For example, Figure 1, Panel C shows a possible spatial
template for above for the chair in Figure 1, Panel A, that extends upward and outward
from the topside of the chair. Note, however, that the size and shape of this region are
speculative. One of the critical goals of the current chapter is to explore the factors that
help determine the shape and spatial extent of such regions.
Spatial templates reflect two important assumptions of the mapping of the term onto
space around the reference object: first, that the use of a spatial term does not correspond
118 LANGUAGE, COGNITION AND SPACE

to a single location in space but rather encompasses a region. Theoretically, this idea is
consistent with Miller and Johnson-Laird (1976) conceptualization of the location of
an object as containing an area of space immediately surrounding it, referred to as its
penumbra or region of interaction (Morrow and Clark, 1988). Two objects are said to
be in a spatial relation with each other when these areas overlap (see also Langacker,
1993, 2002). Second, a spatial term does not apply equally across this region (Logan
and Sadler, 1996; Hayward and Tarr, 1995); rather, some locations are preferred over
others. Theoretically, this idea is consistent with Hayward and Tarr’s (1995) treatment
of the meaning of spatial terms as category representations with prototypes and graded
membership (see also the idea of a preferred subspace, Herskovits, 1986 and a proto-
scene, Tyler and Evans, 2003).

3.0 Previous empirical approaches to understanding spatial regions

There have been theoretical treatments of spatial regions (e.g., Herskovits, 1986; Miller
and Johnson-Laird, 1976; Vandeloise, 1991). For example, Herskovits (1986) suggests
that there are ideal meanings of spatial prepositions based on geometric descriptions
that are then applied in a given context via particular use types that may serve to modify
these underlying descriptions (for similar ideas, see Coventry and Garrod, 2004; Tyler
and Evans, 2003). As applied to the particular regions associated with projective terms,
she discusses how context, the presence of additional objects, and characteristics of the
objects may serve to identify preferred sub-spaces. However, there has been relatively
little systematic empirical work supporting these ideas that explicitly addresses how the
size and shape of the regions are defined, and whether they can be modified. In this
section, I’ll briefly describe two empirical approaches adopted in the previous literature
that examine spatial regions, and discuss some of their limitations as applied to the
questions of interest. The new methodology presented in Section 4.0 will integrate
components from each of these approaches.

3.1 Spatial templates

One approach to examining spatial regions has been to ask participants to rate the
acceptability of a spatial term as describing placements of located objects at various
locations around the reference object (Hayward and Tarr, 1995; Logan and Sadler, 1996).
For example, Logan and Sadler (1996) presented participants with a located object (the
letter ‘O’) that was placed across trials within various locations in an invisible 7 X 7 grid
whose center contained a reference object (the letter ‘X’). The task was to rate how well
a given spatial term described the relation between the two objects. Figure 2, Panel A
shows a sample trial. The endpoints of the scale were labeled as 1 = bad and 9 = good;
intermediate values were permitted. The ratings were then plotted as a function of the
placement of the located object within the grid, as shown in Figure 2, Panel B. The divot
 PARSING SPACE AROUND OBJECTS 119

in the plot at cell 4,4 corresponds to the location of the reference object. Logan and
Sadler (1996) referred to this plot as a spatial template.

Rate: X is above O

b Good Region

Acceptable Region

Acceptability
Rating

Bad Region

Row
Column

Figure 2. Panel A. A sample trial from Logan and Sadler (1996) in which participants are asked to rate
how well the sentence ‘X is above O’ describes the relative locations of the X and O in the display. The
7X7 grid defines the possible placements of the X across trials; the grid was not visible to participants.
Panel B. The spatial template for above from data from Logan and Sadler (1996). See text for descrip-
tions of the good, acceptable and bad regions.
120 LANGUAGE, COGNITION AND SPACE

Within the spatial template three distinct regions were identified. The peak of the
template comprised the ‘good’ region, and corresponded to placements along the
axis. The regions flanking the peak showed a drop in acceptability, and comprised
the ‘acceptable’ regions. Finally, the flat area at the opposite endpoint with uniformly
low acceptability ratings comprised the ‘bad’ region. Logan and Sadler (1996) derived
spatial templates for many different terms, and showed that the overall shapes and
sizes for projectives such as above and left were remarkably similar, differing only in
orientation. Hayward and Tarr (1995), Carlson-Radvansky and Logan (1997), and Regier
and Carlson (2001) obtained spatial templates using a variety of reference objects, and
showed global similarity in their size and shape, suggesting a common underlying
mechanism that defined the space independently of the objects. For example, Regier
and Carlson (2001) presented the attention vector sum model (AVS) in which the shape
of the spatial template was defined by the joint components of attention and a vector
sum representation of direction. Carlson, Regier, Lopez and Corrigan (2006) further
extended AVS to incorporate the functional characteristics of the objects, thus making
it sensitive to the objects being related.
Although this approach offers a means of visualizing the region associated with a
given spatial term, it has several possible limitations. First, it is not clear that a rating of
acceptability will translate into actual use in a more naturalistic task. While one might
infer that ratings at the endpoints may indicate whether a speaker would use the term
(presumably yes for placements in the good region, and no for placements in the bad
region), it is not clear whether intermediate ratings (as in the acceptable region) would
necessarily translate into selection, nor whether such selection would be constant within
the acceptable region. Second, the spatial extent of the template is largely defined by
the experimenter before the study during construction of the grid that contains the
placements of the located object. This is potentially problematic, as the grid may not
directly encompass the boundaries of a given region. For example, in the Logan and
Sadler (1996) plot for above (Figure 2, Panel B), there is no drop-off in acceptability as
a function of distance within the good region, suggesting that all placements within
this region are acceptable. However, Regier and Carlson (2001) found that ratings did
vary within the good region when a reference object with spatial extent was probed at
multiple locations using different distances. Moreover, Carlson and Van Deman (2004)
showed faster response times to placements of the located object in the good region that
were closer to the reference object than those that were farther away from the reference
object, indicating a potential effect of distance. Thus, it is not clear that the edge of the
spatial template as constructed by the experimenter will necessarily reflect the bound-
ary of the region. A final limitation is that spatial templates have been collected using
a 2D projection of space; however, most of our everyday use of these projective terms
involves mapping them onto 3D space. It is not clear whether such 2D projections will
necessarily generalize to the 3D case.
 PARSING SPACE AROUND OBJECTS 121

3.2 Regions around oneself

A second approach to examining the spatial regions defined by projective spatial

terms has been to ask participants to explicitly divide the space around their bodies.
Franklin, Henkel and Zangas (1995) had participants stand in the center of a circular
room constructed by hanging a curtain from the ceiling, thereby eliminating any
visible sides. Their task was to indicate the boundaries of the regions corresponding
to front, back, left and right around themselves. To do this, they held a pointer, and
were asked to move the pointer as far to one side as possible while staying within a
given region. For example, to identify the left edge of the front region, participants
were told to move the pointer to the left as far as possible but so that the pointer still
indicated front. The pointer extended outside the curtain where a protractor was
printed on the floor; this enabled the experimenter to record the angle corresponding
to the pointer, and to identify the size of the various regions. For example, for the left
edge of the front region, if a participant was facing 60 degrees and placed the pointer
at 110 degrees, this would indicate that the front region extended 50 degrees to the left
from their midline. Across trials, each participant indicated each boundary of each
region. This enabled Franklin et al. (1995) to determine the sizes of each region; a
schematic is shown in Figure 3 with the viewer standing in the center, facing the top
o the page. Franklin et al. (1995) found that the space around oneself was not divided
into equally spaced 90 degree regions. Rather, the front region was the largest (124
degrees), followed by the back region (110 degrees) and then left and right regions
(91 and 92, respectively). The regions for front and back did not differ significantly
from each other, but both were significantly larger than the regions for left and right,
which also did not differ from each other.

FRONT

Left Right

Back

Figure 3. A schematic of the front, back, left and right regions obtained by Franklin, Henkel and
Zangas (1995).
122 LANGUAGE, COGNITION AND SPACE

This approach offers a direct means of assessing the boundaries of the various regions.
Interestingly, the overlap between the front and left/right regions indicates that there
areas in which multiple spatial terms may be possible, consistent with an interpretation
of these spatial terms as having fuzzy boundaries within a categorical representa-
tion (Hayward and Tarr, 1995; for other work on overlapping regions, see Carlson-
Radvansky and Logan, 1997; Herskovits, 1986). Another benefit of this approach is
that it was conducted in 3D space. However, there are also several limitations. First,
it does not enable one to assess the spatial extent of the regions. Participants did not
indicate a distance associated with the regions, but only the borders of the regions.
Indeed, a constant distance was imposed by the walls of the room, and there was an
implicit assumption that the regions would extend at a minimum to the walls at all
angles (see Figure 3). Yet, the spatial template data from Logan and Sadler (1996)
suggest that this may not be correct. Specifically, the plot in Figure 2, Panel B, suggests
that ratings drop as a function of angle and distance, as one moves from the good
to the acceptable region, and one moves within the acceptable region. Moreover,
the identity of the objects may impact the extent of these regions as well. Miller and
Johnson-Laird (1976) suggest that objects evoke distance norms that represent typical
values derived from interactions with other objects. Morrow and Clark (1988) refer
to these areas as zones of interaction. Carlson and Covey (2005) showed that the
distances inferred between two objects that were spatially related depended not only
on the spatial term used to describe the relation (e.g., front versus back as discussed in
Section 2.0), but also on the size and shapes of the objects. For example, the distance
estimates inferred for descriptions relating large objects (i.e., ‘The St. Bernard is in
front of the tree.’) were consistently larger than the distance estimates inferred for
descriptions relating small objects (i.e., ‘The squirrel is in front of the flower.’). This
finding suggests that the size of the regions observed by Franklin et al (1995) may
be specific to the space surrounding the particular object that they investigated (the
participants themselves), and may not generalize to other objects.

4.0 A new methodology for determining size and shape of the spatial region

4.1 Combining components from previous approaches

In addition to the limitations specific to each of the two approaches described in section
3.0, more generally, there has been no systematic attempt to examine how spatial regions
may vary as a function of factors asserted to be relevant (Herskovits, 1986; Langacker,
1987; Miller and Johnson-Laird, 1976; Tyler and Evans, 2003; Vandeloise, 1991), and
known to impact other aspects of spatial term interpretation. These factors include the
identity of the objects (Carlson-Radvansky, et al, 1999), its functional characteristics
(Carlson et al., 2006; Carlson-Radvansky et al., 1999; Coventry and Garrod, 2004), the
presence of additional objects (Carlson and Logan, 2001; Carlson and Hill, submitted)
and the type of reference frame used to define the spatial term (Carlson-Radvansky
 PARSING SPACE AROUND OBJECTS 123

and Irwin, 1993; Carlson-Radvansky and Radvansky, 1996). This section presents a
new methodology to assess the role that these factors may play in defining these spatial
regions. The validity of the new methodology can be evaluated by determining whether
it is sensitive to manipulations of these established factors.
The new methodology builds on features of the two previous approaches described
in Section 3. Specifically, from the spatial template approach, we adopt the idea of
probing at multiple locations, drawn from the best, acceptable and bad regions.
However, rather than collect acceptability ratings, we asked participants to directly
indicate placement that corresponds to the best (prototypic) use of a given spatial
term at various angles from the reference object. In this manner we obtain a direct
measure of this distance at multiple locations. From the Franklin et al. (1995) para-
digm we adopt the idea of directly assessing the boundaries, asking participants to
indicate the farthest points at which a spatial term still applies. Combining the best
and farthest measures enables us to obtain a fairly direct means of representing the
spatial extent of the regions. In addition, we also asked participants to indicate whether
alternative terms also apply at given locations, as a way of getting at the fuzziness
of the boundaries and the degree of overlap with other spatial terms. We developed
the methodology in a number of studies by focusing on the size and shape of front,
given that this spatial term is considered privileged relative to the other horizontal
relations (i.e., back, left and right (Clark, 1973; Fillmore, 1971; Franklin, Henkel and
Zengas, 1995; Garnham, 1989).

4.2 The specific methodology and data

To measure the spatial region corresponding to front, we placed a small dollhouse

cabinet that served as a reference object in the center of a 102 cm X 82 cm uniform
white foam board. The white board was placed on a large conference table, and the
participant was seated at one end of the table facing the cabinet. The set up is shown
in Figure 4. Eleven lines were identified that radiated out from the reference object,
numbered from left to right, counterclockwise, as shown in Figure 5. These 11 lines
can be defined in terms of angular deviation (0 – 90 degrees, in either direction; i.e.,
unsigned) from the front of the cabinet, and categorized with respect to the regions
(good, acceptable, bad) within which they fall on the front spatial template of Logan
and Sadler (1996). Specifically, Lines 5–7 were directly in front of the cabinet, at 0
degrees, located in the good region, with lines 5 and 7 at the edges of the region, and
line 6 at its center. Lines 2 and 8 were each at 22.5 degrees, lines 3 and 9 were each at
45 degrees and lines 4 and 10 were each at 67.5 degrees from the front of the cabinet;
these lines all fell into the acceptable region. Finally, lines 1 and 11 were each at 90
degrees from the front of the cabinet, located in the bad region. The lines were not
visible to participants; marks on the edges of the white foam board indicated the
endpoints of the lines for the experimenter.
124 LANGUAGE, COGNITION AND SPACE

Figure 4. Experimental set-up in which the participant is seated in front of a white board contain-
ing the reference object, and indicates distance judgements associated with the spatial term by
pointing to a location on the dowel. The dowel is marked in centimeters on the side not visible to
participants.

1
11

2 10
3 4 8 9
5 6 7
Figure 5. The cabinet is the reference object; measures were collected along each of lines 1 – 11 in a
random order. Lines 1 and 11 are within the bad region; Lines 2–4 and 8–10 are within the acceptable
region; Lines 5–7 are within the good region.

On each trial, the experimenter placed a square dowel rod in position along one of the
11 lines. The lines were not visible to the participant; the experimenter lined up the rod
with one end at the cabinet and the other end at a mark on the edge of the white foam
board (not visible to the participant) that indicated the appropriate angle. One side of
the dowel was marked in centimeters, starting with 0 at the cabinet, and ending at 67 cm,
just past the edge of the white board. For each placement of the dowel, participants were
asked to make three judgments pertaining to their definition of front with respect to the
cabinet. First, participants indicated by pointing to a location on the dowel the distance
that corresponded to the best use of front. The experimenter read this value on the dowel,
and recorded it. This measure was intended to define the ideal or prototypical distance
(Hayward and Tarr, 1995). Second, participants indicated along the dowel the farthest
distance for which front would still be deemed acceptable. The experimenter read this
 PARSING SPACE AROUND OBJECTS 125

value on the dowel, and recorded it. This measure was intended to define the outer extent
of the front region. Third, at this farthest distance, participants were asked to indicate
whether they would prefer to use an alternative spatial term to describe an object at this
location rather than front. If there was a preference for an alternative term, the participants
reported the term, and were instructed to move back along the dowel toward the cabinet
and indicate the point at which front became preferred to this alternative. This measure was
intended to take into account the fact that some locations could be defined with respect
to competing spatial terms. As such, the farthest distance at which front may be used may
not be the same as the farthest distance at which front is preferred relative to these other
terms. That is, just because a participant may indicate that front may be acceptable, this
does not necessarily mean that a participant would use the term front to describe an object
at that location. Each participant provided these three measures for each of the 11 lines,
with the sequence of lines randomly determined for each participant.
The best and farthest data can be summarized by averaging the values per line
across participants and then plotting these means on Lines 1–11. Connecting the means
reveals the spatial regions defined by the best and farthest distances, as shown in Figure
6. With respect to the competing term measure, the data are interesting but complicated,
given that not all participants supplied competing terms, or indicated a new preferred
front distance. For the sake of the current chapter, we will focus on the best and farthest
measures only. In the studies that we describe in the next section, we were interested
in how these regions changed as a function of various manipulations that are known
to impact the interpretation of projective spatial relations.

Best
Farthest

1 11

2
10

9
3

5 6 7
Figure 6. Sample best and farthest front data, averaged across participants and connected to form
the corresponding regions.
126 LANGUAGE, COGNITION AND SPACE

5.0 Factors that impact the size and shape of the spatial region

Using the basic methodology described in Section 4, across experiments we examined

the impact of a diverse set of factors on the size and shape of the front region, including
the identity of the objects, the functional characteristics of the objects, the presence
of additional objects in the display, and the type of reference frame that defined the
spatial term.

5.1 Identity of the reference object

Past theoretical and empirical work (Tyler and Evans, 2003; Herskovits, 1986; Langacker,
1987; Miller and Johnson-Laird, 1976; Vandeloise, 1991) has shown that the identity
of the reference object has a significant impact on the way in which a spatial term is
applied to space around it. Such influence has been observed with respect to where the
reference frame is imposed on the reference object (Carlson-Radvansky et al, 1999), with
respect to the distance inferred between the objects (Carlson and Covey, 2005; Morrow
and Clark, 1988); and with respect to the locations that are deemed acceptable around
the object (for extensive review of object effects, see Coventry and Garrod, 2004). At
the outset of this project we were interested in understanding how the same physical
space (the white board) may be interpreted differently as a function of the reference
object being used. Previous work has shown that the absolute size of the reference object
makes a difference in the distances associated with spatial terms, with shorter distances
associated with smaller objects (Carlson and Covey, 2005; Morrow and Clark, 1988).
We were interested in whether conceptual size would have a similar impact. To assess
this, we contrasted a scaled-down model object (dollhouse cabinet) with an actual size
object (lotion bottle), closely equating the physical sizes of the objects.1 These objects
are shown in Figure 7. If participants scale the whiteboard space with respect to the

Figure 7. Dollhouse cabinet and lotion bottle matched in actual size and shape but mismatched in
conceptual size.
 PARSING SPACE AROUND OBJECTS 127

conceptual size of the object, then we should observe differences in the best and farther
measures for the front region; however, if participants define the front region with respect
to physical size, there should be no differences in these measures.
Figure 8 shows the best regions for the cabinet (Panel A) and the bottle (Panel B),
with the two regions superimposed and flipped vertically to enable an assessment of
how well the regions overlap (Panel C). Figure 9 shows the data in an alternate form,
plotting the best distances for each object as a function of line in panel A, and the farthest
distances for each object as a function of line in panel B. We excluded lines 1, 2, 10,
11 from the plots because most responses for these lines corresponded to a 0 distance,
reflecting the fact that front would not be used for positions on these line. These plots
render differences among the contrasting conditions easiest to see; accordingly, the data
in the remaining sections will be presented and discussed in this manner. There was no
difference in the best measure as a function of object (cabinet or lotion bottle). However,
in the farthest measure, a significant difference occurred within the good region (lines
5,6,7), with front extending farther for the lotion bottle than the cabinet. This suggests
that participants may have been scaling the size of the regions to the conceptual size
of the objects, with the actual size object having a larger zone of interaction than the
model sized object.

Figure 8. Panel A. Plot of best front for the cabinet. Panel B. Plot of best front for lotion bottle. Panel C.
Superimposed plots for cabinet and lotion bottle; differences in size and shape are of interest.
128 LANGUAGE, COGNITION AND SPACE

Figure 9. Panel A. Plot of best front as a function of line for cabinet and lotion bottle. Panel B. Plot for
farthest front as a function of line for cabinet and lotion bottle.

5.2 Functional parts of the reference object

Previous research has suggested that not only the identity of the reference object but
also the manner in which the reference and located objects interact may impact the
way in which spatial terms are mapped onto space. For example, Carlson-Radvansky
et al. (1999) demonstrated that the best placement for a located object was influenced
by the location of a prominent functional part of the object. Carlson and Kenny (2006)
 PARSING SPACE AROUND OBJECTS 129

further showed that this influence depended not only on the association between the
objects, but also upon their ability to interact. To see whether functional effects would
be observed in this new methodology, we contrasted best and farthest measures for front
for two versions of the dollhouse cabinet that differed only with respect to the side at
which the door opened. The two cabinets are shown in Figure 10, Panel A. If the way in
which one might interact with the object impacts the way in which the front region is
defined, then one would expect the best measure to be influenced by the side at which
the door opens. Specifically, the good region should not extend as far on the opening
side, as one needs to be closer to that side in order to interact with (e.g., reach into) the
cabinet. In contrast, if one investigated the back region with respect to the cabinet, one
would not expect such a difference due to side, because one typically doesn’t interact
with the back of the cabinet. Figure 10, Panel B shows the backs of the cabinets – note
that the door handles can be seen; in addition the cabinets were presented with the
doors slightly ajar. Thus, information about the doors was available for defining the back
regions. However, the prediction was that this would not influence the size or shape of
the regions. Finally, the predicted effect on the front region was expected to be limited
to the best measure; the farthest measure reflects the putative boundary of front, and
would be presumably beyond the area of interaction with the object.

Door handle on left Door handle on right

(Line 5) (Line 7)

Figure 10. Panel A. Two dollhouse cabinets. The one on left has door handle on the left (from the reader’s
perspective), aligned with line 5. The one on the right has the door handle on the right (from the reader’s
perspective), aligned with line 7. Panel B. The corresponding backs of the dollhouse cabinets; backs were
removed and doors were slightly ajar so that participants could see how the doors would open.
130 LANGUAGE, COGNITION AND SPACE

Figure 11, Panels A and B show the best and farthest data for front. The best data clearly
show an interaction with the door of the cabinet. When the door opens on the right (from
the reader’s perspective) (dotted line), the location of the best front along the line close to the
handle (line 7) was closer to the cabinet than for line close to the opposite edge of the cabinet
(line 5); however, when the door opened on the left (from the reader’s perspective) (solid
line), the location of the best front was closer to the cabinet by the door (line 5) than on the
opposite side (line 7). This asymmetry in the best front region is consistent with the way in
which one might interact with the object. Moreover, no such asymmetry was observed for
the farthest front measure (Panel B), nor for either the best or farther back measures, plotted
in Panels A and B of Figure 12, respectively. These data replicate the influence of functional
parts on defining spatial regions around a reference object within the new methodology.

Figure 11. Plot of best front as a function of line for the two dollhouse cabinets. Panel B. Plot for
farthest front as a function of line for the two dollhouse cabinets.
 PARSING SPACE AROUND OBJECTS 131

Figure 12. Plot of best back as a function of line for the two dollhouse cabinets. Panel B. Plot for
farthest back as a function of line for the two dollhouse cabinets.

5.3 The addition of a located object

In the studies described in sections 5.1 and 5.2, a reference object was placed in the
middle of the display board and participants indicated the best and farthest distances
associated with front along the dowel using their finger. However, most often spatial
descriptions include two objects, the located object and the reference object. In this
study, we asked participants to indicate the best and farthest front by placing a located
object at the desired distance along the dowel. Previous research has shown that the
identity of the located object and the manner in which it interacts with the reference
object has a significant impact on the way in which spatial terms are applied to space
132 LANGUAGE, COGNITION AND SPACE

around the reference object (Carlson and Kenny, 2006; Carlson-Radvansky et al, 1999).
Therefore, we contrasted two located objects: a doll and a dog that were both scaled
relative to the dollhouse cabinet that was used as a reference object (i.e., these were sold
together as a playset). The objects are shown next to the cabinet in Figure 13, Panel A.
Figure 13, Panel B shows a sample participant placing these objects during the task.
Given Carlson and Covey’s (2005) results that the distance associated with a spatial term
depended upon the size of the objects, we expected the best and farthest measures for
front to be smaller for the dog (a smaller object) than for the doll.

Figure 13. Panel A. Doll as located object next to cabinet as reference object (on left) and dog as
located object next to cabinet as reference object (on right). Panel B. Placement of doll (on left) and
dog (on right) during the experimental task.

Figure 14, Panels A and B show the data for the best and farthest measures, respectively.
For the best measure, two effects are readily apparent. First, adding a located object
compresses the best front, relative to the condition in which no object was placed and
participants indicated the distance with their fingers. Second, averaging across lines,
there was a small but reliable difference, such that the best front for the dog was closer
to the cabinet than the best front for the doll. For the farthest measure, there were also
differences due to placing an object. First, relative to not placing an object, the distances
associated with placements of the dog were much smaller, but of the same general shape.
This stands in contrast to the data for the lines in the good region with the best measure
(contrast lines 5–7, Panels A and B) in which distances for placing the dog were much
 PARSING SPACE AROUND OBJECTS 133

reduced. Second, the differences between placing the doll and not placing an object
depended upon line. In the good region (lines 5–7), there was not much difference; in
contrast, in the acceptable and bad regions, the distances dropped off steeply when there
was no object to place but remained relatively large when placing the doll. Third, the
distances for the dog were uniformly shorter across lines than the distances for the doll.
In summary, across both measures there were systematic effects of adding a locating
object, with its characteristics (identity or size) impacting the way in which the term
front was applied to space around the reference object.
a

Figure 14. Panel A. Plot of best front as a function of line for conditions where no object was placed
(finger indicated location), doll was placed and dog was placed. Panel B. Plot of farthest front as a
function of line for conditions where no object was placed (finger indicated location), doll was placed
and dog was placed.
134 LANGUAGE, COGNITION AND SPACE

5.4 Reference frame used to define front

The source of information used to define the axes of a reference frame, and thereby set
the orientation and direction parameters, is typically used to define the type of reference
frame. For example, within Levinson’s (1996) typology, an absolute reference frame
uses features of the environment that are invariant to the viewer or reference objects to
assign directions; the intrinsic reference frame uses the predefined sides of an object to
assign directions; and the relative reference frame uses the perspective of a viewer or
other object in the scene to assign directions to space around the reference object (see
Levinson, 1996, for other notable differences among the types of reference frames). We
were interested in whether the front region defined with respect to an intrinsic reference
frame based on the cabinet would be of a different size or shape than the front region
when defined with respect to a relative reference frame based on the participant. This
is interesting because when the participant faces the cabinet to perform the task (see
Figure 4), their front regions overlap in physical space. Thus, any observed differences
would be due to the way in which the particular reference frames were imposed on the
space, rather than to the space itself.
Figure 15 shows the data for best and farthest front measures for space around a
cabinet in Panels A and B, respectively. There is a small but consistent effect of a larger
‘best’ region when the space was defined with respect to the intrinsic frame based on
the cabinet than with respect to the relative frame based on the participant’s front.
A similar trend was observed when comparing the intrinsic and relative reference
frames with the lotion bottle. These effects may be due to the fact that greater atten-
tion is paid to the object when front is defined by the intrinsic frame than relative
frame, thereby emphasizing its zone of interaction. No such effect was observed in
the farthest measure.

6.0 Conclusion

In this chapter I have presented a new methodology for examining how regions
associated with the projective term front are defined. Several factors that have been
previously shown to impact the interpretation of such spatial terms within alternative
approaches were examined, and initial findings suggest that effects of these factors can
be observed within this methodology as well. The idea that these regions may change
shape and size as a function of characteristics of the objects being related is consist-
ent with a current dominant theme in spatial language research that incorporates
influences of the objects, the context, and goals into one’s interpretation of a spatial
description (Coventry and Garrod, 2004; Tyler and Evans, 2003; more generally, see
Zwaan, 2004).
 PARSING SPACE AROUND OBJECTS 135

Figure 15. Plot of best front as a function of line when using the intrinsic and relative reference
frames to define the spatial term. Panel B. Plot for farthest front as a function of line when using the
intrinsic and relative reference frames to define the spatial term. The cabinet was the reference object.

Acknowledgements
I thank Meghan Murray, Christina Shreiner, and Padraig Carolan for assistance in data
collection, and Aaron Ashley and Patrick Hill for discussion. Address all correspondence to
Laura Carlson, Department of Psychology, 118-D Haggar Hall, University of Notre Dame,
Notre Dame, IN 46556; [email protected]. Portions of this research were presented at the
2004 meeting of the Association for Psychological Science, Chicago, IL.
136 LANGUAGE, COGNITION AND SPACE

Note
1. An additional constraint also directed our selection of these objects: specifically, that
the objects had functional parts that could be moved from one side to the other (the
door of the cabinet; the nozzle on the lotion bottle; see Section 5.2).

References
Carlson, L. A. (2003)  Using spatial language.  In B. H. Ross (ed.) The psychology of
learning and motivation 127–161. (Vol. 43) New York: Academic Press.
Carlson, L. A. and Covey, E. S. (2005) How far is near? Inferring distance from spatial
descriptions. Language and Cognitive Processes 20: 617–632.
Carlson, L. A. and Hill, P. L. (2008) Processing the presence, placement and prop-
erties of a distractor during spatial language tasks. Memory & Cognition 36:
240–255.
Carlson, L. A. and Kenny, R. (2006) Interpreting spatial terms involves simulating
interactions. Psychonomic Bulletin & Review 13: 682–688.
Carlson, L. A. and Logan, G. D. (2001) Using spatial relations to select an object.
Memory & Cognition 29: 883–892.
Carlson, L. A., Regier, T, Lopez, W. and Corrigan, B. (2006) Attention unites form
and function in spatial language. Spatial Cognition and Computation 6: 295–308.
Carlson, L. A. and van Deman, S. R. (2004) The space in spatial language. Journal of
Memory and Language 51: 418–436.
Carlson-Radvansky, L. A., Covey, E. S. and Lattanzi, K. L. (1999) ‘What’ effects on
‘where’: Functional influences on spatial relations. Psychological Science 10:
516–521.
Carlson-Radvansky, L. A. and Irwin, D. E. (1993) Frames of reference in vision and
language: Where is above? Cognition 46: 223–244.
Carlson-Radvansky, L. A. and Logan, G. D. (1997) The influence of reference frame
selection on spatial template construction. Journal of Memory and Language 37:
411–437.
Carlson-Radvansky, L. A. and Radvansky, G. A. (1996) The influence of functional
relations on spatial term selection. Psychological Science 7: 56–60.
Clark, H. H. (1973) Space, time, semantics, and the child. In T. E. Moore (ed.)
Cognitive development and the acquisition of language. New York: Academic
Press.
Coventry, K. R. and Garrod, S. C. (2004) Saying, seeing and acting: The psychological
semantics of spatial prepositions. New York: Psychology Press.
Fillmore, C. J. (1971) Santa Cruz lectures on deixis. Bloomington, IN: Indiana
University Linguistics Club.
Franklin, N., Henkel, L. A. and Zengas, T. (1995) Parsing surrounding space into
regions. Memory & Cognition 23: 397–407.
Garnham, A. (1989) A unified theory of the meaning of some spatial relational
terms. Cognition 31: 45–60.
 PARSING SPACE AROUND OBJECTS 137

Herskovits, A. (1986) Language and spatial cognition: An interdisciplinary study of the

prepositions in English. Cambridge: Cambridge University Press.
Hayward, W. G. and Tarr, M. J. (1995) Spatial language and spatial representation.
Cognition 55: 39–84.
Langacker, R. W. (1987) Foundations of cognitive grammar. Stanford: Stanford
University Press.
Langacker, R. W. (1993) Grammatical traces of some ‘invisible’ semantic constructs.
Language Sciences 15: 323–355.
Langacker, R. W. (2002) A study in unified diversity: English and mixtec locatives.
In N. J. Enfield (ed.) Ethnosyntax: Explorations in grammar and culture. Oxford:
Oxford University Press.
Levelt, W. J. M. (1984) Some perceptual limitations on talking about space. In A. J.
van Doorn, W. A. van der Grind and J. J. Koenderink (eds) Limits in perception
323–358. Utrecht: VNU Science Press.
Levelt, W. J. M. (1996) Perspective taking and ellipsis in spatial descriptions. In
P. Bloom, M. A. Peterson, L. Nadel and M. Garrett (eds) Language and space
77–108. Cambridge, MA: MIT Press.
Levinson, S. (1996) Frames of reference and Molyneux’s questions: Cross-linguistic
evidence. In P. Bloom, M. A. Peterson, L. Nadel and M. Garret (eds) Language
and space 109–169. Cambridge, MA: MIT Press.
Logan, G. D. and Sadler, D. D. (1996) A computational analysis of the apprehension
of spatial relations. In P. Bloom, M. A. Peterson, L. Nadel and M. Garret (eds)
Language and space 493–529. Cambridge, MA: MIT Press.
Miller, G. A. and Johnson-Laird, P. N. (1976) Language and perception. Cambridge,
MA: MIT Press.
Morrow, D. G. and Clark, H. H. (1988) Interpreting words in spatial descriptions.
Language and Cognitive Processes 3: 275–291.
Regier, T. and Carlson, L. A. (2001) Grounding spatial language in perception: An
empirical and computational investigation. Journal of Experimental Psychology:
General 130: 273–298.
Talmy, L. (1983) How language structures space. In H. L. Pick and L. P. Acredolo
(eds) Spatial orientation: Theory, research and application 225–282. New York:
Plenum Press.
Tyler, A. and Evans, V. (2003) The semantics of English prepositions: Spatial scenes,
embodied meaning and cognition. Cambridge: Cambridge University Press.
Vandeloise, C. (1991) Spatial prepositions: A case study from French. Chicago:
Chicago University Press.
Zwaan, R.A. (2004) The immersed experiencer: Toward an embodied theory of
language comprehension. In B. H. Ross (ed.) The psychology of learning and
motivation 35–62. (Vol. 44) New York: Academic Press.
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 139

6 A neuroscientific perspective on the linguistic

encoding of categorical spatial relations 1
David Kemmerer

Slus, a Mayan speaker of the language Tzeltal, says to her husband, facing an
unfamiliar contraption, ‘Is the hot water in the uphill tap?’ It is night, and we have
just arrived at an alien hotel in a distant, unfamiliar city out of the hills. What does
she mean? She means, it turns out, ‘Is the hot water in the tap that would lie in the
uphill (southerly) direction if I were at home?’ Levinson (2003, p. 4)

1 Introduction

The semantic domain of space arguably consists of three subdomains – shape, motion,
and location. Most readers of the current volume are probably aware that all three of
these subdomains have been intensively investigated by cognitive linguists during the
past few decades. However, many readers may not realize that in recent years cognitive
neuroscientists have begun to use the tools of their trade – especially the lesion method
and hemodynamic methods – to illuminate the brain structures that underlie each
subdomain. The greatest progress has been made in understanding the neural correlates
of the subdomain of shape, but a substantial amount has also been learned about the
anatomical bases of the subdomains of motion and location (for reviews see Kemmerer,
2006, in press, forthcoming). This chapter focuses on the subdomain of location and
attempts to integrate new findings from linguistics and neuroscience.
At the very outset, it is important to note that much of the neuroscientific work
on the meanings of locative morphemes has been partly motivated by an interest in
Kosslyn’s (1987) hypothesis that the human brain contains separate systems for comput-
ing two types of spatial relations – coordinate and categorical (for reviews see Jager
and Postma, 2003; Laeng et al., 2003; Postma and Laeng, 2006). Representations of
coordinate spatial relations involve precise metric specifications of distance, orienta-
tion, and size; they are useful for the efficient visuomotor control of object-directed
actions such as grasping a cup; and they may be processed predominantly in the right
hemisphere. In contrast, representations of categorical spatial relations involve groupings
of locations that are treated as equivalence classes; they serve a variety of perceptual
functions, such as registering the rough positions of objects in both egocentric and
allocentric frames of reference; and they may be processed predominantly in the left
hemisphere. It has often been observed that categorical spatial relations are usually
referred to linguistically by words like English prepositions, many of which specify
binary oppositions – e.g., on/off, in/out, left/right, above/below. For instance, Laeng et
al. (2003, p. 308) state that ‘all natural languages seem to have a special class in their
140 LANGUAGE, COGNITION AND SPACE

grammar (i.e., prepositions) devoted to the expression of categorical spatial relations’.

As I demonstrate below, however, prepositions are not the only relevant grammatical
category, and the range of categorical spatial relations that are linguistically encoded
goes well beyond the meanings of English prepositions.
The chapter is organized as follows. In section 2 I summarize recent research
on the kinds of categorical spatial relations that are encoded in the 6000+ languages
of the world and that are also, ipso facto, implemented in the brains of the speakers.
Emphasis is placed on crosslinguistic similarities and differences involving deictic
relations, topological relations, and projective relations, the last of which are organized
around three distinct frames of reference – intrinsic, relative, and absolute. During the
past few decades, a voluminous literature on the meanings of locative morphemes has
emerged, including several new approaches such as the Functional Geometry framework
(e.g., Coventry and Garrod, 2004; Carlson and Van Der Zee, 2005) and the Principled
Polysemy model (e.g., Tyler and Evans, 2003). However, I will draw mostly on recent
typological research, especially studies conducted by the Language and Cognition Group
at the Max Planck Institute for Psycholinguistics (e.g., Levinson, 2003; Levinson and
Wilkins, 2006). Next, section 3 reviews what is currently known about the neuroana-
tomical correlates of linguistically encoded categorical spatial relations, with special
focus on the left supramarginal and angular gyri. In addition, suggestions are offered for
how crosslinguistic data can help guide future research in this area of inquiry. Finally,
section 4 explores the interface between language and other mental systems, specifi-
cally by summarizing studies which suggest that although linguistic and perceptual/
cognitive representations of space are at least partially distinct, language nevertheless
has the power to bring about not only modifications of perceptual sensitivities but also
adjustments of cognitive styles.

2 What types of categorical spatial relations are linguistically

encoded?

Very few languages have a word for ‘space’ in the abstract sense employed by philoso-
phers and scientists such as Newton, Leibniz, Kant, and Einstein. However, current
evidence suggests that all languages have Where-questions (Ulltan, 1978) that tend to
elicit answers in which the figure object (F) – i.e., the thing to be located – is described
as being within a search domain defined by some kind of categorical spatial relation
to a ground object (G) – i.e., a thing that serves as a point of reference (Talmy, 1983).
Several classes of categorical spatial relations are encoded to different degrees in different
languages, and although they interact in complex ways, each one usually constitutes a
fairly independent semantic field that is ‘carved up’ by a specialized set of lexical items
and grammatical constructions (Levinson and Wilkins, 2006).
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 141

2.1 Deictic relations

Deixis involves the many ways in which the interpretation of utterances depends on
aspects of the speech event (Fillmore, 1997). In the present context, the most relevant
deictic expressions are demonstratives – e.g., here vs. there, this vs. that (Burenhult, 2008;
Diessel, 1999, 2005, 2006; Dixon, 2003; Dunn et al., forthcoming). These words specify
the location of F directly in relation to the location of the speech participants, instead of
in relation to some G outside the speech situation. The proper functional characteriza-
tion of demonstratives requires close attention to details of social interaction (Enfield,
2003; Hanks, 2005). However, I will not discuss these complex social parameters here,
since the main focus is on how demonstratives are often used to divide the radial ego-
centric space surrounding the speaker (or addressee) into categorically discrete zones.
Crucially, demonstratives do not encode metrically precise degrees of remoteness from
the deictic center, but rather have abstract meanings that are pragmatically modulated
by either the discourse context or the referential scenario, thereby allowing speakers
to flexibly expand or contract the zones so as to express an unlimited range of distance
contrasts – e.g., here in this room vs. here in this galaxy.
In a sample of 234 languages from diverse families and geographical regions, Diessel
(2005) found that the kind of demonstrative system manifested in English, with a binary
proximal/distal contrast, is actually the most frequent, showing up in 127 (54%) of the
languages. However, this is the minimal type of system, and other languages exhibit
systems of greater complexity. For example, some languages include the addressee as
a possible deictic center. Such person-oriented systems come in several varieties. One
type, exemplified by Pangasinan (Western Austronesian, Philippines), 2 has a three-
way contrast between ‘near speaker’, ‘near addressee’, and ‘far from both speaker and
addressee’, while another type, exemplified by Quileute (Chimakuan, Washington State),
has a four-way contrast between ‘near speaker’, ‘near addressee’, ‘near both speaker and
addressee’, and ‘far from both speaker and addressee’. These person-oriented systems
resemble the English two-term system insofar as they specify just two zones – proximal
and distal. The key difference is that person-oriented systems require the speaker to
perform more elaborate spatial calculations which take into account not only his or
her own egocentric frame of reference, but also that of the addressee. Perhaps for this
reason, person-oriented systems are relatively rare. A more common way to increase
the complexity of a demonstrative system is to partition the dimension of distance into
more fine-grained zones. Eighty-eight (38%) of the languages in Diessel’s sample follow
this strategy by distinguishing between three zones – proximal, medial, and distal.
Spanish and Yimas (Sepik-Ramu, Papua New Guinea) have systems like this. A very
small proportion of languages (less than 4% in Diessel’s sample) go one step further by
distinguishing between four zones – proximal, medial, distal, and very distal. Tlingit
(Na Dane, Yukon) is the most often cited example. There are even reports of languages
with demonstrative systems that encode five distance contrasts (Anderson and Keenan,
1985), but Diessel supports Fillmore (1997), who maintains that systems with more than
four terms invariably combine other semantic parameters.
142 LANGUAGE, COGNITION AND SPACE

These other semantic parameters include visibility, elevation, and geography. A

striking example of how local geographic features can be incorporated into the semantics
of demonstrative systems comes from the Himalayan language Limbu (Kiranti, Nepal),
which has the following terms: ma:dha:mbi means ‘on the slope of the mountain ridge
across the valley from where the speaker is situated’, kona:dha:mbi means ‘on the same
slope of the mountain ridge as the speaker’, and khatna:dha:mbi means either ‘on the
back side of the mountain ridge on which the speaker is situated’ or ‘on the far side of
the mountain ridge across the valley from which the speaker is situated’ (van Driem,
2001). Even more remarkable is Cora (Uto-Aztecan, Mexico), which encodes in mul-
timorphemic words the distance of F relative to the speaker (proximal vs. medial vs.
distal), the location of F relative to the speaker’s line of sight (inside vs. outside), and
the location of F relative to a mountain slope (foot vs. face vs. top) – e.g., mah means
roughly ‘away up there to the side in the face of the slope’ (Casad and Langacker, 1985).

2.2 Topological relations

According to the loose, non-mathematical sense of ‘topology’ employed in research on

spatial semantics, topological relations involve various types of allocentric contiguity
between F and G, such as the notions of penetration and containment encoded by the
English prepositions through and in, respectively. In an influential article building on a rich
tradition of previous work, Landau and Jackendoff (1993) point out that the spatial con-
cepts found in English prepositions are extremely coarse – in other words, very abstract,
schematic, and categorical – since they place few geometric constraints on F and G. They
also argue that these sorts of concepts are likely to be crosslinguistically universal. For
example, based on the observation that English prepositions are insensitive to the specific
shapes of F and G, they state that no language should have a locative element like the
hypothetical sprough, which means ‘reaching from end to end of a cigar-shaped object’,
as in The rug extended sprough the airplane. Similarly, given that English prepositions do
not discriminate between the subregions of Gs that are containers, they propose that no
language will manifest a locative element like the hypothetical plin, which means ‘contact
with the inner surface of a container’, as in Bill sprayed paint plin the tank.
This orthodox view has been challenged by studies that have revealed considerable
diversity in the kinds of topological relations that are lexicalized in various languages.
To begin with the blackest fly in the ointment, Levinson (2003: 63, 72) notes that the
putative non-existence of an expression like sprough is directly contradicted by Karuk
(Hokan, Northwestern California), which has a suffix -vara meaning ‘in through a
tubular space’. Similarly, expressions of the plin type, which specify subregions of G, have
been attested in Makah (Wakashan, Washington State), which has suffixes encoding
locations such as ‘at the rear of a house’, ‘at the base of an upright object’, and ‘at the head
of a canoe’ (Davidson, 1999). Equally if not more threatening to Landau and Jackendoff ’s
theory is Tzeltal (Mayan, Southeastern Mexico), which describes topological relations
with a large but, importantly, closed class of so-called dispositional adjectives that specify
quite detailed, yet still essentially categorical, distinctions involving the location of F
relative to G (Brown, 1994). When combined with the single, all-purpose relational
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 143

marker ta, these words extensively cross-classify spatial arrays that would be described
in English by using semantically more general prepositions like in and on (Table 1).
Thus, if asked ‘Where are the tortillas?’ an English speaker might reply simply ‘On the
table’, a statement that semantically reduces the tortillas to a mere point or shapeless
blob; however, a Tzeltal speaker would probably select one of several terms that encode
geometric information about the appearance of the tortillas, such as latzal (if they are
stacked) or pakal (if they are folded).
Table 1. Examples of Tzeltal dispositional adjectives encoding topological relations that would normally be
described in English as in or on. In each case, ta is a general-purpose marker meaning ‘be located’. (Data reproduced
from Brown, 1994.)

A. Ways of conveying ‘in’ relationships involving containment.

Form Meaning Eliciting F and G

t’umul ta be located, by having been immersed in liquid in a container apple, water in bucket
tik’il ta be located, by having been inserted into a container with a narrow bull, corral
opening
xijil ta be located, of long-thin object, by having been inserted carefully into pencils, cup
a container
xojol ta be located, by having been inserted singly into a close-fitting container coffee bag, pot
tz’apal ta be located, by having been inserted at its end into supporting medium stick, ground
lapal ta be located, of long-thin-sharp object, by having been inserted through safety pin, cloth
a flexible object

B. Ways of conveying ‘on’ relationships involving contact with, and support by, a horizontal surface.

Form Meaning Eliciting F and G

pachal ta be located, of a wide-mouthed container canonically ‘sitting’ bowl, table
waxal ta be located, of a tall oblong-shaped container or solid object canonically bottle, table
‘standing’
pakal ta be located, of a blob with a distinguishably flat surface lying ‘face’ down dough, table
lechel ta be located, of a wide flat object lying flat frying pan, table
chepel ta be located, of a full (bulging) bag supported underneath netbag, table
cholol ta be located, of multiple objects arranged in a row beans, table

Although languages differ greatly in the kinds of topological relations they encode, there
are underlying patterns. In a recent study, nine unrelated languages3 were investigated
by comparing native speaker responses to a standardized set of 71 pictures showing a
wide range of topological relations (Levinson and Meira, 2003). Results indicated that
crosslinguistically the labels for pictures were not randomly distributed but instead
tended to cluster, suggesting that the topological domain forms a coherent similarity
space with a number of strong ‘attractors’, i.e., taxonomically basic-level categories that are
144 LANGUAGE, COGNITION AND SPACE

statistically likely to be recognized by languages – in particular, notions such as contain-

ment, attachment, superadjacency, subadjacency, and proximity. Several generalizations
about the organization of this abstract similarity space emerged from the study. First,
each core concept has a prototype structure. For example, at the center of the cluster of
containment pictures were scenes in which F is enclosed within G (e.g., a dog in a cage);
scenes involving partial two-dimensional containment on a planar surface (e.g., a dog
in a yard) were more peripheral, implying that English is somewhat unusual in using in
for such topological relations. Second, the core concepts are arranged as neighbors along
gradients in the similarity space, making some conflations of categories more natural than
others. For instance, English on embraces both superadjacency (e.g., a cup on a table)
and attachment (e.g., a picture on a wall), Berber di embraces both attachment (e.g., a
picture on a wall) and containment (e.g., an apple in a bowl), and Spanish en embraces
all three categories; however, there should not be, and do not as yet appear to be, any
languages with a spatial morpheme that applies to superadjacency and containment while
excluding attachment, since the latter concept is intermediate between the other two
along the relevant gradient of the abstract similarity space. Third, each core concept can
be further fractionated, leading to more fine-grained categories of topological relations.
For example, the cluster of pictures for superadjacency included scenes both with and
without contact (e.g., a cup on a table, and a lamp above a table), suggesting that languages
are likely to use the same morpheme for these kinds of relations – a tendency that seems
somewhat surprising from the perspective of English, since on and above/over divide the
superadjacency category into separate subcategories distinguished by the presence or
absence of contact between F and G. Levinson and Meira also report many intriguing
cases of category fractionation in other languages, such as the exotic Tiriyó morpheme
awee, glossed ‘astraddle’, which applies to the subset of attachment pictures in which F
is suspended from a point on G and hangs down on either side of it (e.g., a coat on a
hook, an earring dangling from a person’s ear, a pendant on a chain, clothes drying on a
line, a balloon on a stick, and a tablecloth on a table). Further analyses of crosslinguistic
similiarities and differences in the subdomain of topological relations can be found in
the detailed case studies compiled by Levinson and Wilkins (2006).

2.3 Projective relations

Projective relations involve locating F within a search domain that radiates out some
distance from G along a specified angle or line. This class of categorical spatial rela-
tions breaks down into several subclasses, each of which exhibits substantial, but not
unconstrained, crosslinguistic variation. The following summary is based mainly on
Levinson’s (2003) analysis. According to Levinson (2003, p. 76; see also Levinson and
Wilkins, 2006), languages use, to varying degrees, three frames of reference for encoding
(primarily) horizontal projective relations: ‘the intrinsic system, which projects out a
search domain from a named facet of a landmark object; the relative system, which
imports the observer’s bodily axes and maps them onto the ground object thus deriving
named angles; and the absolute system, which uses a fixed set of bearings or a conceptual
‘slope’ to define a direction from a ground object’.
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 145

2.3.1 The intrinsic frame of reference

The first locative strategy has two steps: the speaker identifies a salient part or facet of
G – e.g., the ‘front’ – and then extracts from the designated component an angle which
extends outward a certain distance, thereby defining a search domain within which
F can be found – e.g., The ball is in front of the house. In English this system operates
mainly by imposing on G a six-sided, box-like ‘armature’ that yields a front, back, top,
bottom, and two lateral (i.e., left and right) sides as the major intrinsic parts. Functional
criteria are often used to identify, for instance, the ‘front’ of G based on factors like the
typical direction of the perceptual apparatus (for animate entities), the typical direction
of motion (for vehicles), or the typical direction of encounter (for houses, TVs, etc.).
Some objects resist this decompositional approach because they appear to lack intrinsic
asymmetries – e.g., English speakers do not construe trees and mountains as having
fronts and backs. But judgments of this nature vary across languages – e.g., in Chamus
(Nilo-Saharan, Kenya) the front of a tree is the side it leans toward, or, if it is vertical,
the side with the biggest branch or the most branches, and in Kikuyu (Nilo-Saharan,
Kenya) the front of a mountain is the side opposite its steepest side (Heine, 1997, p. 13).
It is crosslinguistically common for locative terms employing the intrinsic frame
of reference to derive historically from body-part terms (Svorou, 1994; Heine, 1997; for
recent crosslinguistic work on body part terms, see Majid, Enfield, and van Staden, 2006,
as well as the critique by Wierzbicka, 2007). This can be seen in the English example used
above – The ball is in front of the house – and in a number of fixed English expressions like
the face of a cliff, the mouth of a cave, the eye of a hurricane, the nose of an airplane, the
head of a nail, the neck of a guitar, the arm/leg of a chair, etc. In many languages, however,
the body-part-based intrinsic system is quite complex, requiring regular linguistically
driven visual analysis of the axial geometry as well as the major and minor protrusions
of inanimate objects so that the relative appropriateness of different body-part terms
can be computed instantly on the basis of these inherent properties, i.e., independent
of the object’s orientation or the speaker’s viewpoint. Perhaps the best-studied language
of this type is Tzeltal (Levinson, 1994), in which even a G as seemingly nondescript as
a stone may be assigned a ‘face’, a ‘nose’, an ‘ear’, a ‘back’, a ‘belly’, or any of about fifteen
other quasi-metaphorical body parts in order to specify that F is located within a search
domain projected from one of these facets – e.g., an s-jol ‘head’ is a protrusion that can
be found at one end of the major axis of G and that has a gently curved, circular outline
with only minor concavities on either side.4

2.3.2 The relative frame of reference

To describe spatial arrays in which F is at some remove from G but G is classified as

‘unfeatured’ by the intrinsic system of the given language, the front/back and left/right
axes of the observer’s body can be introduced to provide a frame of reference for struc-
turing the scenario. This increases the complexity of the spatial relations from binary
(F and G) to ternary (F, G, and the observer). Thus, whereas The ball is in front of the
house specifies a binary relation in which F is located with respect to an intrinsic facet
of G, The ball is in front of the pole specifies a ternary relation in which F is located with
146 LANGUAGE, COGNITION AND SPACE

respect to a non-intrinsic facet of G that can only be identified by taking into account
the observer’s perspective.
The type of relative system found in English involves imposing on G the mirror
reflection of the observer’s bodily axes (Figure 1A). A mirror flips the front/back axis
but not the left/right axis of the object it reflects. To designate F as being in front of or
in back of G, the observer’s front/back axis is mapped onto G under 180º rotation, so
that The ball is in front of the pole means ‘From this viewpoint, the ball is in a search
domain projected from the side of the pole that ‘faces’ me’. To designate F as being left
or right of G, directions are projected laterally from G along angles that correspond
to the observer’s left/right axis. Besides the English system, there are two other logical
possibilities for organizing the relative frame of reference on the horizontal plane, and
both are utilized by other languages (Levinson, 2003: 84–89). One strategy, exemplified
by some dialects of Tamil (Dravidian, India), involves mapping the observer’s bodily
axes onto G under complete 180º rotation, generating not only front/back reversal but
also left/right reversal, so that The ball is in front of the pole has the same meaning as it
does in English, but The ball is to the left of the pole means that the ball is located in the
region that English speakers would consider ‘to the right’ (Figure 1B). The other strategy,
exemplified by Hausa (Chadic, Nigeria), involves mapping the observer’s bodily axes
onto G without any rotation whatsoever, so that The ball is in front of the pole means that
the ball is located in the region that English speakers would consider ‘in back of ’, but
The ball is to the left of the pole means the same thing as it does in English (Figure 1C).

A. Reflection (e.g., English) B. Rotation (e.g., Tamil) C. Translation (e.g., Hausa)

B B F

L Ground R R Ground L L Ground R

F F B
ž ž ž
F F F

L Observer R L Observer R L Observer R

B B B

Figure 1. Crosslinguistic variation in the organization of the relative frame of reference on the hori-
zontal plane. In the illustration of each type of system, the observer is shown at the bottom and the
ground object at the top, with the observer’s line of sight indicated by an arrow. Abbreviations: F =
front, B = back, L = left, R = right.
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 147

2.3.3 The absolute frame of reference

The third type of angular specification on the horizontal plane involves an absolute
frame of reference that provides a set of fixed bearings or cardinal directions, similar
to north, south, east, and west. These bearings define ‘an infinite sequence of parallel
lines – a conceptual ‘slope’ – across the environment’ (Levinson, 2003, p. 90). To indicate
the location of F with respect to G, one projects an angle from G to F, assesses the
orientation of this angle in relation to the grid of cardinal directions, and selects the
appropriate term – e.g., something like The ball is north of the pole. Absolute systems
are fundamentally geocentric, and languages often base terms for cardinal directions on
stable environmental features like mountain slopes, river drainages, and prevailing wind
patterns. For example, returning yet again to Tzeltal, it has an absolute system that is
anchored in the mountain incline of the local landscape, giving rise to three directional
terms: ajk’ol ‘uphill’ (roughly south), alan ‘downhill’ (roughly north), and jejch ‘across’
(either east or west) (Brown and Levinson, 1993, forthcoming). It is important to note
(since this issue has been previously misunderstood – see the debate between Li and
Gleitman, 2002, and Levinson et al., 2002) that although the terminology of absolute
systems derives from environmental landmarks, such systems are fully abstracted, and
in order to use them spontaneously and accurately, speakers must constantly monitor
their spatial orientation by running a kind of mental compass. This is a remarkable
neurocognitive capacity, as revealed by the anecdote about the Tzeltal speaker, Slus, in
the epigraph of this chapter. Another vital point is that unlike the English north/south/
east/west system, which has extremely limited use, the absolute systems under discussion
are regularly employed to describe spatial arrays at every level of scale, from inches to
miles. This is clearly shown in Levinson’s (2003, p. 114) description of Guugu Yimithirr
(Pama-Nyungan, Australia), which uses exclusively the absolute frame of reference for
characterizing horizontal projective relations:

In GY, in order to describe someone as standing in front of the tree, one says
something equivalent (as approximate) to ‘George is just north of the tree’, or, to
tell someone to take the next left turn, ‘go north’, or, to ask someone to move over
a bit, ‘move a bit east’, or, to instruct a carpenter to make a door jamb vertical,
‘move it a little north’, or, to tell someone where you left your tobacco, ‘I left it on
the southern edge of the western table in your house’, or, to ask someone to turn off
the camping gas stove, ‘turn the knob west’, and so on. So thoroughgoing is the use
of cardinal directions in GY that just as we think of a picture as containing virtual
space, so that we describe an elephant as behind a tree in a children’s book (based
on apparent occlusion), so GY speakers think about it as an oriented virtual space:
if I am looking at the book facing north, then the elephant is north of the tree, and
if I want you to skip ahead in the book I will ask you to go further east (because
the pages would then be flipped from east to west).
148 LANGUAGE, COGNITION AND SPACE

2.3.4 The vertical dimension

Finally, with regard to the linguistic encoding of projective relations along the vertical
dimension, the three frames of reference – intrinsic, relative, and absolute – usually coin-
cide and yield the same answer to the question ‘Where is F in relation to G?’ (Levinson,
2003, p. 75). For example, consider a scene in which a fly hovers above a bottle. F is
‘above’ G according to all three criteria: it is located within the search domain that radiates
from the top of the bottle (intrinsic frame); it is higher than the bottle in the observer’s
visual field (relative frame); and it is higher than the bottle along the vertical axis defined
by gravity (absolute frame). However, as a number of experiments have shown (e.g.,
Friederici and Levelt, 1990; Carlson-Radvansky and Irwin, 1993; Carlson, 1999), the three
frames of reference can be manipulated independently of each other (e.g., by rotating
either G or the observer, or, more radically, by shifting the entire array to a zero gravity
environment) to create special situations in which they yield conflicting answers to the
Where-question. Also, as noted earlier, although English clearly distinguishes above/over
from on according to whether F contacts G, this may be the result of splitting into two
subcategories the crosslinguistically more common (and perhaps conceptually more
basic) category of superadjacency, which is neutral with respect to contact and is directly
encoded in languages like Japanese and Arrernte (Pama-Nyungan, Australia). This is
one of several ways in which the vertical dimension interacts with topology. Another
manifestation of this interaction is that over and under are not synonymous with above
and below, respectively, because the former prepositions have a topological component
that makes them more suitable than the latter for describing spatial arrays that involve
an encompassment relation – e.g., it is more felicitous to say that a penny is under than
below an inverted cup on a table (Coventry et al., 2001).

2.4 Summary

Two major generalizations emerge from this review of the kinds of categorical spatial
relations that are encoded in languages. First, there is a huge amount of crosslinguistic
variation regarding the specific concepts that are lexicalized, suggesting that every
language has its own unique spatial ontology with idiosyncratic notions ranging from
Limbu’s ma:dha:mbi (‘F is on the slope of the mountain ridge across the valley from
where the speaker is situated’) to Tiriyó’s awee (‘F is astraddle G’) to Tzeltal’s ajk’ol (‘F
is uphillwards, i.e., roughly south, of G’). Second, despite this tremendous diversity, a
number of patterns can be identified that lend coherence to each of the semantic fields
comprising the overall conceptual domain. For instance, in the field of deictic relations,
over 50% of languages appear to have demonstrative systems that specify a binary
proximal/distal contrast; in the field of topological relations, a relatively small number
of core concepts tend to recur across languages and hence constitute statistical attrac-
tors for lexicalization; and in the field of projective relations, languages typically have
complex sets of expressions that instantiate up to three frames of reference – intrinsic,
relative, and absolute.
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 149

3 What are the neuroanatomical correlates of linguistically

encoded categorical spatial relations?

Very little research in cognitive neuroscience has explored which brain structures sub-
serve the rich variety of categorical spatial relations that are lexicalized in languages
around the world. Nevertheless, all of the studies that have addressed this issue suggest
that the left inferior parietal lobule (IPL) is an especially important cortical region. It is
well-established that the visual system consists of two major subsystems – the so-called
‘what’ pathway that projects from the occipital lobe to the ventral temporal lobe and
processes complex information about shape, color, and texture that is necessary for
conscious object perception and recognition; and the so-called ‘where’ pathway that
projects from the occipital lobe to the parietal lobe and processes complex information
about space that is necessary for efficient sensorimotor interaction with objects (for a
review see Milner and Goodale, 2006). In an influential article, Landau and Jackendoff
(1993) used this distinction as the basis for speculating that the meanings of English
locative prepositions are represented in the left IPL. The studies summarized below not
only corroborate this proposal but allow it to be made more precise by suggesting that
the critical neuroanatomical structures are the supramarginal gyrus and, perhaps to a
lesser extent, the angular gyrus.

3.1 Studies implicating the left inferior parietal lobe

3.1.1 Supramarginal gyrus

Damasio et al. (2001) report a positron emission tomography (PET) study in which
English speakers viewed drawings of static spatial relations between objects (e.g., a
cup on a table) and performed two tasks: naming F, and naming the spatial relation
between F and G with an appropriate preposition. When the condition of naming
objects was subtracted from that of naming spatial relations, the largest and strong-
est area of activation was in the left supramarginal gyrus (SMG). The authors do not
indicate which prepositions were targeted for production, but it appears that a mixture
of topological and projective prepositions were included, which suggests that the SMG
activation reflects semantic processing of both types. More recently, a functional mag-
netic resonance imaging (fMRI) study also found significant SMG activation during a
task requiring semantic processing (within the relative frame of reference) of the Dutch
equivalents of the terms left and right (Noordzij et al., 2008).
Additional evidence comes from a neuropsychological study conducted by Tranel
and Kemmerer (2004; see also Kemmerer and Tranel, 2000, 2003, and Kemmerer, 2005).
They administered a set of tests that require production, comprehension, and semantic
analysis of 12 English prepositions (encoding topological relations as well as several
kinds of projective relations) to 78 brain-damaged subjects with lesions distributed
throughout the left and right cerebral hemispheres, and then compared the lesion
sites of the subjects who were impaired on the tests with the lesion sites of those who
150 LANGUAGE, COGNITION AND SPACE

were unimpaired. Poor performance was linked specifically with damage in the left
SMG and the left frontal operculum. The involvement of the left SMG strengthens the
hypothesis that this region plays an essential role in representing the spatial meanings
of English prepositions. The investigators did not, however, conduct separate analyses
to determine whether the different semantic classes of prepositions dissociate from each
other behaviorally and neuroanatomically. As for the involvement of the left frontal
operculum, it may reflect either or both of two functions: phonological encoding, pos-
sibly in Brodmann area 44 (e.g., Amunts et al., 2004), and semantic working memory,
possibly in Brodmann areas 45 and/or 47 (e.g., Devlin et al., 2003; Thompson-Schill et
al., 1999; Wagner et al., 2001).
To my knowledge, no other studies of spoken languages have identified a strong
association between the left SMG and morphemes that denote categorical spatial
relations;5 however, further evidence for precisely this association comes from two
functional neuroimaging studies of locative classifier constructions in American Sign
Language (ASL; Emmorey et al., 2002) and British Sign Language (BSL; MacSweeney et
al., 2002). Locative classifier constructions are complex coding devices that exploit the
three-dimensional medium of signing space in the following ways: the relative positions
of the hands in front of the body correspond schematically and iconically to the relative
positions of F and G in the physical world, and the shape of each hand indicates the
general class to which each object belongs (Emmorey, 2003). For example, to express
the equivalent of The bike is near the house, the referential handshapes for house and
bike are articulated sequentially (G preceding F), and then the classifier for vehicles
(thumb, middle, and index fingers extended) is placed directly adjacent to the classifer
for large bulky objects (five fingers spread and curved) to indicate topographically that F
is ‘near’ G. To investigate the neural substrates of this unique form of spatial description,
Emmorey et al. (2002) conducted a PET study in which deaf native ASL signers viewed
the same kinds of drawings of spatial relations that were used in Damasio et al.’s (2001)
PET study, and performed two tasks: naming F, and naming the spatial relation between
F and G with an appropriate locative classifier construction. Relative to naming objects,
naming spatial relations engaged the left SMG; moreover, the centroid of activation was
similar to that found for English speakers in Damasio et al.’s (2001) study, suggesting that
it reflects semantic processing. In another study, MacSweeney et al. (2002) used fMRI to
investigate the neural systems underlying comprehension of BSL sentences containing
locative classifier constructions. Compared to sentences without such constructions,
activation was observed in the same sector of the left SMG as in Emmorey et al.’s (2002)
study, providing additional support for the hypothesis that this cortical area contributes
to the semantic processing of linguistically encoded categorical spatial relations.

3.1.2 Angular gyrus

Neuroimaging and neuropsychological studies suggest that the left angular gyrus (AG)
is also involved in the linguistic representation of categorical spatial relations, but
perhaps to a more limited degree than the left SMG. Baciu et al. (1999) report an fMRI
study in which significantly stronger left than right AG activation was observed while
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 151

subjects judged whether a dot was presented above or below a bar. This task has a core
linguistic component because, as the instructions clearly indicate, the two categories
that must be discriminated are directly encoded by the projective prepositions above
and below. This particular spatial contrast may seem natural and intuitive to English
speakers, but it is by no means crosslinguistically universal, since some languages do
not have morphemes that distinguish ‘above’ from ‘on’ or ‘below’ from ‘in’ (Levinson,
2003, p. 73; Levinson and Meira, 2003, p. 507). Hence the left AG activation may reflect,
in part, the essentially lexicosemantic process of classifying the location of the dot as
falling within one of two projected search domains – ’above’ or ‘below’ the line – that
are both familiar categories in the spatial ontology of the subject’s native language.6
Another linguistically encoded categorical spatial contrast that has been linked,
albeit loosely, with the left AG is the distinction between left and right within the intrinsic
frame of reference. Lesions centered in the left AG sometimes produce Gerstmann
syndrome, which comprises the following four symptoms: left/right confusion, finger
agnosia, agraphia, and acalculia (Gerstmann, 1957; Mayer et al., 1999; Mazzoni et al.,
1990; Morris et al., 1984; Roeltgen et al., 1983; Varney, 1984). The symptom of left/right
confusion is usually manifested as difficulty pointing to left and right body parts on
command. However, the relevance of this particular deficit to the issue of the neural
correlates of the meanings of left and right is limited in two ways. First, knowledge of the
actual meanings of the terms is not always disrupted; instead, what seems to be impaired
are certain cognitive operations that are necessary to apply the meanings appropriately
in certain situations, such as the ability to mentally rotate visual images of the body in
space (Bonda et al., 1995; Mayer et al., 1999; Zacks et al., 1999). For example, Mayer
et al.’s (1999) subject performed well (15/16 correct) when asked to point with either
hand to designated left and right parts of his own body, but performed poorly (11/16
correct) when asked to point with a specified hand to designated left and right parts of
a line drawing of a human body that was facing him and hence had a 180º reversal of
left and right sides relative to his own body. Second, studies of Gerstmann syndrome are
generally restricted to the use of left and right to refer to the intrinsic sides of the human
body under various conditions; they do not pursue the inquiry further by systematically
assessing whether the subject understands how the terms are also used – in English but
not in all languages (like Guugu Yimithirr and Tzeltal) – to specify regions of space
that are (a) projected outward from the intrinsic sides of the body (e.g., The ball is on
your left), (b) projected outward from the intrinsic sides of inanimate objects (e.g., The
ball is on the car’s left-hand side), and (c) projected outward from the sides of unfaceted
objects by importing the speaker’s own left/right bodily axis as a frame of reference (e.g.,
The ball is to the left of the pole).

3.2 Further neuroanatomical questions raised by linguistic typology

The studies reviewed above suggest that the left IPL is a key cortical region for represent-
ing the meanings of locative expressions. But when these studies are considered in the
context of the preceding typological survey of the kinds of categorical spatial relations
152 LANGUAGE, COGNITION AND SPACE

that are encoded crosslinguistically, it immediately becomes clear that the research done
so far is merely spadework, and that most of this rich neurocognitive terrain remains
to be mined. For example, at this point it is not known whether the three major classes
of categorical spatial relations – deictic, topological, and projective – are subserved by
separate neural networks within the left IPL. Many questions can also be raised about
the specific neural organization of each class, as suggested below.

3.2.1 Deictic relations

I am not aware of any studies that have explored the neural correlates of demonstratives
that specify egocentrically-anchored deictic spatial relations, although in a previous
paper (Kemmerer, 1999) I pointed out that this topic is interesting in light of the mount-
ing evidence for separate circuits representing, on the one hand, near or peripersonal
space which extends roughly to the perimeter of arm’s reach, and on the other hand,
far or extrapersonal space which extends outward from that fuzzy boundary (for a
review see Berti and Rizzolatti, 2002; see also Longo and Lourenco, 2006, and Makin
et al., 2007). The representational division of near and far sectors of space may derive
from computational differences in the forms of sensorimotor control that are typical
for each sector – i.e., primarily visually guided manual activity in the near sector, and
primarily visual search and object scanning or ‘parsing’ in the far sector. It is tempting to
speculate that this fundamental division is causally relevant to the fact that the majority
of languages worldwide have demonstrative systems that encode a binary proximal/
distal contrast.7 It is also important to bear in mind, however, that demonstratives are
not restricted to quantitative spatial distinctions such as within vs. beyond arm’s reach;
instead, objective distances are semantic variables that are assigned values on-the-fly by
pragmatic factors, thereby allowing speakers to expand or contract the referential range
of demonstratives as needed – e.g., as noted by Levinson (1983, p. 80), the statement
Place it here ‘may have quite different implications of precision if said to a crane opera-
tor or a fellow surgeon’. In addition, some languages have demonstrative systems that
carve the radial space surrounding the speaker into three or, as in the unusual case of
Tlingit, even four concentric zones, thereby violating the two-way perceptual distinction.
Perhaps the abstract meanings of demonstratives are subserved by the left IPL, just like
the other types of linguistically encoded categorical spatial relations described above.
But for demonstrative systems that incorporate geographical information, such as the
Limbu and Cora systems involving mountain slopes, the semantic structures may recruit
not only the dorsal ‘where’ pathway extending into the left IPL, but also the ventral
‘what’ pathway extending into the inferotemporal cortex (Milner and Goodale, 2006).

3.2.2 Topological relations

Further research on the neural correlates of linguistically encoded topological rela-

tions could benefit greatly by utilizing carefully designed stimuli that take into account
theoretically important semantic dimensions, like the standardized set of 71 pictures
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 153

that Levinson and Meira (2003) employed in their crosslinguistic comparison (see also
Levinson and Wilkins, 2006). By conducting high-resolution functional neuroimaging
studies with such materials, it may be possible to test the hypothesis that the conceptual
similarity space discovered by Levinson and Meira (2003) – a similarity space organized
in terms of notions such as containment, attachment, superadjacency, subadjacency,
and proximity – is neuroanatomically implemented in the form of a topographically
structured cortical map in the left IPL, most likely the SMG. Within this map, the
representational dimensions of the conceptual space might be captured, albeit in a
warped manner, by the physical distribution of cortical columns (Kohonen and Hari,
1999; Simmons and Barsalou, 2003; Graziano and Aflalo, 2007; Kriegeskorte et al.,
2008). This is, however, an admittedly bold conjecture.
Another hypothesis is that the inferotemporal cortex contributes to representing
the detailed geometric features of objects that Tzeltal incorporates into the meanings
of dispositional adjectives. Besides encoding various forms of allocentric contiguity
between F and G, such as containment or surface contact and support, many disposi-
tional adjectives also indicate, in a manner much more specific than Indo-European
languages, the shape or configuration of F relative to G (see Table 1). These terms are
semantically somewhat similar to the classifiers that are prevalent in sign languages, and
Emmorey et al. (2002) report that in their PET study the production of locative classifier
constructions engaged not only the SMG but also the left posterior inferotemporal
region – a finding which supports the view that the same region might contribute to
the geometric component of the meanings of Tzeltal dispositional adjectives.

3.2.3 Projective relations

Projective relations may constitute the subdomain of spatial representation with the
greatest potential for interdisciplinary cross-talk between linguistic typology and cogni-
tive neuroscience, because research on the central issue of frames of reference is highly
developed in both areas of inquiry (for the best linguistic overview, see Levinson, 2003;
for excellent neuroscientific overviews, see Hillis, 2006; Previc, 1998; Robertson, 2004).
The direction of influence can certainly go both ways, but here I restrict the discussion
to a small sample of the many ways in which recent findings from linguistic typology
can generate intriguing questions about the neural substrates of linguistically encoded
categorical spatial relations involving intrinsic, relative, and absolute frames of reference.
An important discovery in linguistic typology is that terms for projective relations
involving the intrinsic frame of reference often derive historically from body-part terms.
Moreover, in some languages the application of such terms to the facets of inanimate
objects, for the purpose of anchoring a search domain within which F can be located,
usually requires a complex visuospatial analysis of axial and contour features – e.g., in
Tzeltal an s-ni ‘nose’ is a pointed extremity or an extremity having a sharp convexity,
and an x-chikin ‘ear’ is a flattened protrusion. What is the neural basis of terms like
these? One hypothesis that warrants investigation (Kemmerer and Tranel, 2008) is that
the meanings of such terms depend on shape-sensitive regions of the posterior lateral/
154 LANGUAGE, COGNITION AND SPACE

inferior temporal cortex that receive input from the recently discovered ‘extrastriate
body area’ (EBA), which appears to be especially important for the visual categorization
of human body parts (e.g., Peelen and Downing, 2007).
The fMRI study by Noordzij et al. (2008) implicates the left SMG in the use of left
and right to designate projective relations involving the relative frame of reference.
Would the same type of activation be observed when Tamil speakers perform the
same task? As noted earlier, Tamil employs a strategy of rotation rather than reflection,
so that a sentence like The triangle is to the left of the circle means that the triangle is
located within a search domain that English speakers would consider ‘to the right’
(see Figure 1B).
Perhaps the best example of how linguistic typology can inspire future research on
the neural representation of categorical spatial relations involves the systems of cardi-
nal direction terms analogous to north/south/east/west that speakers of languages like
Tzeltal and Guugu Yimithirr use habitually to specify the angular location of F relative
to G according to an absolute frame of reference. Such linguistic behavior requires
a mental compass that constantly computes one’s orientation within a conventional
framework of fixed bearings. Many nonhuman species have evolutionarily specialized
sensory devices that enable them to use absolute coordinates for navigation – e.g.,
some species of migratory birds have light-absorbing molecules in their retinae that
are sensitive to the magnetic field of the earth and that may enable the birds to see
this information as patterns of color or light intensity (Ritz et al., 2004); sea turtles
have the biological equivalent of a magnetically based global positioning system that
allows them to pinpoint their location relative to geographically large target areas
(Luschi et al., 2007); and locusts perceive polarization patterns in the blue sky and
use them as cues for spatial orientation (Heize and Homberg, 2007). But for people
in ‘absolute’ communities the mental compass that generates their superb sense of
direction – a sense comparable in accuracy to that of homing pigeons (Levinson,
2003, p. 232) – is presumably not genetically programmed but may instead be a
‘knock-on’ effect of the intensive training in orientation tracking that comes with
speaking a language that regularly employs cardinal direction terms to describe
spatial arrays at every level of scale (Levinson, 2003, p. 278; see also Haun et al.,
2006a, 2006b). It is reasonable to suppose that relevant brain areas include parietal
as well as hippocampal and entorhinal structures that have been implicated in both
constructing landmark-based cognitive maps of the environment and monitoring
one’s movement through them (e.g., Ekstrom et al., 2003; Hartley et al., 2003; Janzen
and van Turennout, 2004; Hafting et al., 2005; Leutgeb et al., 2007; Spiers and Maguire,
2006). However, because the use of the mental compass does not require input from
visually perceived landmarks (as illustrated in the epigraph of this paper), other
neural systems must also be recruited, presumably to carry out the computations
that underlie dead-reckoning – that is, keeping track of distances traveled along each
angular heading. Identifying these systems is clearly an exciting direction for future
research (for important new clues, see Sargolini et al., 2006; Heyman, 2006; Jeffrey
and Burgess, 2006).
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 155

3.3 Summary

Research on the neuroanatomical substrates of linguistically encoded categorical spatial

relations has only recently begun, but the studies conducted so far consistently point
to the left IPL as an essential region. Taking into account the view from linguistic
typology, which provides not only a well-developed theoretical framework but also
detailed semantic analyses of the variety of spatial coding systems manifested in the
languages of the world, can lead to many new questions regarding the neural basis of
this rich conceptual domain.

4 How do linguistically encoded categorical spatial relations

interact with perception and cognition?

The final topic of discussion involves the interaction between linguistic and perceptual/
cognitive representations of categorical spatial relations. Very little is currently known
about the nature of this interaction, but the existing data suggest that it is quite compli-
cated (Papafragou, 2007). Two main points are elaborated below. First, several studies
with both normal and brain-damaged populations indicate that the kinds of categorical
spatial distinctions that are encoded for non-linguistic perceptual/cognitive purposes
are to some extent separate from the diverse spatial categorization systems of languages
around the world. Second, a number of other studies suggest that the unique spatial
ontology of one’s language nevertheless has the power to influence one’s perceptual/
cognitive representations of categorical spatial relations by both decreasing sensitivity
to distinctions that are not captured by one’s language and increasing sensitivity to
distinctions that are. The fact that these two sets of findings are not easy to reconcile
is a clear sign that we are still far from understanding the intricasies of the interaction
between linguistic and non-linguistic representations of space.

4.1 Linguistic and perceptual/cognitive representations of categorical spatial

relations are to some extent distinct

Although English distinguishes on from above/over, many other languages – perhaps

even the majority (Levinson and Meira, 2003) – have morphemes that encode the
general notion of superadjacency, which is neutral with respect to whether F contacts
G. Korean is one such language. To investigate whether this form of crosslinguistic
variation influences non-linguistic spatial memory, Munnich et al. (2001) asked native
speakers of English and Korean to perform two tasks with the same stimuli, which
consisted of spatial arrays showing a ball in any of 72 locations superadjacent to a table.
In the naming task, subjects completed the sentence ‘The ball is ___ the table’ (or the
equivalent sentence in Korean). In the memory task, they viewed an array for 500 ms,
and then after a 500 ms delay they saw another array which they judged as being either
the same as or different from the initial one. In the naming task the English speakers
156 LANGUAGE, COGNITION AND SPACE

consistently employed the lexical contrast between on and above/over, whereas the
Korean speakers rarely mentioned the contact/noncontact distinction. In the memory
task, however, the two subject groups had almost identical patterns of accuracy for all
72 locations, including an advantage for locations aligned with the surface of the table.
This study therefore suggests that non-linguistic spatial memory is not constrained by
whether the contact/noncontact distinction is linguistically encoded on a regular basis
throughout one’s life. In other words, even though Korean does not force speakers to
fractionate the category of superadjacency according to the presence or absence of
contact between F and G, this spatial distinction is nevertheless perceptually salient
enough to influence the operation of recognition memory in Korean speakers.
Neuropsychological data also support the view that linguistic and perceptual/
cognitive representations of categorical spatial relations are at least partially separate.
As noted earlier, Tranel and Kemmerer (2004) found maximal lesion overlap in the
left SMG for a group of brain-damaged subjects who had pervasive and severe defects
in the knowledge of the meanings of English prepositions. In a follow-up experiment
with these subjects, non-linguistic visuospatial processing was assessed by admin-
istering a battery of standardized neuropsychological tests, including three subtests
from the Wechsler Adult Intelligence Scale-III (Matrix Reasoning, Block Design, and
Object Assembly), the Benton Facial Recognition Test, the Benton Judgment of Line
Orientation Test, the Hooper Visual Organization Test, the Complex Figure Test (copy),
and the Benton Three-Dimensional Block Construction Test (Benton and Tranel,
1993; Tranel, 1996). Overall, the subjects performed extremely well on the various
tests. Although two of the tests – the Benton Facial Recognition Test and the Benton
Judgment of Line Orientation Test – emphasize sensitivity to coordinate spatial relations,
the remaining tests arguably require an appreciation of categorical spatial relations.8
Moreover, Kemmerer and Tranel (2000) describe a subject with a large right-hemisphere
lesion affecting frontoparietal and temporal regions who manifested a dissociation
that was the opposite of the kind manifested by the brain-damaged subjects in Tranel
and Kemmerer’s (2004) study – namely, intact knowledge of the meanings of English
prepositions but impaired nonlinguistic visuospatial processing of coordinate as well
as categorical spatial relations. Taken together, these findings constitute evidence for
what Jager and Postma (2003, p. 513) call ‘a tripartition between perceptual coordinate
spatial codes, perceptual categorical spatial codes, and verbal categorical spatial codes’.
Additional neuropsychological evidence for this ‘tripartition’ comes from Laeng
(1994), who evaluated the performance of 60 brain-damaged subjects, 30 with unilateral
left-hemisphere (LH) lesions and 30 with unilateral right-hemisphere (RH) lesions, on
the following tasks. First, subjects were shown a drawing of two objects bearing a certain
spatial relation to each other (e.g., a large cat to the left of a small cat), and after a short
delay they were shown another drawing and were asked to make a same/different judg-
ment (analogous to the recognition memory task in Munnich et al.’s 2001 study); half of
the drawings were different, and the change was along either the categorical dimension
(e.g., a large cat to the right of a small cat) or the coordinate dimension (e.g., a large cat
to the left of a small cat, but a different distance away). Second, once again subjects were
shown a drawing of a spatial relation, but this time after a short delay they were shown
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 157

two other drawings and were asked to decide which was more similar to the initial one;
alterations were either categorical or coordinate. On both tasks, LH-damaged subjects
had greater difficulty detecting categorical than coordinate changes, and RH-damaged
subjects exhibited the opposite pattern. Importantly, Laeng (1994) also found that the
LH-damaged subjects’ scores on several aphasia tests – including the Token Test, which
has commands that incorporate prepositions (e.g., ‘Point to the square to the left of
the blue circle’) – did not correlate with their scores on the nonlinguistic spatial tests,
supporting the view that linguistic and perceptual representations of categorical spatial
relations are to some extent distinct. Further research is clearly necessary, however, to
explore the nature of this distinction in greater detail.

4.2 Linguistic representations of space can influence perceptual/cognitive

representations of space

The studies reviewed above suggest that the kinds of categorical spatial distinctions that
are encoded for nonlinguistic purposes are at least partially separate from the spatial
ontology of one’s language. However, a number of recent studies suggest that language
can nevertheless influence perceptual/cognitive representations of space by modulating
sensitivity to certain distinctions. These studies support the ‘Whorfian hypothesis’ that
language modifies thought – a hypothesis that was widely embraced during the 1950s
and 1960s, fell into disrepute during the 1970s and 1980s, and was resurrected in the
mid-1990s because of theoretical and methodological advances that have continued to
develop (e.g., Gentner and Goldin-Meadow, 2003; Gilbert et al., 2006).

4.2.1 Language can decrease sensitivity to certain categorical spatial distinctions

As summarized above, Munnich et al. (2001) found that even though Korean does
not lexicalize the contact/noncontact distinction, speakers are still sensitive to it for
nonlinguistic purposes such as recognition memory. However, there is also evidence
that in some cases sensitivity to a particular categorical spatial distinction is present
in infancy but then gradually diminishes during an early stage of language acquisition
because the distinction is not captured by the target language being learned. This type
of scenario is illustrated by a study that focused on the following contrast between
English and Korean strategies for describing actions involving topological relations of
containment (McDonough et al., 2003). The English expression put in specifies that F
ends up occupying an interior region of G, but is neutral with respect to whether F fits
tightly or loosely within G. In Korean, on the other hand, the notion of containment is
subdivided into two different categories: kkita designates the creation of a tight-fitting
relation between F and G (e.g., putting a cassette in a case), and nehta designates the
creation of a loose-fitting relation between F and G (e.g., putting an apple in a bowl).
Using a preferential looking paradigm as an indirect measure perceptual categorization,
McDonough et al. (2003) found that infants as young as 9 months of age, from both
English- and Korean-speaking environments, can discriminate between tight and loose
158 LANGUAGE, COGNITION AND SPACE

containment events (see also Hespos and Spelke, 2004). This kind of spatial sensitivity
is clearly useful for infants growing up in Korean-speaking environments, but it is
ultimately less valuable for infants growing up in English-speaking environments, and
in fact when adult speakers of each language were given the same preferential looking
task, the Korean speakers exhibited sensitivity to the tight/loose distinction, but the
English speakers did not. In another experiment that evaluated the adult speakers’
recognition of the distinction more explicitly, subjects observed the enactment of three
tight containment events and one loose containment event, and then answered the
question ‘Which is the odd one?’ Significantly more Korean- than English-speaking
adults based their choice on degree of fit (80% vs. 37%).
The investigators interpret their findings as evidence that when language-specific
spatial categories are being learned, the perceptual judgments that are necessary to use
them efficiently become increasingly rapid and automatic. Thus Korean speakers implic-
itly monitor the tightness of fit of containment relations because the grammatical system
of their language regularly forces them to encode distinctions along this parameter.
However, spatial sensitivities that are not needed in order to use the local language may
fade – e.g., English speakers can safely ignore the tight/loose contrast most of the time.
As McDonough et al. (2003) point out, the loss of sensitivity to the tight/loose contrast
is remarkably similar to another dramatic instance of perceptual tuning that takes place
during early language development, namely the loss of phonetic contrasts that are not
phonemic in the target language (Kuhl, 2004; Kuhl et al., 1992; Werker and Tees, 1984).
Linguistically induced downgrading of spatial acuity is also illustrated by Levinson’s
(2003: 152–4) report that Tzeltal speakers have difficulty distinguishing mirror stimuli
– e.g., b vs. d. This intriguing perceptual deficiency may derive from the fact that Tzeltal
makes no use of the egocentrically anchored relative frame of reference, relying instead
on two other locative strategies (apart from dispositional adjectives for describing
topological relations): body-part terms based on the intrinsic frame of reference for
describing closely contiguous projective relations, and directional terms based on the
absolute frame of reference for describing more distant projective relations. Because
the mirror stimuli employed in the experiment were multicomponent line figures,
they were most likely processed according to the intrinsic system, which is essentially
orientation-free and viewer-independent. The Tzeltal speakers’ difficulty in detecting
the difference between unreflected and reflected images cannot be attributed to low
education, since educationally matched speakers of Totonac, a Mayan language that
does use the relative frame of reference, performed the task much like Dutch speakers.
From the perspective of cognitive neuroscience, it is striking that the behavior of the
Tzeltal speakers resembles that of brain-damaged English and Italian speakers who
have selectively impaired mirror-stimulus discrimination (Davidoff and Warrington,
2001; Priftis et al., 2003; Turnbull and McCarthy, 1996; McCloskey, 2009). A direc-
tion for future research would be to carefully compare the presumably linguistically
induced decrement in discriminating mirror stimuli exhibited by Tzeltal speakers
with the clearly neurologically induced form of mirror stimulus agnosia exhibited by
brain-damaged subjects.
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 159

4.2.2 Language can increase sensitivity to certain categorical spatial distinctions

There are also reasons to believe that language can cause speakers to become more
attuned to particularly subtle or non-obvious types of spatial relationships. According
to Bowerman and Choi (2003, p. 417), ‘In cases like this, an important stimulant to
comparison can be hearing the same word. As the child encounters successive uses of
the word, she ‘tries’ (although this process is presumably rarely if ever conscious) to
align the referent situations and work out what they have in common. Sometimes …
there is no existing concept that does the job, and the child has to construct a new one to
account for the distribution of the word’. An excellent example is the Tiriyó morpheme
awee which refers to situations in which F is suspended from a point on G and hangs
down on either side of it, hence treating as equivalent such superficially diverse spatial
arrays as a necklace around a person’s neck, a tablecloth draped over a table, and a
clothespin dangling from a line. It is not known whether infants are sensitive to this
highly language-specific spatial category, but it seems likely that they are not and that
they must therefore gradually construct the concept through multiple exposures to
awee when acquiring Tiriyó. Another good example is the Chamus strategy of treating
the intrinsic front of a tree as either the side it leans toward or, in case it is perfectly
vertical, the side with the biggest branch or the most branches. It seems safe to assume
that these are features of trees that English speakers do not usually register, although
Chamus speakers must attend to them in order to use the grammatical system of the
language appropriately. In this manner language can be said to provide ‘on-the-job
training for attention’ (Smith et al., 2002). As Majid (2002) observes, it is useful to think
of this form of linguistically driven perceptual tuning as similar to the novice-to-expert
shift in categorization abilities that is known to engender more refined representations
for the target domain (Palmeri et al., 2004).
The most systematic and intensive investigation of linguistic influences on cognitive
representations of space has been conducted by Stephen Levinson and his colleagues
(Levinson, 2003; Majid et al., 2004; Pederson et al., 1998). Although this area of inquiry
is controversial (see the debate between Li and Gleitman, 2002, and Levinson et al.,
2002), several experiments suggest that there may be deep cognitive consequences of
speaking a language that employs predominantly either the relative frame of reference,
like English and Dutch, or the absolute frame of reference, like Guugu Yimithirr and
Tzeltal, for describing projective relations. The central experimental method involves
a rotation paradigm which makes it possible to identify the frame of reference that
subjects use to carry out various types of nonlinguistic cognitive tasks, such as memory
tasks that probe both recognition and recall, maze task that require tracking motion
and path direction, and reasoning tasks that evaluate transitive inference. To take a
straightforward example, subjects are first seated at a table on which three toy animals
are lined up headed leftward, or south, and then they are rotated 180º and seated at a
different table where they must arrange an identical set of toy animals so that they are
just as before, with an emphasis on remembering the linear order. If subjects orient the
animals in a leftward direction, they are invoking an egocentric frame of reference, but
if they orient the animals in a rightward (i.e., southerly) direction, they are invoking
160 LANGUAGE, COGNITION AND SPACE

an absolute frame of reference. When performing this as well as all other nonlinguistic
cognitive tasks involving the rotation paradigm, subjects overwhelmingly follow the
coding pattern of their language. Such results have been obtained with speakers of a
variety of ‘relative’ languages – e.g., English, Dutch, Japanese, and Yukatek (Mayan,
Mexico) – and ‘absolute’ languages – e.g., Guugu Yimithirr, Tzeltal, Arrernte, Hai//om
(Khoisan, Namibia), Longgu (Austronesian, Solomon Islands), Balinese (Austronesian,
Indonesia), and Belhare (Sino-Tibetan, Nepal).
Levinson (2003: 290–291) argues that these effects are due to the fact that relative
and absolute frames are incommensurable – e.g., from the proposition ‘The knife is to
the right of the fork’ one cannot derive the proposition ‘The knife is to the south of the
fork’, or vice versa:

Once a language has opted for one of these frames of reference and not the other,
all the systems that support language, from memory, to reasoning, to gesture, have
to provide information in the same frame of reference. If I remember an array as
‘The knife is to the right of the fork’ but live in a community where no left/right
terminology or computation is part of everyday life, I simply will not be able to
describe it. For my memory will have failed to support the local description system,
in, say, terms of north and south. The use of language thus forces other systems to
come into line in such a way that semantic parameters in the public language are
supported by internal systems keeping track of all experience coded in the same
parameters.

Despite Levinson’s assertions, this area of research remains quite contentious.

Nevertheless, there is sufficient data to motivate questions regarding the neural substrates
of linguistically driven cognitive restructuring. For example, although neuropsychologi-
cal studies have shown that some brain-damaged subjects with impaired knowledge of
the meanings of English prepositions can still accomplish tasks requiring nonlinguistic
processing of categorical spatial relations (Kemmerer and Tranel, 2000; Tranel and
Kemmerer, 2004), it is unknown how such subjects would perform on the various
kinds of nonlinguistic ‘space games’ that Levinson and his colleagues have developed
around the rotation paradigm. How would an English-speaking brain-damaged subject
with severely disrupted knowledge of left and right perform on the ‘animals in a row’
task described above? And what would the results reveal about the interface between
linguistic and cognitive representations of space? These questions, and many others that
involve integrating linguistic typology and cognitive neuroscience, await future research.

4.3 Summary

Experimental studies with normal as well as brain-damaged subjects suggest that the
meanings of locative expressions are language-specific semantic structures that are
activated primarily when a person packages his or her conceptualizations of space in a
manner that can easily be communicated in words – a process that Slobin (1996) calls
‘thinking for speaking’. These linguistic representations are at least partially distinct
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 161

from the perceptual/cognitive representations used in many visuospatial and visuomo-

tor tasks such as recognizing, drawing, and constructing complex spatial arrays. At the
same time, however, recent findings from the neo-Whorfian movement suggest that
the unique way in which one’s language structures space has implications for other
mental systems, bringing about not only modifications of perceptual sensitivities but
also adjustments of cognitive styles. I consider it a safe bet that most if not all of the
readers of this article are usually oblivious to their orientation with respect to north,
south, east, and west; however, there are a great many cultures in which people can
instantly indicate cardinal directions like these. Such profound cognitive differences
may be largely due to linguistic differences. The correct theory of the interface between
linguistically and nonlinguistically encoded categorical spatial relations remains a
topic of future research, and I submit that the most progress will be made through
interdisciplinary efforts that include the mutually informing perspectives of linguistic
typology and cognitive neuroscience.

5 Conclusion

People worldwide talk on a daily basis about the three-dimensional spatial world that
surrounds them, and most of the time the content of their discourse concerns broadly
defined spatial categories – e.g., The book is right here in front of me – as opposed to
metrically exact notions – e.g., The book is 53 centimeters from my chest. This is obvi-
ously a natural state of affairs. After all, coordinate details usually cannot be consciously
quantified with precision, and even if they could be, taking them into account in ordinary
linguistic communication would require an astronomical number of locative morphemes
to express all of the possible spatial relations – a situation that would be computation-
ally unmanageable and pragmatically otiose, not to mention utterly impossible in lan-
guages that completely lack the lexical and grammatical resources for complex counting
(Gordon, 2004). But even though almost all spatial discourse is restricted to schematic,
coarse-grained distinctions, there is nevertheless a vast range of coding possibilities, and
languages vary tremendously in how they carve up this multidimensional conceptual
domain, while still conforming to certain overarching tendencies. In this chapter I have
attempted to describe this rich field of semantic diversity from the perspective of cognitive
neuroscience. I have also suggested ways in which typological data can help guide future
research on how linguistically encoded categorical spatial relations are implemented in
the brain, and on how they interact with perceptual and cognitive representations of
space. The upshot of the chapter is captured by the following question: Would research
on the neural substrates of spatial representation be substantially different if the dominant
language in the world were, say, Tzeltal instead of English?

Notes
1 Portions of this chapter are reproduced from Kemmerer (2006).
2 Here and in what follows, the family and geographical area of languages that may not be
familiar to the reader are provided in parentheses.
162 LANGUAGE, COGNITION AND SPACE

3 Basque (Isolate, Europe), Dutch (Indo-European, Europe), Ewe (Niger-Congo, West

Africa), Lao (Tai-Kadai, Southeast Asia), Lavukaleve (Isolate, Solomon Islands), Tiriyó
(Cariban, South America), Trumai (Isolate, South America), Yélî Dnye (Isolate, Papua
New Guinea), Yukatek (Mayan, Mexico).
4 The search domain is always restricted, however, to the region directly adjacent to the
designated part of G, because when F is separated from G by a larger distance (even a
few inches, in most cases), a different set of locative terms is applied – specifically, terms
for cardinal directions, as described below in the subsection on the absolute frame of
reference.
5 Since this chapter went to press, the following new studies have come to my attention:
Wu et al. (2007), Chatterjee (2008), Amorapanth et al. (in press).
6 In a related study, Carlson et al. (2002) measured event-related brain potentials (ERPs)
while subjects judged the appropriateness of above for describing the location of a dot
relative to a watering can in a series of pictures in which the intrinsic and absolute
frames of reference were systematically manipulated.
7 See Coventry et al. (2008) for a recent psycholinguistic study that investigated this topic.
See also Bonfiglioli et al. (2009) for another perspective.
8 An important caveat, however, is that none of the tests was specifically designed to
distinguish between impaired processing of categorical spatial relations and impaired
processing of coordinate spatial relations.

References
Amorapanth, P., Widick, P. and Chatterjee, A. (in press). The neural basis for spatial
relations. Journal of Cognitive Neuroscience.
Amunts, K., Weiss, P.H., Mohlberg, H., Pieperhoff, P., Eickhoff, S., Gurd, J.M.,
Marshall, J.M., Shah, N.J., Fink, G.R. and Zilles, K. (2004) Analysis of neural
mechanisms underlying verbal fluency in cytoarchitectonically defined stere-
otaxic space – the roles of Bodmann areas 44 and 45. Neuroimage 22: 42–56.
Anderson, S. and Keenan, E. (1985) Deixis. In T. Shopen (ed.) Language typology
and syntactic description Vol. 3, Grammatical categories and the lexicon 259–307.
Cambridge, UK: Cambridge University Press.
Baciu, M., Koenig, O., Vernier, M.-P., Bedoin, N., Rubin, C. and Segebarth, C. (1999)
Categorical and coordinate spatial relations: fMRI evidence for hemispheric
specialization. NeuroReport 10: 1373–1378.
Benton, A.L. and Tranel, D. (1993) Visuoperceptual, visuospatial, and visuoconstruc-
tive disorders. In K.M. Heilman and E. Valenstein (eds) Clinical neuropsychology
165–213. (Third edition.) New York: Oxford University Press.
Berti, A. and Rizzolatti, G. (2002) Coding near and far space. In H.-O. Karnath,
D. Milner and G. Vallar (eds) The cognitive and neural bases of spatial neglect
119–30. Oxford, UK: Oxford University Press.
Bonda, E., Petrides, M., Frey, S. and Evans, A. (1995) Neural correlates of mental
transformations of the body-in-space. Proceedings of the National Academy of
Sciences USA 92: 11180–11184.
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 163

Bonfiglioni, C., Finocchiaro, C., Gesierich, B., Rositani, F. and Vescovi, M. (2009) A
kinematic approach to the conceptual representations of this and that. Cognition
111: 270–274.
Bowerman, M. and Choi, S. (2003) Space under construction: Language-specific
spatial categorization in first language acquisition. In D. Gentner and S. Goldin-
Meadow (eds) Language in mind 387–428. Cambridge, MA: MIT Press.
Brown, P. (1994) The INs and ONs of Tzeltal locative expressions. Linguistics 32: 743–790.
Brown, P. and Levinson, S.C. (1993) ‘Uphill’ and ‘downhill’ in Tzeltal. Journal of
Linguistic Anthropology 3: 46–74.
Brown, P. and Levinson, S.C. (forthcoming) Tilted worlds. Cambridge: Cambridge
University Press.
Burenhult, N. (2008) Spatial coordinate systems in demonstrative meaning. Linguistic
Typology 12: 99–142.
Carlson, L. and Van Der Zee, E. (eds) (2005) Functional features in language and
space: Insights from perception, categorization, and development. Oxford, UK:
Oxford University Press.
Carlson, L., West, R., Taylor, H.A. and Herndon, R.W. (2002) Neural correlates of
spatial term use. Journal of Experimental Psychology: Human Perception and
Performance 28: 1391–1408.
Carlson-Radvansky, L.A. and Irwin, D.A. (1993) Frames of reference in vision and
language: Where is above? Cognition 46: 223–244.
Casad, E. and Langacker, R.W. (1985) ‘Inside and ‘outside’ in Cora grammar.
International Journal of American Linguistics 51: 247–281.
Chatterjee, A. (2008) The neural organization of spatial thought and language.
Seminars in Speech and Language 29: 226–238.
Coventry, K. R., Prat-Sala, M. and Richards, L. (2001) The interplay between geom-
etry and function in the comprehension of over, under, above, and below. Journal
of Memory and Language 44: 376–398.
Coventry, K.R. and Garrod, S.C. (2004) Saying, seeing, and acting: The psychological
semantics of spatial prepositions. New York: Psychology Press.
Coventry, K.R., Valdés, B., Castillo, A. and Guijarro-Fuentes, P. (2008) Language
within your reach: Near-far perceptual space and spatial demonstratives.
Cognition 108: 889–895.
Damasio, H., Grabowski, T.J., Tranel, D., Ponto, L.L.B., Hichwa, R.D. and Damasio,
A.R. (2001) Neural correlates of naming actions and of naming spatial relations.
NeuroImage 13: 1053–1064.
Davidoff, J. and Warrington, E.K. (2001) A particular difficulty in discriminating
between mirror images. Neuropsychologia 39: 1022–1036.
Davidson, M. (1999) Southern Wakashan locative suffixes: A challenge to proposed
universals of closed-class spatial meaning. Paper presented at the Annual Meeting
of the International Cognitive Linguistics Association, Stockholm, Sweden.
Devlin, J.T., Matthews, P.M. and Rushworth, M.F.S. (2003) Semantic processing in
the left inferior prefrontal cortex: A combined functional magnetic resonance
imaging and transcranial magnetic stimulation study. Journal of Cognitive
Neuroscience 15: 71–84.
Diessel, H. (1999) Demonstratives. Amsterdam: John Benjamins.
164 LANGUAGE, COGNITION AND SPACE

Diessel, H. (2005) Distance contrasts in demonstratives. In M. Haspelmath, M.S.

Dryer, D. Gil and B. Comrie (eds) The world atlas of language structures 170–173.
Oxford: Oxford University Press.
Diessel, H. (2006) Demonstratives, joint attention, and the emergence of grammar.
Cognitive Linguistics 17: 463–490.
Dixon, R.M.W. (2003) Demonstratives: A cross-linguistic typology. Studies in
Language 27: 62–112.
Dunn, M., Meira, S. and Wilkins, D. (eds) (forthcoming) Demonstratives in cross-
linguistic perspective. Cambridge: Cambridge University Press.
Ekstrom, A.D., Kahana, M.J., Caplan, J.B., Fields, T.A., Isham, E.A., Newman, E.L.
and Fried, I. (2003) Cellular networks underlying human spatial navigation.
Nature 425: 184–187.
Emmorey, K. (2003) Perspectives on classifier constructions in sign languages.
Mahwah, NJ: Lawrence Erlbaum Associates.
Emmorey, K., Damasio, H., McCullough, S., Grabowski, T., Ponto, L.L.B., Hichwa,
R.D. and Bellugi, U. (2002) Neural systems underlying spatial language in
American Sign Language. NeuroImage 17: 812–824.
Enfield, N.J. (2003) Demonstratives in space and interaction: Data from Lao speakers
and implications for semantic analysis. Language 79: 82–117.
Fillmore, C.F. (1997) Lectures on deixis. Stanford: CSLI Publications.
Friederici, A. and Levelt, W.J.M. (1990) Spatial reference in weightlessness: Perceptual
factors and mental representations. Perception and Psychophysics 47: 253–266.
Gentner, D. and Goldin-Meadow, S. (eds) (2003) Language in mind. Cambridge, MA:
MIT Press.
Gerstmann, J. (1957) Some notes on the Gerstmann syndrome. Neurology 7: 866–869.
Gilbert, A.L., Regier, T., Kay, P. and Ivry, R.B. (2006) Whorf hypothesis is supported
in the right visual field but not the left. Proceedings of the National Academy of
Sciences 103: 489–904.
Gordon, P. (2004) Numerical cognition without words: Evidence from Amazonia.
Science 306: 496–499.
Graziano, M.S.A. and Aflalo, T.N. (2007) Rethinking cortical organization: Moving
away from discrete areas arranged in hierarchies. The Neuroscientist 13: 138–147.
Hafting, T., Fyhn, M., Molden, S., Moser, M.-B. and Moser, E.I. (2005)
Microstructure of a spatial map in the enthorinal cortex. Nature 436: 801–806.
Hanks, W.F. (2005) Explorations in the deictic field. Current Anthropology 46: 191–220.
Hartley, T., Maguire, E.A., Spiers, H.J. and Burgess, N. (2003) The well-worn route
and the path less traveled: Distinct neural bases of route following and wayfind-
ing in humans. Neuron 37: 877–888.
Haun, D.B.M., Call, J., Janzen, G. and Levinson, S.C. (2006a) Evolutionary psychol-
ogy of spatial representations in the Hominidae. Current Biology 16: 1736–1740.
Haun, D.B.M., Rapold, C.J., Call, J., Janzen, G. and Levinson, S.C. (2006b) Cognitive
cladistics and cultural override in Hominid spatial cognition. Proceedings of the
National Academy of Sciences 103: 17568–17573.
Heine, B. (1997) Cognitive foundations of grammar. Oxford, UK: Oxford University Press.
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 165

Hespos, S.J. and Spelke, E.S. (2004) Conceptual precursors to language. Nature 430:
453–456.
Heyman, K. (2006) The map in the brain: Grid cells may help us navigate. Science
312: 680–681.
Hillis, A.E. (2006) Neurobiology of unilateral spatial neglect. The Neuroscientist 12:
153–163.
Jager, G. and Postma, A. (2003) On the hemispheric specialization for categorical and
coordinate spatial relations: A review of the current evidence. Neuropsychologia
41: 504–515.
Janzen, G. and van Turennout, M. (2004) Selective neural representation of objects
relevant for navigation. Nature Neuroscience 7: 673–677.
Jeffery, K.J. and Burgess, N. (2006) A metric for the cognitive map: Found at last?
Trends in Cognitive Sciences 10: 1–3.
Kalaska, J.F., Cisek, P., Gosselin-Kessiby, N. (2003) Mechanisms of selection and
guidance of reaching movements in the parietal lobe. In A.M. Siegel, R.A.
Andersen, H.-J. Freund and D.D. Spencer (eds) The parietal lobes 97–120.
Philadelphia, PA: Lippincott Williams & Wilkins.
Kemmerer, D. (1999) ‘Near’ and ‘far’ in language and perception. Cognition 73: 35–63.
Kemmerer, D. (2005) The spatial and temporal meanings of English prepositions can
be independently impaired. Neuropsychologia 43: 797–806.
Kemmerer, D. (2006) The semantics of space: Integrating linguistic typology and
cognitive neuroscience. Neuropsychologia 44: 1607–1621.
Kemmerer, D. (in press) How words capture visual experience: The perspective from
cognitive neuroscience. In B. Malt and P. Wolff (eds) Words and the world: How
words capture human experience. Oxford, UK: Oxford University Press.
Kemmerer, D. (forthcoming) Visual and motor features of the meanings of
action verbs: A cognitive neuroscience perspective. In R. de Almeida and C.
Manouilidou (eds) Verb concepts: Cognitive science perspectives on verb represen-
tation and processing. Oxford, UK: Oxford University Press.
Kemmerer, D. and Tranel, D. (2000) A double dissociation between linguistic and
perceptual representations of spatial relationships. Cognitive Neuropsychology 17:
393–414.
Kemmerer, D. and Tranel, D. (2003) A double dissociation between the meanings of
action verbs and locative prepositions. Neurocase 9: 421–435.
Kemmerer, D. and Tranel, D. (2008) Searching for the elusive neural substrates of body
part terms: A neuropsychological study. Cognitive Neuropsychology 25: 601–629.
Kosslyn, S. (1987) Seeing and imagining in the cerebral hemispheres: A computa-
tional approach. Psychological Review 94: 148–175.
Kriegeskorte, N., Mur, M., Ruff, D.A., Kiani, R., Bodurka, J., Esteky, H., Tanaka,
K. and Bandettini, P.A. (2008) Matching categorical object representations in
inferior temporal cortex of man and monkey. Neuron 60: 1126–1141.
Kuhl, P.K. (2004) Early language acquisition: Cracking the speech code. Nature
Reviews Neuroscience 5: 831–843.
166 LANGUAGE, COGNITION AND SPACE

Kuhl, P.K., Williams, K.A., Lacerda, F., Stevens, K.N. and Lindblom, B. (1992)
Linguistic experience alters phonetic perception in infants by 6 months of age.
Science 255: 606–608.
Laeng, B. (1994) Lateralization of categorical and coordinate spatial functions: A
study of unilateral stroke patients. Journal of Cognitive Neuroscience 6: 189–203.
Laeng, B., Chabris, C.F. and Kosslyn, S.M. (2003) Asymmetries in encoding spa-
tial relations. In K. Hugdahl and R.J. Davidson (eds) The asymmetrical brain
303–339. Cambridge, MA: MIT Press.
Landau, B. and Jackendoff, R. (1993) ‘What’ and ‘where’ in spatial language and
spatial cognition. Behavioral and Brain Sciences 16: 217–238.
Leutgeb, J.K., Leutgeb, S.L., Moser, M.-B. and Moser, E.I. (2007) Pattern separation in
the dendate gyrus and CA3 of the hippocampus. Science 315: 961–966.
Levinson, S.C. (1983) Pragmatics. Cambridge, UK: Cambridge University Press.
Levinson, S.C. (1994) Vision, shape, and linguistic description: Tzeltal body-part
terminology and object description. Linguistics 32: 791–856.
Levinson, S.C. (2003) Space in language and cognition. Cambridge, UK: Cambridge
University Press.
Levinson, S.C., Kita, S., Haun, D.B.M. and Rasch, B.H. (2002) Returning the tables:
Language affects spatial reasoning. Cognition 84: 155–188.
Levinson, S.C. and Meira, S. (2003) ‘Natural concepts’ in the spatial topological
domain – adpositional meanings in crosslinguistic perspective: An exercise in
semantic typology. Language 79: 485–516.
Levinson, S.C. and Wilkins, D. (eds) (2006) Grammars of space. Cambridge, UK:
Cambridge University Press.
Li, P. and Gleitman, L. (2002) Turning the tables: Language and spatial reasoning.
Cognition 83: 265–94.
Longo, M.R. and Lourenco, S.F. (2006) On the nature of near space: Effects of tool
use and the transition to far space. Neuropsychologia 44: 977–981.
Luschi, P., Benhamoou, S., Girard, C., Ciccione, S., Roos, D., Sudre, J. and Benvenuti,
S. (2007) Marine turtles use geomagnetic cues during open-sea homing. Current
Biology 17: 126-133.
Majid, A. (2002) Frames of reference and language concepts. Trends in Cognitive
Sciences 6: 503–504.
Majid, A., Bowerman, M., Kita, S., Haun, D.B.M. and Levinson, S.C. (2004) Can
language restructure cognition? The case for space. Trends in Cognitive Sciences
8: 108–114.
Majid, A., Enfield, N. and van Staden, M. (eds) (2006) Parts of the body:
Crosslinguistic categorization. Special issue of Language Sciences 28: 137–360.
MacSweeney, M., Woll, B., Campbell, R., Calvert, G.A., McGuire, P.K., David,
A.S., Simmons, A. and Brammer, M.J. (2002) Neural correlates of British Sign
Language comprehension: Spatial processing demands of topographic language.
Journal of Cognitive Neuroscience 14: 1064–1075.
Makin, T.R., Holmes, N.P. and Zohary, E. (2007) Is that near my hand? Multisensory
representation of peripersonal space in human intraparietal sulcus. Journal of
Neuroscience 27: 731–740.
 A NEUROSCIENTIFIC PERSPECTIVE ON THE LINGUISTIC ENCODING OF CATEGORICAL SPATIAL RELATIONS 167

Mayer, E., Martory, M.-D., Pegna, A.J., Landis, T., Delavelle, J. and Annoni, J.-M.
(1999) A pure case of Gerstmann syndrome with a subangular lesion. Brain 122:
1107–1120.
Mazzoni, M., Pardossi, L., Cantini, R., Giorgetti, V. and Arena, R. (1990) Gerstmann
syndrome: A case report. Cortex 26: 459–467.
McCloskey, M. (2009) Visual reflections: A perceptual deficit and its implications.
Oxford, UK: Oxford University Press.
McDonough, L., Choi, S. and Mandler, J.M. (2003) Understanding spatial relations:
Flexible infants, lexical adults. Cognitive Psychology 46: 229–259.
Milner, A.D. and Goodale, M.A. (2006) The visual brain in action. (Second edition.)
Oxford, UK: Oxford University Press.
Morris, H.H., Luders, H., Lesser, R.P., Dinner, D.S. and Hahn, H. (1984) Transient
neuropsychological abnormalities (including Gerstmann’s syndrome) during
cortical stimulation. Archives of Neurology 34: 877–883.
Munnich, E., Landau, B. and Dosher, B.A. (2001) Spatial language and spatial repre-
sentation: A crosslinguistic comparison. Cognition 81: 171–207.
Noordzij, M.L., Neggers, S.F.W., Postma, A. and Ramsey, N. (2008) Neural correlates
of locative prepositions. Neuropsychologia 46: 1576–1580.
Palmeri, T.J., Wong, A.C.N. and Gauthier, I. (2004) Computational approaches to the
development of perceptual expertise. Trends in Cognitive Sciences 8: 378–386.
Papafragou, A. (2007) Space and the language-cognition interface. In P. Carruthers,
S. Laurence and S. Stich (eds) The innate mind: Foundations and the future.
Oxford, UK: Oxford University Press.
Pederson, E., Danziger, E., Wilkins, D., Levinson, S., Kita, S. and Senft, G. (1998)
Semantic typology and spatial conceptualization. Language 74: 557–589.
Peelen, M.V. and Downing, P.E. (2007) The neural basis of visual body perception.
Nature Reviews Neuroscience 44: 1515–1518.
Postma, A. and Laeng, B. (2006) New insights in categorical and coordinate process-
ing of spatial relations. Neuropsychologia 8: 636–648.
Previc, F.H. (1998) The neuropsychology of 3-D space. Psychological Bulletin 124:
123–164.
Priftis, K., Rusconi, E., Umilta, C. and Zorzi, M. (2003) Pure agnosia for mirror
stimuli after right inferior parietal lesion. Brain 126: 908–919.
Ritz, T., Thalu, P., Phillips, J.B., Wiltschko, R. and Wiltschko, W. (2004) Resonance
effects indicate a radical-pair mechanism for avian magnetic compass. Nature
429: 177–180.
Robertson, L. (2004) Space, objects, minds, and brains. New York: Psychology Press.
Roeltgen, D.P., Sevush, S. and Heilman, K.M. (1983) Pure Gerstmann’s syndrome
from a focal lesion. Archives of Neurology 40: 46–47.
Sakata, H. (2003) The role of the parietal cortex in grasping. In A.M. Siegel, R.A.
Andersen, H.-J. Freund and D.D. Spencer (eds) The parietal lobes 121–140.
Philadelphia, PA: Lippincott Williams & Wilkins.
Sargolini, F., Fyhn, M., Hafting, T., McNaughton, B.L., Witter, M.P., Moser, M.B. and
Moser, E.I. (2006) Conjunctive representation of position, direction, and velocity
in entorhinal cortex. Science 312: 758–762.
168 LANGUAGE, COGNITION AND SPACE

Slobin, D.I. (1996) From ‘thought and language’ to ‘thinking for speaking’. In
J.J. Gumperz and S.C. Levinson (eds) Rethinking linguistic relativity 70–96.
Cambridge, UK: Cambridge University Press.
Smith, L.B., Jones, S.S., Landau, B., Gershkoff-Stowe, L. and Samuelson, L. (2002)
Object name learning provides on-the-job training for attention. Psychological
Science 13: 13–19.
Spiers, H.J. and Maguire, E.A. (2006) Thoughts, behaviour, and brain dynamics
during navigation in the real world. NeuroImage 31: 1826–1840.
Svorou, S. (1994) The grammar of space. Amsterdam: John Benjamins.
Talmy, L. (1983) How language structures space. In H. Pick and L. Acredolo (eds)
Spatial orientation: Theory, research, and application 225–282. New York:
Plenum Press.
Thompson-Schill, S.L., D’Esposito, M. and Kan, I.P. (1999) Effects of repetition and
competition on activity in left prefrontal cortex during word generation. Neuron
23: 513–522.
Tranel, D. (1996) The Iowa-Benton school of neuropsychological assessment. In I.
Grant and K.M. Adams (eds) Neuropsychological assessment of neuropsychiatric
disorders 81–101. (Second edition.) New York: Oxford University Press.
Tranel, D. and Kemmerer, D. (2004) Neuroanatomical correlates of locative preposi-
tions. Cognitive Neuropsychology 21: 719–749.
Turnbull, O.H., Beschin, N. and Della Sala, S. (1997) Agnosia for object orientation:
Implications for theories of object recognition. Neuropsychologia 35: 153–163.
Turnbull, O.H. and McCarthy, R.A. (1996) Failure to discriminate between mirror-
image objects: A case of viewpoint independent object recognition? Neurocase
2: 63–72.
Tyler, A. and Evans, V. (2003) The semantics of English prepositions: Spatial scenes,
embodied experience and cognition. Cambridge: Cambridge University Press.
Ulltan, R. (1978) Some general characteristics of interrogative systems. In J.
Greenberg (ed.) Universals of human language Vol. IV 211–248. Stanford, CA:
Stanford University Press.
Van Driem, G. (1987) A grammar of Limbu. Berlin: Walter de Gruyter.
Varney, N.R. (1984) Gerstmann syndrome without aphasia. Brain and Cognition 3: 1–9.
Wagner, A.D., Pare-Blagoev, E.J., Clark, J. and Poldrack, R.A. (2001) Recovering
meaning: Left prefrontal cortex guides controlled semantic retrieval. Neuron 31:
329–338.
Werker, J.F. and Tees, R.C. (1984) Cross-language speech perception: Evidence
for perceptual reorganization during the first year of life. Infant Behavior and
Development 7: 49–63.
Wierzbicka, A. (2007) Bodies and their parts: An NSM approach to semantic typol-
ogy. Language Sciences 29: 14–65.
Wu, D.H., Waller, S. and Chatterjee, A. (2007) The functional neuroanatomy of the-
matic role and locative relational knowledge. Journal of Cognitive Neuroscience
19: 1542–1555.
Zacks, J., Rypma, B., Gabrieli, J.D., Tversky, B. and Glover, G.H. (1999) Imagined
transformations of human bodies: An fMRI investigation. Neuropsychologia 37:
1029–1040.
 Part IV
Theoretical approaches to spatial
representation in language

169
7 Genesis of spatial terms
Claude Vandeloise

A parallelism is often established between the production of a language by a culture

(phylogeny) and its reproduction by children (ontogeny). The basic spatial words in
different languages will be used in this article in order to investigate the similarities and
the discrepancies between the two processes. Concerning the creation of spatial words,
section 1 establishes a contrast between what I call external lexical formation, in which
a word is associated to an extra-linguistic concept, and internal lexical formation, that
proceeds by division or union of established lexical categories. In section 2, I will discuss
a hierarchy in the formation of spatial terms in languages of the world (Levinson and
Meira 2003) inspired by an implicational scale for the creation of basic color terms
proposed by Berlin and Kay (1968). MacLaury (1993) motivates this development by
creating a hierarchy involving a process of internal lexical formation by division. I will
compare these hypotheses to another hierarchy proposed in Vandeloise (2003, 2005).
This hierarchy establishes a basic contrast between the relation of localization, conveyed
in French by the preposition à, and the dynamic relations of control, expressed by in
and on.
Three modes of lexicon development are investigated in section 3. Whereas the
creation of basic color terms may go from the most general to the most specific, as
illustrated by MacLaury, the creation of words often evolves in the reverse direction,
from the application of a word to very specific situations to its extension to more general
uses. In contrast to the former mode of creation that operates by division, the latter
mode of internal lexical formation proceeds by union. If external lexical creation anchors
a word in the middle of a hierarchy of concepts, both processes can occur to create
supercategories and subcategories. In contrast to the linguistic community that builds
its language from scratch, infants pick their first spatial words inside a complete and
well-structured language. In section 4, I will attempt to explain how the different levels
of abstraction of spatial words can influence the acquisition of spatial words.

1 External and internal lexical formation

According to one of the main dogmas of structuralism, the meanings of words emerge
negatively, from their differences with other words in the language. These differential
meanings are called values (Saussure 1916). I will come back to them when I speak
of internal lexical formation. This conception of meaning, however, poses an obvi-
ous logical problem once one considers the production of language and the first
words created in the lexicon. This problem has not been urgent as long as language
creation was considered a taboo subject, unworthy of linguists’ attention. Once this
interdiction is transgressed, though, one must admit that, according to the differential

171
172 LANGUAGE, COGNITION AND SPACE

hypothesis, the first words can only be created by pairs (x, y), with x determining the
value of y and vice versa. This may make sense for pairs like here and there or yes and
no. But if one admits that among the first words appear also terms for actions (like
eat) or names for persons (like Peter), complementary words designating any action
that is not eating, or any human who is not Peter, are more difficult to conceive. The
meaning of these words cannot emerge from differences in a system but may only be
explained by the extra-linguistic stimulations that make these terms convenient to
ensure the good functioning of the society in which they emerge. This is what I call
external lexical formation. It occurs when the members of a society share a common
interest in an aspect of their environment or of their social life; when they are able to
recognize this aspect in a sufficiently similar way; and when they associate a term to
this aspect of their lives.
The existence of external lexical formation, mainly based on similarities between the
occurrences in the world designated by a word, does not preclude a very important role
for internal lexical formation. In this case, a term is applied to aspects of environment
or social life because differences with aspects of the world designated by the available
words in the lexicon begin to appear pertinent for the ease of communication. Linguists
have been much more interested in internal lexical formation. Its functioning is much
better documented than the development of external lexical formation. The domain
of colors is a perfect field to observe this type of formation. The work of Berlin and
Kay (1968), devoted to basic color terms and to their hierarchical appearance in the
development of language, will be essential for this article. This book was equally an
important source of inspiration for the typology proposed by Levinson and Meira (2003)
that will be discussed in the next section. According to these authors’ interpretation
of Berlin and Kay’s implication scale, color terms appear in the following order in the
languages of the world:

White +Black→ Red→ Green or Yellow→ Yellow or Green→ Blue→ Brown→ Purple
Pink
Orange
Grey

This means that a language that possesses a color term on the right of the scale neces-
sarily includes all others to the left of this term.
The formation of basic color terms in this implicational scale cannot be explained
by internal lexical formation only. At the beginning of the scale, an internal lexical
formation of white and black might be justified by the contrast between day and night
(Wierzbicka, 1990). In this case, however, it would not be a genuine color contrast. Taken
together, white and black might be opposed to a word meaning colorful but certainly not
to red alone as proposed in the above scale. At the end of the implication scale, brown
is also very unlikely to be created from a category including brown and blue by internal
lexical formation. Some amount of external lexical formation, then, must be involved
in the creation of the first and the last ‘basic color terms’.
 GENESIS OF SPATIAL TERMS 173

The creation of basic color terms has been carefully observed by MacLaury (1993),
who compares the evolution of two Mayan languages, Tzeltal of Tenejapa and Tzotzil
of Novenchuuc, from a system of two color terms to a system of six color terms. He
proposes an interpretation of Berlin and Kay’s implicational scale that is more compatible
with internal lexical formation (see chart 1).

dark light
cool warm
1 2

dark cool light warm

green blue red yellow

5 6 3 4

Chart 1. MacLaury’s implicational scale

In chart 1, the places of black and white, the two first terms of the implicational scale,
are occupied by the category of dark or cool colors and by the category of light or
warm colors, respectively. Red and yellow, third and fourth in the implicational scale,
are the result of the split of warm colors whereas green and blue, fifth and sixth in
the implicational scale, are the result of the split of cool colors .1 What is the destiny
of dark and light at the second level remains an open question: do these categories
remain without linguistic representation, or are they conveyed by words equivalent
to black and white?
The category of cool colors that gathers green and blue in chart 1 is often called ‘grue’.
The split of the category of cool colors into green and blue provides an exemplary case
of internal lexical formation. At the beginning, suppose that green is indifferently used
for the green or blue tokens of the ‘grue’ category. Inside this general category appears a
new word blue that is used by innovative speakers for blue objects only. At a first stage,
green can still be used for blue objects, in such a way that blue may be considered as a
hyponym of green. With evolution, and the disappearance of more conservative speakers
who prefer green to blue, a second stage appears at which green can no longer be applied
to blue objects and restricts itself to ‘grue’ objects that are not blue, i.e. to green objects.
In this way, at the third and final stage, the connection between green and blue is severed:
green applies only to green objects and blue to blue objects.
The case of the ‘grue’ category is a perfect example of internal lexical creation,
because the similarities between green and blue makes plausible the existence of a
category including the two colors. In contrast, at the origin of the implicational scale of
174 LANGUAGE, COGNITION AND SPACE

Berlin and Kay, it is difficult to imagine a category gathering black and white from which
these terms emerge. White and black, then, are examples of external lexical formation
and green and blue are examples of internal lexical formation, as illustrated in Figure 1.
External and internal lexical formation will prove useful in section 2 for the comparison
of English spatial words in and on with Spanish en and sobre.

White Black Green Blue

Figure 1. Internal vs. external lexical creation

2 Creation of spatial words

Berlin and Kay (1968) provide an implicational scale according to which basic color
terms develop in the languages of the world. This may provide a first insight in lexical
formation. In this article, I will be concerned with the creation of spatial terms. I will
present the analysis of Levinson and Meira (2003) before arguing for an alternative
solution based on preceding articles (Vandeloise 2003, 2005). Like Berlin and Kay (1968),
Levinson and Meira use a sample of genetically unrelated languages. Informants in
each language were asked to ascribe an adposition in their language to a booklet of 71
line-drawings known under the name of ‘topological relations picture series’. As in the
case of the attribution of basic color terms, the choices tended to cluster and were not
randomly distributed as they would be if there were no crosslinguistic generalizations.
The five main clusters are labeled IN, NEAR/UNDER, ON/OVER, ATTACHMENT
and ON-TOP. On the basis of these data, Levinson and Meira propose the following
implicational scale for spatial terms.

AT < IN < ON < OVER< ON TOP < ATTACHED < INSIDE < SPIKED
UNDER NEAR HANGING
DISTRIBUTED OVER

They elaborate this implicational scale to show how different languages can develop
spatial terms in different ways. I modify the presentation of their analysis (Figure 18,
p. 512), in order to make the comparison with my solution easier.
 GENESIS OF SPATIAL TERMS 175

AT1

AT2 IN-2D
IN-3D

AT3 UNDER ON1 IN1 INSIDE

OVER
ON TOP
ATTACHMENT

AT4 NEAR

Option 1 Option 2

ON2 OVER ON2’ ATTACHMENT

ON TOP ON TOP
ATTACHMENT OVER

Chart 2. Levinson and Meira’s implicational scale

AT1 is a unique spatial notion that covers all the spatial relations and corresponds to
the adposition ta in a language like Tzeltal or di in Indonesian (Feist 2004). AT2, AT3
and AT4 are more and more specific notions. Thus, AT2 covers all spatial relationships
with the exception of those conveyed by IN-2D and IN-3D and AT4 is a residue that
excludes all the preceding notions, including NEAR. These processes correspond to
internal lexical formation by division. Vertically aligned notions, such as IN-3D and
IN-2D, as well as ON1, OVER, ON-TOP and ATTACHMENT, correspond to composite
notions. Levinson and Meira split IN-2D and IN-3D because they attribute two foci to
the category IN, one focus specifying containment in a three-dimensional container
and the other inclusion in a two-dimensional plane. I will come back to this decision
later in this section. Indices attached to ON1 and ON2 do not appear in Levinson and
Meira’s chart but they will make the exposition easier. At the last level of specification
of chart 2, two options are offered for the decomposition of the complex concept ON1,
OVER, ON TOP and ATTACHMENT.
Levinson and Meira use capitals AT, IN and so forth to represent the ‘central mean-
ings of the relevant sort’ (footnote 2, p. 486) associated to the basic topological notions
conveyed by at, in and so forth.2 ATTACHMENT is an exception to this convention. The
authors are obliged to use this notional term instead of a preposition because English
has no specific adposition to convey attachment. The use of capitals may be a handy
way of introducing the prototypes of spatial relations but it raises some questions. First,
why is AT chosen to represent the most general category in chart 2? The preposition
at appears nowhere in the data or in the article and it is certainly not a good example
of an inclusive spatial preposition. As we will see later, the Old English preposition œt
might fit this role better. Second, there are discrepancies between the data coming from
the experiments summarized in the map proposed by Levinson and Meira (p. 505)
176 LANGUAGE, COGNITION AND SPACE

and chart 2. Indeed, IN represents a coherent cluster in the map but it splits in IN-2D
and IN-3D in the chart. As a matter of fact, this is so because all the members in the
IN-cluster correspond to IN-3D. Therefore, one may doubt whether IN-2D represents a
‘central meaning of the relevant sort’. The reason why IN-3D and IN-2D are grouped in
the chart is not because they are notionally related but because an identical preposition
is assigned to them in English. On the other hand, NEAR/UNDER, as well as ON/
OVER, corresponds to one cluster in the map of data but NEAR and UNDER, as well
as ON and OVER, are disjointed in the chart. In my alternative, instead of AT, IN (with
or without contact), I will use explicit notions like LOC(ALIZATION), CONTROL and
so forth. For each notion, I will provide an example of a preposition attached to this
notion in a language of the world.
Levinson and Meira are uniquely concerned with basic topological relationships
between the located target and the landmark that locates it. For example, among the
clusters in the map of data, ON/OVER is characterized by superadjacency (with or
without contact), UNDER by subadjacency (with or without contact) and NEAR
by proximity. Contiguity and coincidence are further topological notions present
in the analysis. In contrast, the cluster IN corresponds to full containment (p. 508).
Containment is certainly not a topological notion but the distinction between con-
tainment and the topological notion of inclusion is often blurred in the literature
and many scholars appear to use them indifferently. In chart 2, Levinson and Meira
make a distinction between IN-3D (a notion close to CONTAINMENT) and IN-2D
(close to INCLUSION). In this way, they introduce a further topological notion in
their analysis. The authors mention that a reviewer of their article ‘questions to what
extent ‘attachment’ (and indeed other notions like ‘containment’ and ‘support’) are
really spatial as opposed to mechanical in conception’ (footnote 9, p. 487). The authors
admit that some doubt is in order. The alternative proposed in chart 3 reinforces the
contrast between topological basic categories and puts the role of force and energy
to the forefront. Therefore, the first dichotomy established in chart 3 distinguishes
between LOC (topology3) and CONTROL (dynamics). The discrepancies in the lin-
guistic representation of these notions in the languages of the world create a problem
for the typology proposed in chart 2. Indeed, the Spanish preposition en conveys
both ON and IN notions. But in chart 2, ON and IN do not have a common direct
hyperonym. If these notions belong to different branches of the structure, one may
wonder why so many languages, like Spanish and Modern Greek, have a common
adposition to designate both notions. The common status of IN and ON as opposed
to AT constitutes the main discrepancy between my analysis and that of Levinson’s
and Meira’s. Another famous example concerning control is Korean in which two
verbs correspond to the English preposition in: the verb kkita that conveys tight fit,
as opposed to loose containment conveyed by the verb nehta.4 Chart 3 makes this
distinction possible.
Levinson and Meira exclude very pervasive spatial relations such as projective
notions IN FRONT and BEHIND from their inquiry because ‘projective concepts
belong to a different conceptual subdomain, where coordinate systems or frames of
reference are necessary’ (p. 488). It is true that projective prepositions do not cor-
respond to topological notions, but control prepositions like in and on do not either.
 GENESIS OF SPATIAL TERMS 177

Projective spatial prepositions are basic spatial prepositions that will be introduced in
chart 3. On the other hand, like Levinson and Meira, I will limit myself to static spatial
relations. However, chart 4 describing the evolution of the Old English preposition
œt will show how kinetic prepositions such as from and to might be incorporated
in the genesis of spatial terms. UNDER and OVER will also be excluded from basic
spatial categories in chart 3 for reasons I will give below. These two categories have a
peculiar status in the analysis of Levinson and Meira. First, UNDER is the only basic
category to appear simultaneously with another category (ON1) in the partition of
AT2. No explanation is provided for this exception. 5 Second, OVER appears first
as a component of the composite concept ON1/ON TOP/OVER/ATTACHMENT.
Depending on the languages, this composite category can split in two different ways.
The former option, with OVER excluded, appears in English whereas Levinson and
Meira attribute the latter option to Yucatec and Ewe. In their chart, they put ON1,
OVER, ON-TOP and ATTTACHMENT at the same level. The relationship between
ON1 and ON-TOP is unclear. The only definition provided for the latter notion is
‘location above the eye-level’ (p. 512), a definition contradicted by the utilization of
ON-TOP for a picture representing a table covered by a tablecloth.6 As a matter of
fact, except for this example, the pictures to which ON-TOP is ascribed correspond to
the prototypical uses of on in English. Therefore, instead of being at the same level as
ON-TOP, ATTACHMENT and OVER, ON1 might be considered as a more general
notion, including these three categories. One may furthermore cast in doubt the useful-
ness of the category ON-TOP. At any rate, in contrast to OVER and ATTACHMENT,
ON-TOP never dissociates itself from ON1 at the last level of specification.
In chart 3, the equivalent of AT1 is called RELATION IN SPACE. These relations
imply accessibility in space between two material entities; between a material and a
spatial entity; or between two spatial entities. A spatial entity may be a place occupied
by a material entity or a portion of space that material entities might occupy. Linguistic
communities attribute names to geographic spatial entities. In the case of material enti-
ties, accessibility is guaranteed by contact or proximity.7 When a spatial landmark is
involved in the relationship, there is coincidence or proximity of the target with the
landmark. This coincidence is often partial since the landmark is usually larger than
the target. Coincidence between two material entities is impossible. In many respects,
proximity appears to be the most general ingredient of a relation in space. As a matter
of fact, if contact and coincidence are considered as limit cases of proximity, this notion
might be chosen to characterize relations in space at the most general level. Therefore,
the occurrence in chart 2 of NEAR (representing proximity) at the same level as the
most specific notion AT4 is surprising. How can one and the same concept appear
together at the most general and at the most specific level? Because it is associated to
the primitive notion of proximity, near to appears deceptively as a basic expression.
However, far from being basic, the syntax of close to or near (to) in English or of près
de in French demonstrate that these locutions, though related to the primitive concept
of proximity, are complex notions. Near in English may be an adjective as well as a
preposition and there are discussion in French (Gunnarson 1986) about whether près
should be treated as an adverb, an adjective or a preposition. In the genesis I propose
in chart 3, projective relationships in the vertical axis and in the horizontal plane will
178 LANGUAGE, COGNITION AND SPACE

appear instead of NEAR. As a matter of fact, near (to) might be considered as a late
hyperonym for all the prepositions involving proximity in the horizontal plane.
The most important division in chart 3 separates CONTROL (that implies an
exchange of energy between the landmark and the target) from a residue of spatial
relations, called LOC1, that do not involve such an exchange of energy.8 By this division,
LOC1 is deprived of all the relations in space involving contact between two material
entities, since – if one forgets magnets and radiations – contact is a necessary condi-
tion for control. LOC2, then, is left with the spatial relationships involving at least one
spatial entity on the one hand; and with the relationships between material entities
that are not in contact on the other hand. Thus, LOC2 means that the target (partially)
coincides with a spatial landmark; or that it is close to a spatial or material landmark.
At the corresponding stage of the development, chart 2 subtracts NEAR from AT3. If
PROXIMITY were similarly subtracted from LOC2 in chart 3, LOC3 would be restricted
to the spatial relationships containing at least one spatial entity and implying coincidence
of the target and the landmark. Reasons to avoid the presence of PROXIMITY at this
stage have been evoked in the preceding paragraph. Categories split because a subset
of their members attracts more attention than the others or because there is a need for
explicitness. In conformity with this principle, IN-3D in chart 2 or CONTROL in chart 3
are relations in space more constrained and prominent than AT2 or LOC1, respectively.
But why should relationships of proximity be more prominent than relationships of
coincidence? Quite the contrary! Coincidence with the landmark locates the target
more precisely than proximity which needs specification. This specification is, I believe,
the role of projective prepositions. Actually, in the vertical axis, proximity may be too
strong a word since the sun is above the earth does not involve proximity between the
sun and the earth. Separation between the target and the landmark may be sufficient
for the use of projective prepositions. I will make a distinction between separation in
the vertical axis (VERTICAL SEPARATION) and separation in the horizontal plane
(HORIZONTAL SEPARATION). Whereas VERTICAL SEPARATION admits material
landmarks (the lamp is over the table) as well as spatial landmarks (the airplane is above
Paris), HORIZONTAL SEPARATION prefers material landmarks and is used with
difficulty with spatial landmarks: ?The car is to the left of Paris. Also, in French, au-dessus
and en dessous maintain a connection with coincidence since, in contrast to plus haut
and plus bas, these prepositions require coincidence of the vertical projection of the
target with the landmark. For these reasons, in chart 3, I will first subtract VERTICAL
SEPARATION from LOC2 and then, subtract HORIZONTAL SEPARATION from
LOC3. In internal lexical division, the emergence of a new term makes the use of the
old term obsolete. This is true in the vertical axis in the French examples below:
(1) La lampe est au-dessus/*à/?près de la table
(2) La chaise est devant/ *à/ près de la table
(3) Le chat est à gauche/*à/ près de la table

Près de, in contrast, is compatible with the projective horizontal prepositions.

 GENESIS OF SPATIAL TERMS 179

RELATION IN SPACE
(ta)

LOC1 CONTROL
(à1)1 (en)

LOC2 VERTICAL SEPARATION CONTAINMENT SUP(PORT)1

(à2) (au-dessus/en dessous) (in) (on)

LOC3 HORIZONTAL SEPARATION2 TIGHT FIT LOOSE FIT SUP2 ATTACH.

(à3) (devant/derrière) (kkita) (nehta) (op) (aan)
(à gauche/à droite)
1
Besides static location, the French preposition à can also introduce the goal of the target like the
preposition to.
2
Separation along the frontal direction and along the lateral direction might need to be treated separately.

Chart 3. A hierarchy of concepts

Chart 3 can be understood as a hierarchy of concepts going from the most abstract
level to the most concrete levels. Languages like Tzeltal (Brown 1994) have only one
adposition at the first level – ta – that introduces any spatial relations in space and
leaves to verbs and nouns the elaboration of these relations. At this abstract level,
the only spatial term opposes relations in space to the other grammatical functions,
marked by cases like nominative, accusative and so forth. All the notions below LOC1
are illustrated by prepositions in the same language. I have chosen French because à is
more clearly related to localization (Vandeloise 1991) than at or in. Whenever a more
specific preposition is added, the extension of the most general preposition diminishes.
The process going from à1 to à3 proceeds by subtraction like the evolution from AT1
to AT4 in the analysis of Levinson and Meira. LOC1 splits in LOC2 and VERTICAL
SEPARATION; and LOC2 in LOC3 and HORIZONTAL SEPARATION.
The nature of the development going from LOC1 to the more specific levels in the
chart is different from the development of CONTROL. Whereas the former notion
evolves by division, the latter develops by specification. In contrast to the development
of LOC1, the development of CONTROL in chart 3 is exemplified by prepositions in
different languages. General control is conveyed by the Spanish preposition en. The
prepositions in and on in English, and the Korean verbs kkita and nehta as well as
the Dutch prepositions op and aan, correspond to more and more specific types of
control. Whereas in support, conveyed by on, the bearer controls the burden in the
vertical direction only, containment, conveyed by in, requires control in more than
one direction. Kkita and nehta mark tight fit and loose containment of the target in the
landmark respectively9 while op and aan convey direct support and indirect support,
respectively. I propose this hierarchy in Vandeloise (2003) to show that the relativity
in the description of space illustrated by Spanish, English, Korean and Dutch is less
180 LANGUAGE, COGNITION AND SPACE

dramatic than claimed by Bowerman (1996). Spanish en, English in and Korean kkita
convey control at different levels of specificity.
In the development of localization, the evolution occurs mainly by internal lexical
formation. In the case of control, some languages might overlook the most abstract
notions and immediately establish a connection at the level of more specific concepts.
Thus, whereas in the development of localization, a specific preposition (like au-dessus)
reduces the scope of a more general preposition (like à), in the case of control, there is no
evidence that a general preposition of control existed at an earlier stage of French, even
though dans and sur specify the Spanish preposition en. The more specific prepositions
dans and sur, covering approximately the scope of the Spanish preposition en, might
have appeared simultaneously, or at least independently, by external lexical formation.
This is the reason why the examples illustrating the development of control prepositions
in chart 3 are taken from different languages. This does not mean that the expression
of control never developed in the same way as the expression of localization. I would
like to remain neutral on this point.
From AT1 to AT4, the extension of the preposition of general localization shrinks
each time a more specific preposition appears. The history of languages might support
this type of development. Notably, the evolution of the preposition œt in Old English
illustrates the mode of production by internal formation. Besides the meaning of the
present preposition at, œt could convey (1) the origin of movement; (2) proximity to
a living being; and (3) the goal of a movement. It progressively lost these meanings to
the profit of the prepositions with or by, from and to. The first shift occurred around
1500, the second in the sixteenth century and the concurrence with to lasted until Early
Modern English (Lindkvist 1978). This evolution is represented in chart 4.

œt1

œt2 from

œt3 with

œt4 to

Chart 4. Evolution of AT

Œt2 has all the meanings of œt1 with the exception of the origin; œt3 has all the mean-
ings of œt2 but its landmark cannot be a living being; and œt410 has all the meanings of
œt3 with the exception of the goal. In contrast to œt, the French preposition à does not
leave the introduction of the goal of the target to another preposition.
A comparison between the case of en and sobre in Spanish and between in and on
in English might reveal two different modes of lexical creation. In Spanish, one may
use the preposition en for an object placed on a table, but if the object is placed on a
chest of drawers, sobre must be used instead of en. Indeed, using the latter preposition
 GENESIS OF SPATIAL TERMS 181

would imply a reference to the interior of the drawers, as a preferred option. To avoid
the confusion with the objects contained by the drawers, sobre must be chosen.

(4) El libro está en la mesa

(5) El libro está sobre la cómoda

El libro está en la cómoda (inside a drawer)11

Therefore a need for clarification pushes Spanish to work with two prepositions like
English. What is different between the two languages is not that Spanish has only
one preposition to describe CONTAINMENT and SUPPORT, but that English does
not allow one to use in when on is adequate. The diagrams in Figure 2 illustrate the
distribution of the prepositions in and on in English and of the prepositions en and
sobre in Spanish.

(A) (B)

in on en sobre

Figure 2. Distribution of prepositions in English and Spanish

To be accurate, the schema describing in and on requires an intersection since, in some

cases, speakers hesitate between these two prepositions to describe the same situation.
Whereas schema (A) is compatible with external lexical formation, in which in and on
are directly attached to support and containment, schema (B) is a case of internal lexical
formation at its first stage. This means that sobre has not reached the stage in which it
would prevent en being chosen when the conditions for the use of sobre are met. This
preposition is preferred to en only in the cases in which an ambiguity must be avoided.
In the case of in and inside, morphology shows that the formation of in is likely to
precede the formation of inside since it is much easier to imagine the addition of –side
to in than to build in from inside by truncation. Whereas in can be used for the interior
of closed containers (sentence 7a), for the interior of open containers (sentence 8a), for
the material of containers (sentence 9a) and for masses (sentence 10a), inside can only
be used in sentences (7b) and (9b):
(7) a. The jewels are in the box
b. The jewels are inside the box

(8) a. The wine is in the glass

b. *The wine is inside the glass
182 LANGUAGE, COGNITION AND SPACE

(9) a. The termites are in (the wood of) the cupboard

b. The termites are inside (the wood of ) the cupboard

(10) a. The fish is in the water

b. *The fish is inside the water

Therefore, inside may be considered as a hyponym of in. I am not aware of examples

in which inside must be used instead of in in order to avoid ambiguity, as was the case
for sobre and en in Spanish. As long as this need is not felt by the speakers, a split of in
and inside similar to the split of the category ‘grue’ in green and blue has not occurred.
Hyponymy, then, does not necessarily lead to separation. Therefore, IN1 does not cor-
respond to IN-2D and IN-3D except those cases in which INSIDE can be used, as it is
suggested in chart 2. I conclude this section with other discrepancies between chart 2
and the analysis proposed in chart 3.
According to the analysis of Levinson and Meira, IN-3D (containment) and IN-2D
(inclusion in a plane) appear simultaneously at the second level of abstraction. The first
notion implies the control of the target by the landmark whereas the second notion
localizes the target in a two-dimensional landmark. The interaction occurs between two
material entities in the former case while the landmark is a spatial entity in the latter
case. In chart 3, IN-3D corresponds to CONTAINMENT. The notion corresponding
to IN-2D should be in the LOC part of the chart and, indeed, an important function of
LOC3, conveyed by à in French is to locate a material entity (Jean) or a spatial entity
(Montmartre) in a spatial entity (Paris):

(11) Jean est à Paris

(12) Montmartre est à Paris

As a matter of fact, the contrast between a material landmark and a spatial landmark
might determine the difference between IN-2D and IN-3D better than the contrast
between two-dimensional and three-dimensional. Indeed, the dimensionality of a spatial
entity is a matter of conceptualization and the two-dimensional wood in sentence (13)
looks rather three-dimensional in sentence (14):

(13) The rabbits play in the wood

(14) The birds fly in the wood

Interestingly, English uses in to translate (11) and French uses en – coming from the
Latin preposition in – in front of feminine country names as well as in front of masculine
country names beginning with a vowel:

(15) John is in Paris

(16) Jean est en France
 GENESIS OF SPATIAL TERMS 183

Further hesitation between at and in to locate a target in geographic entities appears

in the development of English. Indeed, whereas in was used for this function in Old
English, œt introduces countries and large areas in Middle English and survives in Early
Modern English to disappear in the nineteenth century (Lindkvist 1978). In order to
explain these variations, I would like to claim that chart 3 captures only the prototypical
values of the basic spatial prepositions. According to my analysis of dans (‘in’), the first
function of this preposition is the representation of the relationship CONTAINER/
CONTENT (Vandeloise 1994, 2005). It accounts for the initial value of in in chart 3.
From this initial value, in develops different meanings that can be more or less close to
the prototypical value of other basic spatial preposition in the chart (Vandeloise 1995).
Thus, what Levinson and Meira call IN-2D might be a later development of IN-3D.
Whereas the landmark in IN-3D is a material entity with boundaries that allow physical
control of the target, the landmark of IN-2D is a spatial entity. Spatial entities may have
determinate boundaries – think of countries! – but they are virtual rather than material.
Therefore, like à in French, IN-2D orientates itself toward localization and may compete
with AT2. Even with a spatial landmark, however, the French preposition dans keeps
the memory of its first function. Compare sentences (17) and (18):

(17) ?Hans est dans Paris

(18) Les soldats sont dans Paris

Whereas sentence (17) looks odd, the use of dans in sentence (18) is perfect because
the idea of a conflict evoked by the soldiers makes control more salient.12
Two notions introduced in the analysis of Levinson and Meira – UNDER and OVER
– do not appear in chart 3. Numerous studies have been dedicated to over (Lakoff 1987,
Brugmann 1988, Dewell 1994, Tyler and Evans 2001, Deane 2005). In contrast to Tyler
and Evans, Dewell (1994) treats this preposition as a path preposition. This would be a
sufficient condition to ignore over in a chart devoted to static spatial prepositions. One
may also doubt whether this preposition belongs to basic prepositions since, besides
English, Levinson and Meira do not mention another language with the category OVER.
Brugman (1988) and Lakoff (1987) associate over to above and across. In fact, the two
pictures illustrating OVER in Levinson and Meira’s data might as well be described by
above. However, in the analysis of Levinson and Meira, the link of over with above is
ignored and OVER is considered as a notion that confines the scope of ON2 in languages
like English, in the same way as ATTACHMENT does for languages like Dutch.
Like OVER, UNDER has a particular status in chart 2 since it is introduced simul-
taneously with ON1. As with IN, UNDER can convey control between two material
entities when there is contact (sentence 19), or localize a target relative to the landmark
(sentence 20):

(19) The red book is under the yellow book

(20) The shoes are under the table
184 LANGUAGE, COGNITION AND SPACE

Under looks like a converse of on in sentence (19) since this sentence implies that the
yellow book is on the red book.13 However, as illustrated by sentence (20), the converse
relation between on and under is not as complete as the converse relation between
the projective prepositions in front and in back. It would be easy to integrate under in
chart 3. Its first meaning might be introduced below SUPPORT, in the same way as the
prepositions au-dessus and en dessous are introduced below VERTICAL SEPARATION.

CONTROL

CONTAINMENT SUPPORT
(in) (on/under)

Chart 5. Incorporating under

The meaning of under in sentence (20) might then be considered as an extension of

its meaning in sentence (19) (Vandeloise 1991, chapter 12), just as IN-2D may be an
extension of IN-3D. Compared to chart 2, this alternative presents the advantage of
justifying the simultaneous introduction of on and under by their common relationship
to SUPPORT. However, this might suggest too strong a connection between on and
under and I will ignore the notion UNDER in chart 3.

3 Three modes of development

The implicational scale of Berlin and Kay for the basic terms of colors describes the
order of appearance of these terms in the formation of languages. With the assump-
tion that languages evolve from little sets of words to their complete lexicon, one may
assume that a language with a system of seven basic color words is more evolved in
this domain than a language with a system of five words. When the new terms occur
through internal lexical formation by division, the development of languages can only
go from the top to the bottom, i.e. from the most general terms to the most specific
ones. If only internal lexical formation were involved in the creation of spatial terms,
the same conclusions might be drawn for the typology of Levinson and Meira in chart 2
and for the conceptual hierarchy proposed in chart 3. This means that Korean would be
a development of English, itself a development of Spanish. But then, in and on in English
should derive from a word conveying the same situations as en in Spanish, just as green
and blue are created by the split of the category ‘grue’. And kkita and nehta in Korean
would be created by internal lexical formation from a word with a larger distribution
corresponding to English in.14 If we do not have evidence in the history of English and
Korean for such a development, this may simply mean that the formation of spatial
terms is not parallel to the formation of basic color terms, and that there are different
modes of genesis of spatial terms. Indeed, besides internal lexical formation, external
lexical formation plays a role in their creation. In this case, in and on in English, as well
as kkita and nehta in Korean, do not have to be the result of the split of a larger category.
 GENESIS OF SPATIAL TERMS 185

They may have been created separately because the speakers of these languages attach
a communicative virtue to the categories represented by these words.
With external lexical formation, the first spatial terms can appear at any level of
generality in the hierarchy proposed in chart 3. Whereas Spanish might attach en directly
to control, English may attach in to containment and Korean can associate immediately
kkita to tight fit. The process of internal lexical formation proposed by MacLaury pro-
ceeds by division: a larger category ‘grue’ is replaced by two more specific categories
designated by green and blue. This type of formation, therefore, can only go from the
top to the bottom of the hierarchy. But, if some languages create words at a high degree
of specificity by external lexical formation, there may be a different type of internal
lexical formation going from the bottom of the hierarchy to the top. Besides internal
lexical formation by division, then, there might be a mode of internal lexical formation
by union. This mode of formation is internal because it relies on the existence of two
more specific words. In contrast to internal lexical formation by division, however,
internal lexical formation by union goes toward the top of the hierarchy. It can begin
from the bottom of the hierarchy, with the most specific terms, or in the middle with
intermediary notions. With these three modes of lexical formation, the developments
illustrated in chart 6 are logically possible in languages.

$ ĺĺ

% ĸĺ

& ĺĺ

Chart 6. A hierarchy of formation of spatial terms

In the case of schema (A), the creation of spatial terms begins with RELATIONS IN
SPACE, a concept that gathers LOC and CONTROL. Schema (B), beginning in the
middle of the hierarchy, is very reminiscent of the relationship between basic categories,
supercategories and subcategories proposed by Rosch (1973). One goes from basic cat-
egories to supercategories by abstraction and to subcategories by specification. Schema
(C) goes from the most concrete concepts to the most abstract.
Which of schemas (A), (B) and (C) is dominant in the creation of language? Schema
(A) is illustrated by the development of basic color terms proposed by MacLaury in chart
1, in which the number of words increases from the top to the bottom of the hierarchy.
The implicational scale of Levinson and Meira suggests a similar development for spatial
terms. One may also surmise that languages have fewer words at their beginning than
when they are fully developed. This parallelism pleads in favor of schema (A). Other
arguments, however, show that schema (C) has a dominant role in the creation of lan-
guages. Indeed, Lévy-Bruhl (1922) claims that ‘primitive’ thought is characterized both
by its concreteness and the absence of general concepts. For example, many Amerindian
languages do not have a general term for walking but they have many more specific
186 LANGUAGE, COGNITION AND SPACE

verbs that specify the direction, the trajectory or the manner of walking. According to
Merleau-Ponty (1945), Maoris have 3000 terms for colors, not because they distinguish
numerous colors, but because they do not recognize the same color when it belongs
to different objects and use different words for it. Concrete specific concepts, then,
might be at the origin of many words. As far as schema (B) is concerned, numerous
experiments in cognitive psychology by Rosch and her colleagues (1975) demonstrate
the preponderance of basic categories over subcategories and supercategories. If one
may recognize basic categories in the middle of chart 3, this might plead for schema
(B). CONTAINMENT and SUPPORT, then, should be more prototypical notions than
CONTROL and TIGHT FIT or ATTACHMENT. Experiments by Choi et al. (1999) cast
some doubt about the predominance of CONTAINMENT over TIGHT FIT. Indeed,
English infants demonstrate more interest in the latter relation than in the former. If
TIGHT FIT were universally dominant, Spanish children should begin their journey
in language by limiting the use of en to the most specific contexts before enlarging its
distribution to CONTROL. As far as attachment is concerned, Levinson and Meira
found that many languages consider it a central topological notion. An explanation
for this predominance might be that, with the exception of fruits attached to trees,
ATTACHMENT is mainly an artificial way of stabilizing the target. This is in contrast to
CONTAINMENT and SUPPORT that occur frequently in the nature. ATTACHMENT,
then, would contrast with all the natural spatial relationships.

4 Language acquisition

At the beginning of the nineteenth century, ‘primitive’ thought was often compared
to the thought of children (Lévy-Bruhl 1922). In this way, ontogeny, the acquisition of
one language by one child, would reproduce phylogeny, the creation of a language by a
civilization. However, there is an obvious difference between language creation and its
recreation by the child since, in contrast to the community that must begin a language
from scratch, the child is immediately confronted with a completely developed language.
Furthermore, whereas the creation of a language requires production, children first learn
a language through understanding and reproduction. In this section, I will first attempt
to understand how the acquisition of a language without the help of a pre-linguistic
conceptual system could occur. This eventuality appears very unlikely. Therefore, in the
second part of this section, I will evaluate the incidence of schemas (A), (B) and (C)
and of the pre-linguistic concepts in chart 3 on the acquisition of different languages.
An extreme form of determinism claims that no structured thought can exist
without language. Therefore, only language can help to learn language. However, the
first use of a word W by a child must be triggered by a situation in the world to which
he associates W. Since, by hypothesis, the concept corresponding to the word W does
not exist before W is acquired, its association with the situation must be referential
or indexical. At the time of anchorage, the knowledge of the word is, of course, very
tentative. Language can help to develop the full knowledge of the word in two ways.
First, when the word W is used for a new situation in which the child would not have
 GENESIS OF SPATIAL TERMS 187

used it, he knows that his language establishes a connection between this situation and
the other occasions on which he uses W. In contrast, when a different word is used for
a situation in which he would have used W, he realizes that his language is sensitive to a
difference that justifies the choice of another word. The use of W will be under-extended
as long as the child does not know all the relevant similarities and overextended as long
as he does not know all the relevant differences.
The strength of linguistic determinism depends a great deal on the nature of the
connections language reveals to the child. Indeed, if they are based on similarities
and differences recognizable in the extra-linguistic situations, language does not so
much create the concept associated to the word as it guides the child through an array
of differences and similarities available in the world. In this case, at each stage of the
development of his acquisition of a word, the child has expectations that correspond to
his knowledge of the word. Does the final stage corresponding to the complete acquisi-
tion of the word – if there is such a thing – have a special linguistic flavor that singles it
out from the preliminary stages? I would rather guess that there is a continuum going
from the anchorage situation to the final stage of knowledge. In this case, there may not
be a clear-cut distinction between the established ‘linguistic’ concept and its elaboration.
If determinism is rejected and pre-linguistic concepts15 are admitted, how can the
acquisition of words expressing containment in languages like Spanish, English and
Korean help us to understand what they are? Since these concepts are pre-linguistic, they
are independent of language and can be shared by the infants speaking each language.16
For example, a Spanish child could be receptive to the notion of TIGHT FIT (associated
to kkita in Korean) and a Korean child could be sensitive to CONTROL (associated
to en in Spanish). In this way, there might be a common set of pre-linguistic concepts
shared by all the children in the world. On the other hand, children might have different
pre-linguistic conceptual systems, even among children learning the same language.
For example, there might be concrete-minded Spanish boys ready to anchor en to
TIGHT FIT whereas other boys, more abstract-minded, would associate it directly to
CONTROL and others, in the middle, would associate en to CONTAINMENT. As a
result, these children should use different schemas in order to reach a complete knowl-
edge of en: concrete-minded boys should use schema (C), going from the concrete to
the abstract, whereas abstract-minded boys would get an almost immediate knowledge
of the distribution of the word. In this way, schemas (A), (B) and (C) constitute the
most economical ways of learning Spanish, English, and Korean respectively, since the
concept corresponding to the level of abstraction chosen by these languages would be
acquired directly. I do not have empirical data answering this question. They would be
very helpful to choose between the existence of a common universal set of pre-linguistic
concepts on the one hand, and the existence of individual variations in the acquisition of
spatial terms on the other hand. Spanish infants under-extending en to TIGHT FIT or to
CONTAINMENT, for example, would provide strong evidence for these pre-linguistic
concepts since these underextensions cannot be justified by their language. The same
thing would be true for English infants limiting the use of in to TIGHT FIT.
Schema (C), proceeding from the most specific to the most abstract concepts,
might be built entirely conceptually, without the help of language, by the child who
188 LANGUAGE, COGNITION AND SPACE

recognizes the commonalities between TIGHT FIT and LOOSE FIT, and afterwards
between CONTAINMENT and SUPPORT. In this way, a child going through this
process of generalization would have the three pre-linguistic concepts at his disposal
before he begins to acquire his language. It is very easy, however, to see how language
can contribute to the building of these concepts. Indeed, suppose that an English child
under-extends the meaning of the preposition in and restricts its use to the representa-
tion of TIGHT FIT. He will quickly realize that adults are also using the same word for
LOOSE FIT. Therefore, he will be inclined to look for similarities that he might otherwise
have overlooked. In this case, one might say that language is a necessary condition, if
not a sufficient one, for the constitution of concepts. A Spanish child who would under-
extend the meaning of the Spanish preposition en and associate it with TIGHT FIT or
CONTAINMENT would also receive plenty of warnings from adult language until he
extends the use of the preposition en to CONTROL, which embraces the whole extension
of the preposition in adult language. Korean children, in contrast, will not find in their
language any incentive to extend TIGHT FIT to CONTAINMENT or to CONTROL.
A Spanish child who underextends en will correct himself more easily than a Korean
child who overextends kkita since the former will receive positive evidence (each time
he hears en used in circumstances he was not using it), whereas the Korean child will
only receive negative data (when adults correct him if he uses kkita inappropriately).

5 Conclusion

The genesis of basic colors (Berlin and Kay 1968, MacLaury 1993) provides hints to
better understand the genesis of spatial terms. Two modes of internal lexical formation
inside the language system (by division and by union) have been opposed to external
lexical formation that attaches words directly to extra-linguistic notions of utmost
importance in the linguistic community.
Before presenting my views on the genesis of spatial terms, I have discussed the
analysis of Levinson and Meira (2003). They exclude projective prepositions from
their investigation because, according to the authors, these prepositions belong to a
different subsystem. The development of spatial terms begins with an all-encompassing
adposition AT covering all the relationships in space. In chart 2, the system enriches
itself through internal lexical formation by division. The new notions introduced are
mainly topological basic categories like ON/OVER (superadjacency with or without
contact), UNDER (subadjacency with or without contact), NEAR (proximity). IN-3D
(containment) and IN-2D (inclusion in a surface) are also notions proposed in the
analysis, even though I believe that containment is a dynamic notion rather than a
topological one. According to my proposition, the dichotomy between CONTROL
(a general dynamic notion) and LOC (a general topological notion of localization)
constitutes the first step in the genesis of spatial terms. As illustrated by the preposition
of Old English œt, this part of the system evolves essentially by internal lexical formation
by division. In contrast to Levinson and Meira, I have introduced the projective notions.
As far as the dynamic spatial system is concerned, different levels of specification may
 GENESIS OF SPATIAL TERMS 189

be observed in different languages. For example, the Spanish preposition en represents

a general notion of CONTROL whereas the English prepositions in and on convey
more specific notions of CONTAINMENT and SUPPORT. No historical data show
that this enrichment occurs by internal lexical formation by division, which means that
IN and ON might occur by external lexical formation. In this case, the comparison of
the different levels of abstraction cannot be done inside one and the same language but
requires a comparison between different languages.
As far as color terms are concerned, one may consider that languages with more
specific terms are a development of languages with more general terms according
to schema (A) in section 3. If language creation was proceeding by internal lexical
formation only, one might draw the same conclusion for spatial terms related to contain-
ment in chart 3. But external lexical formation may attach a word directly to different
levels of abstraction. Such is the case for natural kinds like dogs and birds. Nouns in
basic categories are considered more prototypical than nouns for supercategories and
subcategories and are acquired first. The creation of these words conforms to schema
(B): supercategories and subcategories develop from basic categories by abstraction
and specification respectively. Finally, according to Lévy-Bruhl, human thought at its
beginning evolves from the concrete to the abstract, according to schema (C). This
schema would give precedence to the most specific basic terms.
In the last section of this article, I investigate how the acquisition of language might
help to provide clues about the development of spatial terms. How do children adjust to
the level of abstraction of control terms in the language they are learning: general like
en in Spanish, intermediary like in and on in English or specific like kkita in Korean
and aan in Dutch? Any discrepancies between child and adult language, as well as
the adjustments children are making to reach a complete command of spatial control
terms, may be helpful to understand the genesis of language. Three extreme – and
much caricatured – avenues may be proposed. First, the universal view: before speaking,
all the children in the world first pay attention to the same concepts and, afterwards,
adjust to their language through schemas (A), (B) or (C). Second, the relativist view:
after a period of passive understanding, children are immediately tuned to the level of
abstraction that characterizes their language. And finally, the individualistic view: even in
a single language, different children make different hypotheses and reach the command
of control terms by different ways. It might be useful to keep the three possibilities in
mind when we analyze any data that might be relevant for the genesis of language.

Notes
1 According to MacLaury, the category of warm colors splits before the category of cool
colors and red appears in third position because the perceptual difference between red
and yellow is more conspicuous than the contrast between green and blue.
2 ‘Relevant sort’ might only have a specific sense if there was a consensus about the
central meanings of these prepositions, which is far from being be the case.
190 LANGUAGE, COGNITION AND SPACE

3 Topology here has not a mathematical meaning but refers to static common sense rela-
tionships in space, such as neighborhood and inclusion, as used in Piaget and Inhelder
(1956).
4 Levinson and Meira do not need to be concerned by verbs since they explicitly limit
their analysis to adpositions.
5 Maybe the authors consider that on appears simultaneously with its converse under.
However, in language acquisition, under is understood much later than on (Rohlfing
2003).
6 If the tablecloth covers the table entirely, its situation would be described in English by
the tablecloth is over the table rather than by the tablecloth is on the table.
7 For some spatial relationships, like the situation described by the sun is above the earth
or the airplane is over the house, proximity of the two material entities is not a necessary
condition. In these particular cases, however, accessibility may be obtained by the rays
in the case of sun or by bombs (or landing) in the case of the airplane.
8 Adpositions marking control may help to locate the target but they do it only indirectly.
A sentence like the wine is in the glass is used to indicate that the wine is available for
drinking – as opposed to the wine on the floor. French children are well aware that the
preposition dans conveys localization only indirectly when their answer to the question
‘Where is the King?’ is: Dans sa chemise (‘In his shirt’).
9 Kkita might also be considered as a specification of on when it represents a relation of
tight fit between the target and a horizontal landmark. However, these situations are
extremely rare since, except for magnetic objects, the pressure exerted by the target on
its support is not stronger than its weight. Two horizontal pieces of Lego fitting together
are an example of horizontal tight fit. However, in might be used in this case, in contrast
to on, preferred if one piece is simply put on the other, without adjustment.
10 In Early Modern English, oet had acquired the modern form at.
11 These sentences are adapted from Fortis (2004). Thanks to Ignasi Navarro-Ferrando for
comments on these examples.
12 As noted by an anonymous reader, the control here is exerted by the target (the soldiers)
rather than by the landmark (Paris).
13 In French, the phonetic similarity between sur and sous reinforces the parallelism
between the two spatial relations they convey.
14 Even though this hypothesis looks similar to the hypothesis concerning the common
origin of in and on in English, there is an important difference since in and on are
acquired approximately at the same time by children whereas kkita appears to be
learned earlier than nehta in Korean. These two words, then, do not have the same
status in acquisition.
15 Tye (2000: 176) speaks of ‘perceptual concepts’ that are ‘a matter of having a stored
memory representation that has been acquired through the use of sense organs and
available for retrieval, thereby enabling a range of discriminations to take place ‘
16 Society can introduce differences in the set of pre-linguistic concepts independently of
language. This is the case for societies that have no containers or societies that have only
round symmetrical objects.
 GENESIS OF SPATIAL TERMS 191

References
Berlin, B. and Kay, P. (1968) Basic Color Terms. Berkeley: The University of California
Press.
Bowerman, M. (1996) The origins of children’s spatial semantic categories. In
J. Gumperz and S. Levinson (eds) Rethinking Linguistic Relativity 145–176.
Cambridge: Cambridge University Press.
Brown. P. (1994) The INs and ONs of Tzeltal locative expressions: The description of
stative descriptions of location. Linguistics 32 (4/5): 743–790.
Brugman, C. (1988) The Story of over: Polysemy, Semantics and the Structure of the
Lexicon. New York: Garland Press.
Choi, S., Mc Donough, L., Bowerman, M. and Mandler, J. (1999) Early sensitiv-
ity to language specific spatial categories in English and Korean. Cognitive
Development 14: 241–268.
Deane, P. (2005) Multimodal spatial representation: On the semantic unity of over.
In Beate Hampe (ed.) From Perception to Meaning: Image Schema in Cognitive
Linguistics 235–285. Berlin: Mouton De Gruyter.
Dewell, R. (1994) Over again: Image-schema transformation in semantic analysis.
Cognitive Linguistics 5: 351–380.
Feist, M. (2004) Talking about space: A cross-linguistic perspective. In K.D. Forbus,
D. Gentner and T. Regier (eds) Proceedings of the Twenty-Sixth Annual Meeting of
the Cognitive Science Society. Mahwah, NJ: Lawrence Erlbaum.
Fortis, J-M. (2004) L’espace en linguistique cognitive: Problèmes en suspens. Histoire
Epistémologie Langage XXVI(1): 43–88.
Gunnarson, K.A. (1986) Loin de X, près de X et parallèlement à X. Syntagmes préposi-
tionnels, adjectivaux ou adverbiaux? Le français moderne 54(2): 1–23.
Lakoff, G. (1987) Women, Fire and Dangerous Things. Chicago: The University of
Chicago Press.
Levinson, S. and Meira, S. (2003) Natural concepts in the spatial topological domain
– adpositional meanings. Language 79(3): 485–516.
Lévy-Bruhl, L. (1922) La mentalité primitive. Paris: Presses Universitaires de France.
Lindkvist, K. (1978) AT vs. ON, IN, BY. Stockolm: Almqvist and Wiksell
International.
MacLaury, R. (1993) Social and cognitive motivation of change: Measuring variabil-
ity in color semantics. Language 69(3).
Merleau-Ponty (1945) Phénoménologie de la perception. Paris: Gallimard.
Piaget, J. and Inhelder, B. (1956) The Child’s Conception of Space. London: Routledge
and Kegan Paul.
Rohlfing, K. (2003) UNDERstanding. How Infants Acquire the Meaning of Under and
other Spatial Relational Terms. Bielefield University: PhD dissertation.
Rosch, E. (1973) Natural categories. Cognitive Psychology 4: 328–350.
Rosch, E. and Mervis, C. (1975) Family resemblances: Study in the internal structure
of categories. Cognitive Psychology 93: 10–20.
Saussure, F. (1916) Cours de linguistique générale. Paris: Payot.
Tye, M. (2000) Consciousness, Color and Content. Cambridge, MA: The MIT Press.
192 LANGUAGE, COGNITION AND SPACE

Tyler, A. and Evans, V. (2001) Reconsidering prepositional polysemy networks: The

case of over. Language 77: 724–65.
Vandeloise, C. (1991) Spatial Prepositions: A Case Study in French. Chicago: The
University of Chicago Press.
Vandeloise, C. (1994) Methodology and analyses of the preposition. Cognitive
Linguistics 5(5): 157–185.
Vandeloise, C. (1995) De la matière à l’espace. Cahiers de grammaire 20: 123–145.
Vandeloise, C. (2003) Containment, support and linguistic relativity. In H. Cuyckens,
R. Dirven and R. Taylor (eds) Cognitive Approaches to Lexical Semantics 393–
425. Berlin: Mouton De Gruyter.
Vandeloise, C. (2005) Force and function in the acquisition of the preposition in. In
L. Carlson and E. van der Zee (eds) Functional Features in Language and Space
219–231. Oxford: Oxford University Press.
Wierzbicka, A. (1990) The meaning of colour terms: Semantics, culture and cogni-
tion. Cognitive Linguistics 1: 11–36.
8 Forceful prepositions 1
Joost Zwarts

Introduction

As the title suggests, the focus of this paper is on prepositions with a force-dynamic
aspect, as in the following example sentence:

(1) Alex ran into a tree

This sentence does not mean that Alex ended up inside a tree, but that she came forcefully
into contact with the tree. There is a spatial component in this sentence (Alex followed
a path that ended where the tree was), but there is also a force-dynamic component
(the tree blocked her movement).
This use of into brings together two conceptual domains that are fundamental in
the semantics of natural language: the spatial domain and the force-dynamic domain,
each of which comes with its own intricate system of concepts and relations. The
spatial domain is primarily concerned with location, movement and direction, the
force-dynamic domain with causation, control and interaction. The basic thematic
roles of the spatial domain are Figure and Ground (Talmy 1983), Theme, Goal, and
Source (Gruber 1976), and Trajector and Landmark (Langacker 1987), while the force-
dynamic domain has Agent and Patient (Jackendoff 1987, Dowty 1991) or Agonist and
Antagonist (Talmy 1985).
The interaction of these two domains in the verbal domain has been relatively
well-studied (for example in Jackendoff 1987, Croft 1991, 2009, and others), but this is
different with the prepositions, that seem to be the spatial words par excellence. However,
there is a growing awareness in the study of prepositions and spatial language that
force-dynamic notions do play an important role (Vandeloise 1991, Bowerman and Choi
2001, Coventry and Garrod 2004, Carlson and Van der Zee 2005). It is becoming clear
that geometric notions alone do not suffice to capture the meaning of even very basic
prepositions like in and on, let alone an obviously force-dynamic preposition like against.
However, what is not yet clear is how the role of force-dynamics can be transparently
and adequately modeled in representations of the meaning of prepositions. This paper
makes some specific proposals about how to do this.
In section 1 I will single out a few important phenomena that concern prepositions,
most of which are well-known from the literature, that require reference to forces in one
way or another. I will then argue in section 2 that the general semantic mechanics that
underlies reference to forces can best be captured in terms of vectors (O’Keefe 1996,
Zwarts 1997, Zwarts and Winter 2000). These force vectors will allow an interface between

193
194 LANGUAGE, COGNITION AND SPACE

the force-dynamic part and the geometric part of the semantics of prepositions along lines
worked out in section 3. In a concluding section 4 I will sketch the potentials of the model
in understanding cross-linguistic variation in the domain of containment and support.

1 Forced beyond geometry

In order to illustrate the need for force-dynamics in the semantics of prepositions, this
section will briefly discuss some relevant aspects of the interpretation of the English
prepositions against and in and on and the Dutch prepositions op and aan.

Against

Against is the clearest example of a preposition that is not purely geometric:

(2) Alex bumped against the wall

Dictionaries characterize the meaning of against in such terms as ‘collision’, ‘impact’,

and ‘support’. It typically combines with verbs like crash, lean, push, bang, and rest, verbs
that all involve forces, either dynamically (3a) or statically (3b):

(3) (a) There was a loud bang against the door

(b) The rifle rested against the tree

Against is a relation that always implies physical contact between the Figure and the
Ground. This contact is usually lateral, i.e. from the side, involving horizontal force
exertion. We can see this clearly when we contrast against with on:

(4) (a) Alex leaned against the table

(b) Alex leaned on the table

(4a) refers to a horizontal force, requiring contact with the side of the table, but (4b) to
a downward force, involving the tabletop. Notice finally that the result of the force is
left unspecified when against is used:

(5) (a) Alex pushed against the car

(b) Alex pushed the car (to the garage)

Sentence (5a) does not tell us what the ‘reaction’ of the car is, whether it moves as a
result of the pushing or stays put. It simply leaves the result of Alex’ force open. Notice
the contrast with (5b) in this respect, a construction that allows directional PPs like to
the garage, apparently because the transitive use of push implies that pushing results in
motion of the direct object.
 FORCEFUL PREPOSITIONS 195

In and on

Although probably two of the most common prepositions in English, in and on have
also proved to be the most difficult ones to define in geometric terms (Herkovits 1986).
Intuitively, the geometric condition for in is ‘inclusion’ and the geometric for on ‘con-
tiguity’. But, as Vandeloise (1991) and Coventry and Garrod (2004) have argued, these
conditions are not always necessary for the proper use of in and on, respectively, and
they are not always sufficient either. Here are two well-known examples:

Figure 1 Figure 1a Figure 1b

In Figure 1a the black marble is not included in the (interior of the) bowl, but we still
would describe this situation with sentence (6a) below. In Figure 1b there is contiguity
of the ball with the table, but nevertheless, the description in (6b) is not felicitous.

(6) (a) The black marble is in the bowl

(b) The ball is on the table

So, there are relations without inclusion that we call in, as in (6a), and there are relations
with contiguity that we don’t call on, (6b). These observations have led Vandeloise
(1991) and Coventry and Garrod (2004) to propose that force-dynamic conditions
are needed instead of, or in addition to, the geometric conditions of containment and
contiguity. Even though the black marble in Figure 1a is not included in the bowl, its
position is in some sense controlled by the bowl through the grey marbles. There is a
force-dynamic relation of containment. The position of the ball underneath the table in
Figure 1b is not controlled by the table in the way that would be necessary for on to be
apply, namely by support, a force relation that requires the ball to be on the opposite,
upper side of the tabletop.
196 LANGUAGE, COGNITION AND SPACE

Aan and op

For the third example of the role of force-dynamics, we need to turn to Dutch. The
English preposition on corresponds to two distinct Dutch words, op and aan (Bowerman
and Choi 2001, Beliën 2002):

(7) (a) a cup on the table een kopje op de tafel ‘support’

(b) a bandaid on a leg een pleister op een been ‘adhesion’
(c) a picture on the wall een schilderij aan de muur ‘attachment’
(d) a handle on the door een handvat aan de deur ‘attachment’
(e) a leaf on a twig een blaadje aan een tak ‘attachment’

As Bowerman and Choi (2001) show, Dutch uses aan for spatial relations that involve
attachment (7c-e) while op is used for relations of support (7a) and adhesion (7b). So the
distinction between aan and op is again not purely geometric, but also force-dynamic,
given that relations of attachment, support and adhesion presuppose that the related
objects exert forces on each other.

Extended location

Herskovits (1986) noted that the applicability of on can be extended in an interesting

way, crucially involving force-dynamics again. The English sentence (8a) and its Dutch
translation (8a’) describe the situation in Figure 2a below, even though the cup is really
standing on a book that is lying on the table.

(8) (a) The cup is standing on the table

(a’) Het kopje staat op de tafel
(b) De lamp hangt aan het plafond
The lamp hangs on the ceiling
‘The lamp is hanging from the ceiling’

Figure 2 Figure 2a Figure 2b

 FORCEFUL PREPOSITIONS 197

In the same way, Figure 2b fits the Dutch description in (8b), although the lamp is not
directly hanging from the ceiling, but connected to it by a cable. What we see then is that
these prepositions usually require direct contact (‘contiguity’ or ‘attachment’) between
Figure and Ground, but it is possible to make the relation indirect, by the intervention
of a third object.
So, we have seen a range of force-related notions that seem to play a role in the
semantics of spatial prepositions: ‘impact’, ‘control’, ‘containment’, ‘support’, ‘adhesion’,
‘attachment’. What is the force-dynamic system behind these notions? And, given that
there is interaction with purely spatial concepts (like vertical and horizontal direction,
inclusion and contiguity), how does this force-dynamic system interface with the spatial
system? In other words: what is the geometry of forces?

2 A geometry of forces
Vectors

Since the notion of vector is going to play an important role in our analysis of the geom-
etry of forces, we will start with a brief and informal overview of some core concepts.
Vectors are a powerful tool to analyze geometrical concepts. Essentially, a vector v is
a directed line segment, an arrow, as illustrated in the diagrams in Figure 3. There are
different ways to represent a vector in linear algebra, but for our purposes it is sufficient
to understand it at this basic level. Free vectors have a length and a direction only,
located vectors have a particular starting point. The zero vector has no length and no
direction, but it can have a location.

w
w
v+w
−w

v
v 2v

Figure 3 Figure 3a Figure 3b

In the algebra of vectors two vectors v and w can be added up to form the vector sum
v+w. Figure 3a illustrates how this vector sum forms the diagonal of the parallelogram
of which v and w are the sides. Scalar multiplication is another operation, in which a
vector v is multiplied by a real number s, to form the scalar multiple sv, which is s times
as long as v (see Figure 3c). Each non-zero vector v has an inverse -v of the same length,
but pointing in the opposite direction. With this background, we can now take a closer
look at vectors in the force-dynamic domain.
198 LANGUAGE, COGNITION AND SPACE

Force vectors

The literature about force-dynamics is not extensive, but it would still go too far for the
purposes of this paper to give here even a short overview of what has been written in
Talmy (1985) and works inspired by it, like Johnson (1987), Langacker (1991), Croft
(1991, 2009), Jackendoff (1993), Wolff and Zettergren (2002). I will restrict myself to
extracting from the literature some useful ingredients for a rudimentary model of forces.

(i) The first ingredient is that forces have vector properties. Even though this
is not made explicit by all the authors mentioned above, forces have two
parameters: they have a magnitude (they can be smaller or bigger) and they
have a direction (they point in a particular, spatial, direction). These two
parameters define a vector. The third parameter, less relevant here, is the
location of the force, i.e. the physical point where the force is exerted.

(ii) Usually, a force is exerted by one object, the Agent, on another object, the
Patient. The Agent is what Talmy (1985) calls the Antagonist, the Patient is
the Agonist. Talmy’s terms have not found general currency and I will there-
fore use the more common terms Agent and Patient here, even though this
occasionally leads to somewhat awkward results, as we will see at the end of
the next section.

(iii) The Patient may also have its own force vector. This vector represents the
inherent tendency of the Patient to move in a particular direction. The
tendency of material objects to go downwards, because of gravitation, is an
example of such an inherent force vector (even though, strictly speaking, the
earth is the Agent here).

(iv) Because of the interaction between the forces of Agent and Patient, there is a
resultant vector that determines the result of this interaction. This resultant
vector is simply the sum of the Agent’s and the Patient’s vector (according to
the parallelogram rule) and this sum can be zero, when the forces of Agent
and Patient are equal but opposite.

All of these ingredients can be illustrated with a concrete example, based on the experi-
ment and analysis of Wolff and Zettergren (2002). Consider the example:

(9) The fan prevented the boat from hitting the cone

In their experiment, subjects were asked to judge whether sentences like these applied
to short and simple animations in which different kinds of objects were seen to exert
forces on a moving boat. Wolff and Zettergren found that the conditions for using
causative verbs like prevent could be analyzed in terms of the vector force interaction
of the objects involved. A situation that falls under sentence (9) might look as follows:
 FORCEFUL PREPOSITIONS 199

fan cone

fP
fR
fA

Figure 4

In this picture, fA is the force of the Agent, the fan, blowing against the Patient, the
boat. The boat has its own force tendency fP, that is directed towards the cone. In this
example, the Patient’s force vector is determined by the engine and the rudder of the
boat. When we add up the two vectors we get the resultant vector fR = fA + fP that tells
us where the boat is heading, as a result of the combination of the two forces. All of this
is simple high-school physics, but it allows Wolff and Zettergren to isolate the directional
parameters that determine how people actually apply causative verbs to dynamic scenes:
the directions of fA, fP and fA+fP with respect to a target T.
In the model of Wolff and Zettergren, the relative magnitudes of these force vectors
are essential for understanding how people label particular situations. A stronger force
vector fA results in a stronger sum fA+fP, which will then bring the Patient far enough
away from the target to judge the situation as an instance of prevent. Notice that the
absolute lengths of the force vectors in the spatial diagrams have no direct linguistic
significance. Multiplying all the force vectors in a situation by the same scalar would
represent the same force-dynamic concept. What matters for the understanding of
verbs like prevent are ultimately the relative magnitudes and absolute directions of the
three vectors.
For prevent to be applied to a force-dynamic situation, it is necessary that fP is
directed towards the target T, while fA and fA+fP are not. The verbs cause and enable are
different, in that the result fA+fP is directed towards the target. Enable requires that the
vectors of both Patient and Agent point towards the Target, with cause they are opposite.
See Wolff and Zettergren (2002) for further explanation and evidence concerning this
vector-based force-dynamics of causative verbs. I will turn now to a class of verbs that
refer to forces in a more direct and more spatial way.

Forceful verbs

The first two verbs that I would like to consider are push and pull. Obviously, these two
verbs are opposites, more specifically directional opposites (Cruse 1986):

(10) (a) Alex pushed the pram

(b) Alex pulled the pram

But what is it exactly about their meanings that makes them opposite? It is not the
directions of motion that are opposite, because Alex can push or pull the pram without
200 LANGUAGE, COGNITION AND SPACE

the pram actually moving. In this respect, push and pull are different from opposite
motion verbs like enter and leave or come and go that have opposite spatial trajectories.
The opposition of push and pull is also different from the opposition between cause and
prevent seen in the following examples:

(11) (a) The fan caused the boat to hit the cone
(b) The fan prevented the boat from hitting the cone

where the results are in opposition (hitting the cone vs. not hitting the cone). The
opposition between push and pull lies purely in the opposite directions of the force
vectors involved, relative to the Agent. With push the force vector is pointing away from
the Agent, with pull it is pointing in the direction of the Agent. This is schematically
indicated in the following two figures:

push pull

Figure 5 Figure 5a Figure 5b

The vector is located at that point of the Patient where the Agent exerts its force and
its length represents the magnitude of the force. If there are no other forces interacting
with the pushing or pulling force, the Patient will move in the direction of the force
vector, so either away from the Agent in Figure 5a or towards the Agent in Figure 5b.
The force relation between Agent and Patient is closely related to a purely locative
relation between them. With pushing, the Agent is behind the Patient, with pulling
it is in front of the Patient. We can already see here how force-dynamic and spatial
notions interface in a way that is crucially based on direction and that requires forces
to have spatial direction.
What is the role of the length of the force vectors in Figure 5? As I said above, the
particular scale with which we represent force vectors in spatial diagrams is arbitrary.
However, the magnitude of forces does play a role, in two ways. First, verbs like push
and pull can be modified by an adverb like hard, which suggests that the length of a force
vector has linguistic relevance, although in a non-quantitative way, of course. Second,
on a more conceptual level, we could imagine that there are two people pulling equally
hard on opposite sides. In that case, we need to compare the magnitudes of forces to
conceptualize and describe this situation as one of balance.
Because I am mainly interested here in the directions of the force vectors, relative
to Agent and Patient, and not so much in their location and length, I will use a simpler
 FORCEFUL PREPOSITIONS 201

and much more schematic graphical representation, that abstracts away from the other
two parameters:

(12) push: Agent –-> Patient

pull: Agent <–- Patient

The arrows in (12) represent the spatial directions of the force vector, either pointing
from Agent to Patient, or from Patient to Agent.
Let me make a bit more precise how this could be represented in a formal vector
model. Let us assume that the spatial relation between Agent and Patient is represented
by a spatial vector vPA pointing from the Patient to the Agent (connecting their centers
of gravity, for instance). This vector vPA then gives us the spatial frame with respect
to which we can represent a force vector fA, as indicated in Figure 6a and 6b. What
push and pull express, is how fA is aligned with respect to vector vPA. vPA and fA are
opposite for push, they point in the same direction for pull. This is what (12) intends to
represent in an informal way.

push pull
fA
fA

vPA vPA

Figure 6 Figure 6a Figure 6b

Another pair of opposite force verbs is squeeze and stretch, that are very close to push
and pull. Squeeze can be defined as ‘press from opposite sides’, while stretch is ‘pull in
opposite directions:

squeeze stretch

Figure 7 Figure 7a Figure 7b

Again, there is a close relation with basic spatial notions: the forces of squeeze have
an inward direction with respect to the Patient and the forces of stretch an outward
direction. If there is a resulting change, it is a change of shape or volume, a shrinking or
202 LANGUAGE, COGNITION AND SPACE

expanding. Here also, I will use a more schematic representation of the force-dynamic
relation between Agent and Patient:

(13) squeeze: Agent –> Patient <– Agent

stretch: Agent <– Patient –> Agent

The third and last pair of forceful verbs to be discussed here is lean – hang. Both verbs
can refer to a downward force exerted by the subject, as in the following examples:

(14) (a) Alex was leaning on the table with his elbows
(b) There was a light bulb hanging from the ceiling

The distinction lies in the relative position of the Agent and the Patient. In (14a) the
Agent (Alex, or rather, his elbows) is above the Patient (the table), in (14b) the Agent
(the light bulb) is below the Patient (the ceiling). In one sense, leaning and hanging are
a bit like pushing and pulling. Leaning is like pushing from above and hanging is like
pulling from below. But there are two important differences. The first difference is that
the forces don’t come from within the Agent, but are the result of gravitation. Alex does
not have to do something to the table when he is leaning on it. The second difference is
that the force exerted by the Agent is counterbalanced by an equal but opposite force
of the Patient (indicated by the grey arrow), creating a static situation of balance, as
illustrated in the following two figures:

lean hang
Agent Patient

Patient Agent

Figure 8 Figure 8a Figure 8b

We can see the configuration of Figure 8a as a representation of support: the Patient is

supporting the Agent. Figure 8b, on the other hand, captures an important aspect of
the notion of attachment: the Agent is attached to the Patient.
It is in this situation that the use of the terms Agent and Patient becomes somewhat
awkward. From the perspective of the theory of thematic roles, we would not usually
call the subject of lean or hang an Agent and the table or the ceiling a Patient, because
we cannot say that the subject is doing something to the object of the prepositions.
Talmy’s term Agonist and Antagonist are not appropriate either. I will therefore use
a slightly different representation for situations of leaning and hanging, respectively:
 FORCEFUL PREPOSITIONS 203

lean hang
Figure Ground

Ground Figure

Figure 9 Figure 9a Figure 9b

The objects are labeled Figure and Ground here. The underlining of Figure indicates
that it is this participant that exerts the primary downward force to which the Ground is
reacting. What is not made explicit in the representation is that gravitation is responsible
for the Figure’s force.

Two kinds of arrows

The arrows in the representation that I proposed in the previous section should not be
confused with the arrows that are found in Langacker (1991) and Croft (1991, 2009).
There an arrow is used to indicate the direction in which energy is transmitted from
one object to another. The direction of the arrow is non-spatial and non-vectorial, and
it is always pointing from the Agent to the Patient, or from a more agentive to a less
agentive participant in a situation, e.g. from an Agent to an Instrument. In fact, notions
like Agent, Patient and Instrument can be more or less defined from their position in
a chain of causal relations:

(15) X –-> Y –-> Z

In such a chain, X will be the Agent, Y the Instrument and Z the Patient. In other words,
the arrow is thematic, representing the roles that objects play in a force relation.
In the representation that is used here, and also in Johnson (1987) and Wolff and
Zettergren (2002), the arrow represents the spatial direction of the force with respect
to given objects and dimensions. (15) then means that there is a force working away
from X and towards Y and a force working away from Y towards Z. It does not specify
the origins of these forces: this is where we need to label objects as Agent or Patient, or
underline them to indicate their force-dynamic primacy.
Both representations are justified, but for different reasons and for different pur-
poses. The first kind of arrow is useful for representing the thematic side of causal
relations, particularly for analyzing aspectual and argument structure, as argued
for in Croft’s work. The second kind of arrow is needed for the spatial side of causal
relations and is indispensable for understanding verbs with a directional component,
as we saw in this section, but also for force-dynamic prepositions, as we will see in
the next section.
204 LANGUAGE, COGNITION AND SPACE

3 Prepositional forces
Verbs and prepositions in Dutch

Verbal forces and prepositional forces interact. One area where we can see this clearly is
in some relevant verb preposition patterns in Dutch (Beliën 2002). While trekken ‘pull’
is used with aan, as shown in (16a) and (16b), the opposites duwen ‘push’ or drukken
‘press’ are used with op or tegen, (16a’) and (16b’):

(16) (a) aan de wagen trekken (a’) tegen de wagen duwen

on the car pull against the car push
‘pull the car’ ‘push the car’
(b) aan de bel trekken (b’) op de bel drukken
on the bell pull on the bell press
‘pull the bell’ ‘press the bell’

The choice between op and tegen is subtle, depending on the direction and the granular-
ity of the force. While (16a’) is used for a horizontal force exertion, (17a) below is used
for a force that comes from above. Op in (16b’) is the normal preposition to use when
a bell is pressed with a finger, but tegen is found, as in (17b) when something bigger
exerts a force on the bell, in a non-canonical way:

(17) (a) op de wagen duwen

on the car push
‘push on the car’
(b) tegen de bel drukken
against the bell press
‘press against the bell’

Hangen ‘hang’ and leunen ‘lean’ also correlate with particular prepositions:

(18) (a) aan de wagen hangen (a’) op/tegen de wagen leunen

on the car hang on/against the car lean
‘hang on the car’ ‘lean on/against the car’
(b) aan de bel hangen (b’) op/tegen de bel leunen
on the bell hang on/against the bell hang
‘hang on the bell’ ‘lean against the bell’

Hangen clearly goes with aan, (18a) and (18b), while leunen goes with op and tegen, (18a’)
and (18b’). However, hangen is also possible with op and tegen. Notice the contrasts in
the following examples:
 FORCEFUL PREPOSITIONS 205

(19) (a) Het gordijn hangt aan het plafond

The curtain hangs on the ceiling
‘The curtain is hanging from the ceiling’
(b) Het gordijn hangt op de grond
The curtain hangs on the ground
‘The curtain is hanging on the ground’
(c) Het gordijn hangt tegen het raam
The curtain hangs against the window
‘The curtain is hanging against the window’

The curtain is suspended from the ceiling, and aan is used in (19a) to describe this
relation. However, to describe the situation in which the curtain touches the ground
at the lower end op is used in (19b) and its contact with the window in the vertical
direction is indicated by tegen in (19c). In the remainder of this paper I will ignore the
use in (19b) and (19c).
In order to make sense of the patterns of (16) and (18), the prepositions aan, tegen
and op need to involve a force relation between the Figure and the Ground. The basic
idea is that aan ‘on’ is like pulling and hanging: a relation in which the Figure is at the
same time a kind of Agent, exerting a force on the Ground that is directed towards itself.
I will represent this as follows:

(20) aan: Figure <–- Ground

What characterizes aan is that the force vector is pointing from the Ground towards the
Figure. The Figure is underlined to indicate the division of agentivity in this relation:
it is the Figure that has an intrinsic tendency to move. Tegen ‘against’ and op ‘on’ are
the opposite of aan, in the sense that the force points away from the Figure towards
the Ground:

(21) tegen: Figure –-> Ground

op: Figure –-> Ground

In this respect, tegen and op are like pushing and leaning. The directional nature of
forces allows us to capture the distinction between aan on the one hand from op and
tegen on the other hand, but it also explains why prepositions cooccur with push and
pull verbs in the way they do.
Interestingly, the directional nature of forces has a direct reflex in English in the
the use of the directional preposition from with the verb hang:

(22) The lamp was hanging from the ceiling

The from that usually designates a path of motion away from the Ground is used here
for a force vector pointing away from the Ground. It is difficult to account for this use
if we don’t allow forces to have spatial directions.
206 LANGUAGE, COGNITION AND SPACE

More properties of support and attachment

What we have seen in the previous section is just the basic core of the force-dynamics
of contact prepositions like on in English and and op and aan in Dutch. There are a
number of other observations to make about these prepositions.
The first effect is the contact effect: the Figure and the Ground are in contact or
spatially contiguous. But note that this is not a spatial condition that is separate from
their force-dynamic properties. As we noted already with forceful verbs, spatial contact is
necessary for force-dynamic interaction. The Figure and Ground have to touch to allow
the configurations in (20) and (21) to obtain in the first place. So, the force-dynamic
and spatial components of the relations expressed by prepositions like on and against
are closely tied together.
The second effect, which we already described in section 1, is the chaining effect,
a way of extending the contiguity between two objects. The force interaction between
objects does not need to be direct, but it can be mediated by a third object. In our
schematic representation, we can represent this for op and aan (support and attachment)
as follows:

(23) op: Figure –-> X –-> Ground (support)

aan: Figure <–- X <–- Ground (attachment)

With op the Figure has a pushing relation with X and X with the Ground, with aan
the Figure is pulling X and X is pulling the Ground. The X can only fulfil its role if it is
literally between Figure and Ground, so if it is also a spatial intermediary, which is also
what we see in the situations from section 1, repeated here:

(24) (a) Het kopje staat op de tafel

The cup stands on the table
‘The cup is standing on the table’
(b) De lamp hangt aan het plafond
The lamps hangs on the ceiling
‘The lamp is hanging from the ceiling’

Figure 10 Figure 10a Figure 10b

 FORCEFUL PREPOSITIONS 207

The book in Figure 10a is between the table and the cup, just as the cable in Figure 10b
is between the ceiling and the lamp.
There is a third effect that usually occurs with aan and op. We can call it a default
effect, because it concerns the prototypical use of these prepositions. Again, we need to
refer to the spatial direction of forces to account for this effect. Unless otherwise specified
by the context or the sentence, we assume that aan (attachment) applies in a situation
in which the force vector is downward, because of gravitation. Aan is not just ‘pulling’,
it is downward pulling, i.e. ‘hanging’. This is especially the case if the sentence does not
have an explicit Agent. Also with op (support) the default is downward, as a result of
the gravitational pull. So, this is what we get in prototypical situations:

op aan
Figure Ground

Ground Figure

Figure 11 Figure 11a Figure 11b

This also implies that in the prototypical case, aan (attachment) implies ‘under’, while
op (support) implies ‘above’. These spatial relations follow again from the force-dynamic
specifications. However, it is not difficult to find situations in which the force relations
hold in a different direction, especially with on/op:

(25) (a) The fly is sitting on the wall

(a’) De vlieg zit op de muur
(b) The fly is sitting on the ceiling
(b’) De vlieg zit op het plafond

Finally, with aan and op, we get what we might call stative effects: situations in which
the force that the Figure exerts on the Ground is counterbalanced by an equal but
oppositely directed force exerted by the Ground. We normally interpret the sentences
in (18) as referring to situations of stasis, similar to what we saw with lean and hang:

op aan
Figure Ground

Ground Figure

Figure 12 Figure 12a Figure 12b

208 LANGUAGE, COGNITION AND SPACE

In general, in such a situation of stasis, the force vectors fFigure and fGround are
opposite and of equal length, i.e. fGround = -fFigure, or, in other words: fGround +
fFigure = 0, where 0 is the zero force vector.2
Stasis is not necessary, however. There can also be situations with on, op and aan in
which the counterforce is non-existent or such that no balance results:

(26) (a) Alex trok aan de wagen

Alex pulled on the car
‘Alex pulled the car’
(b’) Alex drukte op de bel
Alex pressed on the bell
‘Alex pressed the bell’

These Dutch sentences don’t specify what happens with the car and the bell. This depends
on particulars of the situation and properties of these objects.

Containment as a force-relation

We have finally come now to the most common and at the same time most complicated
preposition of Dutch and English: in. Vandeloise (1991) and others have argued that
the semantics of this preposition should be understood in terms of containment. Given
what we know now about force vectors, how can we capture containment in these terms,
such that the phenomena in section 1 are accounted for?
The idea is to take our inspiration again from what we see with verbs. We take in
to share important force-dynamic characteristics with squeeze. We have proposed in
section 2 to treat squeeze as a configuration in which there is concavity of forces: the
Patient is between (parts of) the Agent and the Agent’s forces are pointing towards
the Patient. I propose to represent in in a similar way, but since we are talking about
prepositional relations, I will use Figure and Ground:

(27) in: Ground –> Figure <– Ground

This is like a minimal configuration, which says that the Ground exerts forces on the
Figure from at least two opposite sides. Of course, the Ground might enclose the Figure
on all sides (and maybe this is even true for typical containment), but for the time being
I will assume that containment minimally requires what we see in (27). Notice that a
kind of spatial inclusion follows from this force-dynamic configuration. The forces of the
Ground can only come from different sides if the Ground somehow spatially includes
the Figure. Just as with on, aan and op, we see an intimate connection between forces
and locations, made possible by the way force vectors are embedded in space.
Obviously, there are important differences between squeeze and in. The verb squeeze
involves active and dynamic exertion of forces from at least two opposite sides, involv-
 FORCEFUL PREPOSITIONS 209

ing close contact, typically by an animate Agent. The preposition in involves a passive
and stative configuration of forces, not necessarily involving contact, typically by an
inanimate Ground. I believe that many of these differences correlate with the fact that
squeeze is a verb, while in is a preposition. Nevertheless, the two words both take part
in an abstract force-dynamic schema.
The configuration in (27) gives a basic condition for containment. What we see are
only two parts of the Ground, on opposite sides of the Figure. In a sense, (27) gives us
a one-dimensional cross-section of a two-dimensional situation in which the Ground
is a ring around the Figure or of a three-dimensional situation in which the Ground is
all around the Figure.
Even though (27) is very rudimentary, it does give us a way to capture what goes
on in the following two situations, both describable by the black marble in the bowl:

Figure 13 Figure 13a Figure 13b

Figure 13a is the simple situation in which there are two forces of the Ground pointing
from opposite sides towards the Figure, as in (27). However, in Figure 13b, there is
chaining:

(28) in: Figure

         
Ground –> X <– Ground

There are force vectors pointing from the Ground to X, but there is also a force vector
connecting X to the Figure. The force-dynamics of containment by the Ground is
transmitted here through an object X to the Figure. What is interesting here is that the
chaining is not homogeneous. The force-dynamic relation between the black marble
in Figure 13b (the Figure in (28)) and the other marbles (X in (28)) is not itself a
relation of containment, but rather one of support, it seems. But because the grey
marbles contained in the bowl support the black marble this marble is also indirectly
contained in the bowl.
This is only one simple example and it is not clear what will happen with this
primitive model of prepositional force-dynamics when we confront it with the diversity
of uses of topological prepositions like in and on. Nevertheless, as semanticists we
should go beyond simple descriptive labels like ‘containment’ and ‘support’ and look
for the system behind these relations. Modeling such a system will allow us to generate
210 LANGUAGE, COGNITION AND SPACE

testable hypotheses about the role that containment and support play in the semantics
of prepositions. We would predict, for example, that in can also be used in a situation
where Ground and Figure are related through attachment:

(29) in: Ground –> X <– Ground

       
Figure

Whether this is the case remains to be seen, but it illustrates an important point. With
a general unanalyzed notion of containment it remains unclear what is possible and
impossible. A model of force-dynamic relations in which we can manipulate parameters
is more adequate from a semantic point of view.

4 Conclusion

In this paper I have shown that verbs and prepositions are based on the same ‘geometry’
of forces. The notion of geometry can be taken quite literally, because forces are repre-
sented as vectors with a direction in space. This is essential for providing the interface
between probably the two most basic components of natural language semantics: force-
dynamics and space.
One avenue to explore is how this model can help us to model the typological results
of Bowerman and Choi (2001:485). They show that there is a universal hierarchy of
topological static relations that ranges from a typical instance of support (cup on table)
to containment (apple in bowl), with less typical relations in between:

(30) cup bandaid picture handle apple apple

on table on leg on wall on door on twig in bowl
<–––––––––––––––––– ON ––––––––––––––––––––> <–IN–>
<–––––––OP––––––><––––––––––AAN––––––––––––> <–IN–>
<–––––––––––––––––––––––––EN––––––––––––––––––––––––>

Languages carve up this scale in different ways, but terms always correspond to con-
tinuous regions. If a language uses a term X for two situations then it also uses it for
every situation in between. This is illustrated in (30) for English, Dutch and Spanish,
respectively. This continuity property is strongly related to the property of convexity that
Gärdenfors (2000) proposed as a constraint on regions in conceptual spaces, but also to
the notion of connectivity in the semantic map approach in typology (Haspelmath 2003).
My point here concerns not so much the nature of this general property, but rather
the way we could use the force-dynamic schemas proposed here to give us more insight
into the conceptual space underlying prepositions like in and on in various languages,
in other words, in the conceptual space of containment and support. If we compare our
representations for Dutch op, aan and in, we can see the beginnings of a way to model
the hierarchy in (30).
 FORCEFUL PREPOSITIONS 211

(31) op: Figure –-> Ground

aan: Figure <–- Ground
in: Ground –> Figure <– Ground

There are different parameters here that can be manipulated: whether the Figure or
the Ground is the agentive participant, whether the force vector is directed towards
the Ground or towards the Figure and additionally, whether the force vector is typical
downward (as with op) or not. Another parameter is whether the force is simplex (with
op and aan) or complex (with in). In this way, we might hope to get a scale in which (the
prototypes of) op and in are maximally distinct with a gradient in between, in which
the parameters change from op to in:

(32) op in
Force source: Figure …………………………. Ground
Force orientation: Ground ……………………… Figure
Force direction: Down …………………………. Not down
Force complexity: Simplex……………………… Complex

In this way the analysis of forceful prepositions proposed here is not only relevant
for English and Dutch, but for all languages across the world that refer to notions of
containment, support and attachment.
If the approach of this paper is on the right track, then it also sheds an interesting light
on two common and influential ideas in the literature on topological prepositions like
in and on, which go back to work of Herskovits (1986) and Vandeloise (1991). One idea
is that the semantics of in and on is based on a particular type of geometry, namely the
topological one, in which basic relations between spatial regions play a role (as opposed to
the axis-based semantics of projective prepositions like above and behind). In corresponds
to ‘inclusion’ while on corresponds to ‘contiguity’ or ‘connectedness’. Vandeloise came with
an alternative, non-geometric approach based on functional or force-dynamic notions like
‘containment’ and ‘support’. The results of this paper suggest, however, that geometry vs.
function may be a false dichotomy. Spatial geometry and force-dynamics are not mutually
exclusive, but they are both based on a more fundamental notion of vector, which makes
it possible to take a more unified approach towards these conceptual domains.

Notes
1 This paper was presented at the ICLC 9 in Seoul, July 22, 2005. The research for this
paper was financially supported by a grant from the Netherlands Organization for
Scientific Research NWO to the PIONIER project ‘Case Cross-Linguistically’ (number
220–70–003), which is gratefully acknowledged. The comments of an anonymous
reviewer have been very helpful for me in revising the paper.
2 This zero force vector should be kept distinct with the zero spatial vector that might
potentially be used to represent the purely spatial relation of contact between Figure and
Ground. However, as argued in Zwarts and Winter (2000), there are several reasons to
analyze the spatial contact relation of on in terms of non-zero vectors.
212 LANGUAGE, COGNITION AND SPACE

References
Beliën, M. (2002) Force dynamics in static prepositions: Dutch aan, op, and tegen.
In H. Cuyckens and G. Radden (eds) Perspectives on Prepositions 195–209.
Tübingen: Niemeyer.
Bowerman, M. and Choi, S. (2001) Shaping meanings for language: Universal
and language-specific in the acquisition of spatial semantic categories. In M.
Bowerman and S.C. Levinson (eds) Language Acquisition and Conceptual
Development 475–511. Cambridge: Cambridge University Press.
Carlson, L. and van der Zee, E. (eds) (2005) Functional Features in Language and
Space: Insights from Perception, Categorization, and Development. Oxford: Oxford
University Press.
Coventry, K.R. and Garrod, S. (2004) Saying, Seeing and Acting: The Psychological
Semantics of Spatial Prepositions. Psychology Press, Hove.
Croft, W. (1991) Syntactic Categories and Grammatical Relations: The Cognitive
Organization of Information. Chicago: University of Chicago Press.
Croft, W. (2009) Aspectual and causal structure in event representations. In V.
Gathercole (ed.) Routes to Language Development: In Honor of Melissa Bowerman
139–166. Mahwah: New Jersey: Lawrence Erlbaum Associates.
Cruse, D.A. (1986) Lexical Semantics. Cambridge: Cambridge University Press.
Dowty, D. (1991) Thematic proto-roles and argument selection. Language 67(3):
547–619.
Gärdenfors, P. (2000) Conceptual Spaces: The Geometry of Thought. Cambridge, MA:
MIT Press.
Gruber, J.S. (1976) Lexical Structures in Syntax and Semantics. Amsterdam: North-
Holland.
Haspelmath, M. (2003) The geometry of grammatical meaning: Semantic maps
and cross-linguistic comparison. In M. Tomasello (ed.) The New Psychology of
Language 211–243. (Vol. 2) New York: Erlbaum.
Herskovits, A. (1986) Language and Spatial Cognition: An Interdisciplinary Study of
the Prepositions in English. Cambridge: Cambridge University Press.
Jackendoff, R. (1987) The status of thematic relations in linguistic theory. Linguistic
Inquiry 17: 369–411.
Jackendoff, R. (1993) The combinatorial structure of thought: The family of causative
concepts. In E. Reuland and W. Abraham (eds) Knowledge and Language, Volume
II: Lexical and Conceptual Structure 31–49. Dordrecht: Kluwer Academic Press.
Johnson, M. (1987) The Body in the Mind: The Bodily Basis of Meaning, Imagination,
and Reason. Chicago: University of Chicago Press.
Langacker, R.W. (1987) Foundations of Cognitive Grammar. Vol. 1: Theoretical
Prerequisites. Stanford: Stanford University Press.
Langacker, R.W. (1991) Concept, Image, and Symbol: The Cognitive Basis of Grammar.
Berlin: Mouton de Gruyter.
O’Keefe, J. (1996) The spatial prepositions in English, vector grammar, and the
cognitive map theory. In P. Bloom et al. (eds) Language and Space 277–316.
Cambridge, Mass: MIT Press.
 FORCEFUL PREPOSITIONS 213

Talmy, L. (1983) How language structures space. In H. Pick and L. Acredolo (eds)
Spatial Orientation: Theory, Research, and Application 225–282. New York:
Plenum Press.
Talmy, L. (1985) Force dynamics in language and thought. In Papers from the
Twenty-first Regional Meeting of the Chicago Linguistic Society 293–337. Chicago:
University of Chicago.
Vandeloise, C. (1991) Spatial Prepositions: A Case Study from French. Chicago:
University of Chicago Press.
Wolff, P. and Zettergren, M. (2002) A vector model of causal meaning. In W. D. Gray
and C.D. Schunn (eds) Proceedings of the 24th Annual Conference of the Cognitive
Science Society 944–949. Mahwah, New Jersey: Lawrence Erlbaum Associates,
Publishers.
Zwarts, J. (1997) Vectors as relative positions: A compositional semantics of modified
PPs. Journal of Semantics 14: 57–86.
Zwarts, J. and Winter, Y. (2000) Vector space semantics: A model-theoretic analysis
of locative prepositions. Journal of Logic, Language and Information 9: 169–211.
9 From the spatial to the non-spatial: the ‘state’
lexical concepts of in, on and at
Vyvyan Evans

1 Introduction

This paper is concerned with modelling the lexical representation of spatial relations,
particularly as encoded by English prepositions, and examining how these spatial
relations give rise to non-spatial meanings. In previous work Andrea Tyler and I
(Evans and Tyler 2004a, 2004b; Tyler and Evans 2001, 2003) modelled the extensive
polysemy exhibited by prepositions, and sought to provide a principled framework
for characterising their distinct sense-units. We also sought to establish boundaries
between senses as they inhere in semantic memory. In so doing, we attempted to
account for this polysemy in a motivated way, as an outcome of situated language use,
the nature of human socio-physical experience and the relevant cognitive mechanisms
and processes.
Nevertheless, the framework of Principled Polysemy we developed was not prima-
rily concerned with modelling the complexity of the spatio-geometric and functional
semantic properties, and the extremely complex functional knowledge that prepositional
sense-units assist in conveying. This follows as it was primarily concerned with address-
ing perceived methodological weaknesses in early work in cognitive lexical semantics,
as exemplified by the work of Brugman and Lakoff (Brugman [1981] 1988; Brugman
and Lakoff 1988; Lakoff 1987). In particular, it is becoming clear that Tyler and I, in
our work on Principled Polysemy, may, in fact, have underestimated the functional
complexity that ‘spatial’ prepositional sense-units encode.
Accordingly, the goal of this paper is to present a more recent theory of lexical
representation which builds on and refines the framework of Principled Polysemy. This
approach, I argue, better accounts for some of the complexities I will be describing with
respect to the sorts of knowledge structures that prepositions encode, as evidenced
in language use. Following Evans (2004a 2004b; see also Evans 2006, 2009), this
theory employs two central constructs: the lexical concept, and the cognitive model.
In brief a lexical concept is a relatively complex sense-unit which is conventionally
associated with a specific form. Moreover, certain kinds of lexical concepts afford
access to large-scale multi-modal knowledge structures: cognitive models. Cognitive
models constitute relatively stable, non-linguistic knowledge structures, which are
subject to ongoing modification as we continue to interact in the world and in com-
municative settings. Moreover, cognitive models provide the complex informational
characterisation lexical concepts invoke in meaning construction processes. As the
constructs of the lexical concept and the cognitive model are of central importance,

215
216 LANGUAGE, COGNITION AND SPACE

the theory of lexical representation to be presented is termed the theory of Lexical

Concepts and Cognitive Models, or LCCM Theory for short. The theoretical discussion
presented later in the paper is based on more detailed explications of LCCM Theory
(Evans 2006, 2009).
The main analytical focus of the paper is the so-called ‘state’ senses of English prepo-
sitions, as associated with prepositions such as in, at, and on. While these sense-units
presumably derive from, and are certainly related to ‘spatial’ senses encoded by the same
forms, they are not, in and of themselves, primarily spatial in nature. Representative
examples are provided below.

(1) We are in love/shock/pain ‘state’ sense

cf. We are in a room ‘spatial’ sense

(2) We are at war/variance/one/dagger’s drawn/loggerheads ‘state’ sense

cf. We are at the bus stop ‘spatial’ sense

(3) We are on alert/best behaviour/look-out/the run ‘state’ sense

cf. We are on the bus ‘spatial’ sense

In these examples, in, at and on mediate a relation between human experiencer(s) and a
particular state. While some of these expressions, for instance, to be ‘at daggers drawn’
are clearly idiomatic, the contention of cognitive lexical semantics is that while such
expressions may be highly conventionalised, and the source of the idiom may not be
accessible to contemporary language users, the fact that at is employed is, diachronically
at least, motivated (see Evans and Green 2006: chapter 10, for a review; see also Evans,
Bergen and Zinken 2007).
If the perspective offered by cognitive semantics is correct, namely that the use of
in, at and on to encode a ‘state’ meaning is motivated, deriving from historically earlier,
and synchronically, perhaps, more primary ‘spatial’ senses, then there are a number
of issues which await explanation. Firstly, how do we account for the derivation of
non-spatial, what we might dub ‘abstract’ senses from historically earlier spatial senses?
One solution to this problem has been to posit underlying conceptual metaphors as the
solution (Lakoff and Johnson 1999). That is, due to the conceptual metaphor, qua sub-
symbolic knowledge structure, of the sort glossed as states are locations, states of
the type captured in (1) to (3) inclusive are conceptualised as locations. On the metaphor
account, the existence of an independently motivated conceptual metaphor licenses the
development of new polysemous senses associated with in, at and on.
Despite the intuitive appeal of the conceptual metaphor account, this cannot be the
whole story. After all, each of the ‘state’ senses associated with the prepositions evident
in (1)-(3) exhibit distinct patterns in terms of the semantic arguments with which they
collocate. Put another way, the ‘state’ senses associated with the different prepositional
forms: in, on and at, are not equivalent. For instance, the ‘state’ sense associated with in
relates to semantic arguments which have to do with emotional or psychological ‘force’
such as being ‘in love’, ‘in pain’ and so on. In contrast, the semantic arguments associated
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 217

with at have to do, not with emotional force but, rather, with mutual (or interpersonal)
relations, such as being ‘at war’. Meanwhile, on relates to semantic arguments that have
to do with time-restricted activities and actions which involve being currently active in
some sense. These include being ‘on alert’, ‘on duty’, and so forth. That is, the semantic
arguments associated with each of the ‘state’ senses for these prepositions is of a quite
different kind. This suggests that the ‘state’ meanings conventionally associated with each
of these prepositional forms is also of a distinct kind. While this does not preclude a
conceptual metaphor account as part of the story, positing a unified metaphoric account
for examples of the kind provided in (1) to (3) does not, in itself, adequately account
for the linguistic facts.
The challenge, then, for a theory of lexical representation, which assumes that the
‘state’ sense-units are motivated and related, is to account for the fact that i) each of
these prepositions exhibits a conventional ‘state’ lexical concept, and ii) that each of the
‘state’ lexical concepts diverges. Put another way, we must account for the differential
motivation that gives rise to the similar, yet distinct, ‘state’ lexical concepts associated
with each of these prepositions. Thus, the ‘state’ lexical concepts present an intriguing
challenge which, I shall argue, existing theories of lexical representation, notably the
theory of Principled Polysemy, cannot, at present provide an account for. For this reason,
we require a more sophisticated account of lexical representation.
I will employ linguistic data associated with these ‘state’ lexical concepts in order
to provide a reasonably detailed illustration of how LCCM Theory accounts for the
functional complexity of the semantics involved. I argue that LCCM Theory facilitates
i) a revealing descriptive analysis of the ‘state’ lexical concepts of these prepositions,
including the way in which these sense-units are in fact distinct from one another; and
ii) a revealing account of the spatio-geometric and functional knowledge that the core
‘spatial’ lexical concepts associated with in, at and on encode; and finally, in view of this,
iii) a revealing account of how each of the ‘state’ lexical concepts involved is motivated
by, and related to, the core ‘spatial’ lexical concepts associated with each preposition.
A further reason for selecting the ‘state’ lexical concepts as a case study is as follows.
While there is now a voluminous literature on spatial semantics, especially within cogni-
tive lexical semantics, this work has primarily been concerned with examining the range
of distinct sense-units associated with a given preposition, including a now impressive
body of research which has focused on principles for determining sense-boundaries,
including psycholinguistic and corpus-based approaches (e.g., Sandra and Rice 1995
and Gries 2005 and the references therein). However, hitherto, there has been, in relative
terms, comparatively little research on the non-spatial lexical concepts associated with
prepositional forms, and how they are related to one another and derived from spatial
lexical concepts. This lack of research makes an examination of the ‘state’ lexical concepts
of different prepositions an issue worthy of attention.
There are two claims that I make, and which the findings presented serve to
substantiate. Firstly, ‘new’ lexical concepts derive from already extant lexical con-
cepts by virtue of inferential processes, relating to situated instances of language use.
Hopper and Traugott (1993) refer to such a mechanism as pragmatic strengthening:
an inferential process whereby a new semantic representation is abstracted from an
218 LANGUAGE, COGNITION AND SPACE

extant semantic representation in what has been referred to as a bridging context (N.
Evans and Wilkins 2000). A bridging context is a context of use in which the new
lexical concept emerges as a situated inference (or an ‘invited inference’, Traugott and
Dasher 2004). A polysemous relationship thereby holds between the extant and the
derived lexical concept. I argue that the polysemous lexical concepts associated with
the prepositional forms to be examined arise due to new parameters being encoded,
giving rise to distinct lexical concepts. These parameters arise due to the functional
consequences of spatio-geometric properties in situated language use, about which I
shall have more to say below.
The second claim is as follows. The ‘state’ lexical concepts for each prepositional
form are distinct, as revealed by an examination of their lexical profiles: the semantic
and grammatical selectional tendencies exhibited. Moreover, each form has a number
of conventional ‘state’ lexical concepts associated with it, which are different from one
another. Put another way, there are clear differences in terms of ‘state’ lexical concepts
both across and within the prepositions I address here.

2 The functional nature of the spatial semantics of prepositions

The point of departure for this study relates to the functional nature of the semantics
associated with spatial relations as lexicalised by prepositions. Recent work in the
framework of cognitive semantics (e.g., Herskovits 1986, 1988; Vandeloise 1991, 1994)
has shown that the received or traditional view is descriptively inadequate in terms of
accounting for how the core, prototypical or ideal ‘spatial’ sense-units associated with
prepositions are actually used. The received view, which following Herskovits I refer
to as the simple relations model, holds that the prototypical sense-unit associated with
a given preposition straightforwardly encodes purely spatio-geometric properties, i.e.,
‘simple’ relations.
My purpose in this section is to make the case for a functional characterisation of
the ‘spatial’ lexical concept associated with a given preposition. By ‘functional’ I mean
the following. To understand how language users employ the core ‘spatial’ lexical concept
of a preposition we must also allow for non-spatial parameters which form part of the
linguistic content encoded by the lexical concept. The use of the term ‘functional’ is moti-
vated by the observation that such non-spatial parameters are a functional consequence
of humanly relevant interactions with the spatio-geometric properties in question.
Moreover, the way ‘spatial’ lexical concepts are ordinarily employed by language users
would appear to require such a functional understanding if ‘spatial’ lexical concepts are
to be correctly interpreted in context.
Providing a functional account is of further importance as the derived lexical
concepts which result from sense-extensions (such as the ‘state’ lexical concepts of in,
on and at), cannot be adequately accounted for without first recognising that in addi-
tion to spatio-geometric parameters, the core ‘spatial’ lexical concept associated with
a prepositional form also includes functional information. That is, if we assume that
the derived lexical concepts are motivated by the prototypical lexical concept, as is the
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 219

case in cognitive lexical semantics, then we must assume a relatively complex (albeit
schematic) body of ‘functional’ knowledge, if we are to account for the derivation of
extended lexical concepts.
In this section, therefore, I briefly review some of the arguments made by Herskovits,
and Vandeloise for thinking that functional information also constitutes part of the
linguistic content associated with ‘spatial’ lexical concepts for prepositions (see also
Coventry and Garrod 2004; Deane 2005, and Feist This volume for a related perspective).
I begin with Herskovits. In her work she observes that the received view has assumed
that the ‘basic’ function of the spatial sense-units associated with prepositional forms is
to encode purely spatial relations. On this view, the semantic contribution of any given
spatial use of a preposition relates to spatio-geometric properties, typically designating
a relation involving notions such as dimensions, axes or proximity (e.g., Bennett 1975;
Miller and Johnson-Laird 1976 for representative examples).
This general approach, particularly as has been evident in formal and computa-
tional accounts of prepositions, as noted above, Herskovits (e.g., 1988) refers to as the
simple relations model. Yet, as Herskovits shows in detail, the simple relations model
is descriptively inadequate. That is, the ‘simple’ spatial relations posited are unable to
account for the range of spatial representations that prepositions ordinarily designate.
Some of the descriptive shortcomings of the simple relations model relate to phenomena
such as the following.
Firstly, the same preposition often appears to include quite distinct geometric
descriptions:

(4) a. the water in the vase

b. the crack in the vase

The example in (4a) relates to an entity: the water, the trajector (TR), ‘contained’ by
the landmark (LM), the vase. That is, it relates to the volumetric interior of the LM. In
contrast, in (4b) the semantic contribution of in concerns a relation between a ‘negative’
region, namely a lack of substance, a crack, which is not part of the volumetric interior of
the vase, but rather forms part of the landmark-boundary, namely the physical structure
of the vase. Put another way, in relates to quite distinct spatio-geometric relations in
these examples. This is problematic for the simple relations model which assumes that
a given preposition encodes a single spatio-geometric relation,
Secondly, the spatial relations encoded by prepositions often appear to diverge from
straightforward ‘simple’ relations. For instance, the following expression:

(5) the dictionary on the table

can be used unproblematically to refer to a dictionary placed on top of another book

which is ‘on’ the table. That is, the dictionary is not actually ‘on’ the table, but rather ‘on’
the book which is in direct contact with, and therefore ‘on’, the table.
Thirdly, there often appears to be what Herskovits refers to as ‘added constraints’
which apply to prepositions. For instance, in examples of the following kind:
220 LANGUAGE, COGNITION AND SPACE

(6) a. the man at the desk

b. the schoolboy at the bus-stop

the relation implied is more specific than ‘simple’ spatio-geometric relations. That is, the
example in (6a) implies, and is understood to mean, that not only is the TR in question,
the man, in close proximity to his desk, but he is also working at his desk (or at least
in a position to do so). Similarly, in (6b), in addition to the co-locational relation, this
expression implies that the schoolboy is ‘waiting’ at the bus-stop, presumably for a bus.
In other words, part of the meaning of these utterances is functional in nature. The
schoolboy is co-located with the bus-stop in order to catch a bus. Implications such
as these are not explained by the simple relations model. In fact, we seldom employ
prepositions simply to describe a purely spatio-geometric relationship.
Fourthly, there are often unexplained context dependencies associated with preposi-
tions which the simple relations model fails to account for. In an example such as the
following:

(7) Max is at the crèche

this utterance appears only to work when both speaker and addressee are not also present
at the crèche. In the case when the speaker and addressee are located at the crèche, the
following would be more likely:

(8) Max is (somewhere) in the crèche

Finally, there are a number of other restrictions which appear to relate to discursive
salience and/or relevance. Again, these are not accounted for by the simple relations
model. For instance, in a scenario such as that represented by Figure 1, in which there
is an apple located beneath an upturned bowl, the following expression is semantically
anomalous:

(9) #the apple in the bowl

Figure 1. The apple beneath the bowl

Herskovits argues that in view of the failure of the simple relations approach a modified
view of the lexical representation for spatial prepositions is required.
A related perspective has been presented by Vandeloise in his work. Vandeloise
(1991, 1994) argues compellingly that any account of spatial semantics that leaves out
the functional nature of prepositional lexical concepts fails to properly account for
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 221

how they are actually employed. That is, spatio-geometric relations have functional
consequences, consequences which arise from how we interact with objects and entities
in our physical environment, and in our daily lives. To illustrate, take the mundane
example of a cup of coffee. Imagine holding it in your hand. If you move the cup slowly
up and down, or from side to side, the coffee moves along with the cup. This follows as
the cup is a container with a bottom and sides and thus constrains the location of any
entity within these boundaries. Tyler and I (2003) referred to this property of bounded
landmarks as ‘location with surety’.
The force-dynamic properties associated with a cup as a container also show up
in linguistic content, as illustrated by the semantic contribution of the preposition in.
Consider the diagram in Figure 2 drawn from the work of Vandeloise (1994).

Figure 2. A bottle or a light bulb? (adapted from Vandeloise 1994)

Vandeloise observes that the image depicted in Figure 2 could either represent a bottle
or a light bulb. As example (10) shows, we can use the preposition in to describe the
relation between the light bulb (TR) and the socket (LM).

(10) The bulb is in the socket

In contrast however, we cannot use in to describe the relation between a bottle and its
cap, as illustrated by (11).

(11) #The bottle is in the cap

Vandeloise points out that the spatial relation holding between the TR and LM in each of
these utterances is identical, and yet while (10) is a perfectly acceptable sentence (11) is
semantically odd. Vandeloise suggests that it is not the spatial relation holding between
the TR and LM that accounts for the acceptability or otherwise of in. He argues that the
relevant factor is one of force-dynamics: ‘[W]hile the socket exerts a force on the bulb
and determines its position, the opposite occurs with the cap and the bottle’ (Vandeloise,
1994: 173). In other words, not only is the position and the successful function of the
bulb contingent on being in (contained by) the socket, but the socket also prevents the
bulb from succumbing to the force of gravity and falling to the ground. In contrast,
the position and successful functioning of the bottle is not contingent on being in the
cap. This suggests that our knowledge of the functional consequences associated with
located containment affects the contextual acceptability of a preposition such as in.
222 LANGUAGE, COGNITION AND SPACE

3 Principled polysemy revisited

Having begun to consider the functional nature of the spatial semantics of prepositions,
I now reconsider the model of Principled Polysemy as an account of spatial semantics. In
developing this model Tyler and I (e.g., Tyler and Evans 2001, 2003) sought to model the
nature of the lexical representations associated with spatial particles such as prepositions.
In so doing we were concerned with two sorts of issues. Firstly, we were concerned with
accurately describing the nature and range of the distinct (albeit related) lexical concepts
(what we referred to as ‘senses’) associated with lexical categories such as prepositions.
That is, we were concerned with providing a constrained (i.e., principled) methodology
for establishing sense-units and thus sense-boundaries.
Secondly, we were concerned with accounting for how sense-units (lexical concepts
in present terms) arise. We posited that the lexical concepts which populate a semantic
network for a given preposition are diachronically related, and the derivation of ‘new’
lexical concepts (i.e., sense-extension) is motivated (see Evans and Tyler 2004a in
particular). Both these issues required detailed analysis of the lexical representations
associated with the various lexical concepts for a given preposition. Moreover, this in
turn entailed examination of spatio-geometric, and non-spatio-geometric, aspects of
prepositional lexical concepts.
For instance, while an important part of the semantic representation for over in (12)
has to do with the spatio-geometric relationship holding between the TR and the LM,
in (13) an important part of the lexical representation relates to non-spatio-geometric
aspects, i.e., occlusion.

(12) The picture is over the sofa

(13) The veil is over her face

In (12) the semantic contribution of over relates to an ‘above’ relation, which concerns
the spatio-geometric relationship in a 3-dimensional region holding between the TR
and LM. In (13), while part of the linguistic content of over must also encode spatio-
geometric information – as occlusion is a consequence of a physical relationship holding
between artefacts and the vantage point of a perceiver from which the artefacts are
viewed – nevertheless, the semantic contribution of over is more saliently identifi-
able as the functional notion of ‘occlusion’. Examples such as this, in which over is not
interpreted as providing a semantic contribution relating to ‘above’ but rather ‘occlusion’,
provide good evidence that the occurrence of over in (13) is sanctioned by a distinct
lexical concept: we are dealing with a lexical concept which is distinct vis-à-vis the
‘above’ lexical concept which sanctions the use of over in (12).
In our analyses, Tyler and I made the point that functional lexical concepts such as
what we referred to as the Covering Sense of over (i.e., the [occlusion] lexical concept in
present terms), obtain because spatial experience is inherently meaningful for humans.
That is, as human beings we interact with objects around us in our spatial environment
(see Johnson 1987; 2007). Particular spatial relations, as manifested by the linguistic
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 223

content encoded by prepositional lexical concepts, have functional consequences. These

functional consequences we described as arising from experiential correlations, an idea
we borrowed, and adapted from the work of Grady (1997).
For instance, a consequence of the spatio-geometric property associated with over
in examples such as (12), i.e., an ‘above’ relation, is that in certain contexts, occlusion
occurs. To illustrate consider (14):

(14) The tablecloth is over the table

In this example, the use of over is sanctioned by a lexical concept that encodes a spatio-
geometric relation in which the TR is in an ‘above’ relation with respect to the LM.
However, a functional consequence of how we interact with TRs such as tablecloths,
and LMs such as tables, and given the dimensions of tablecloths, such that they often
have a greater extension than tables, is that by virtue of being over (i.e., above), the
tablecloth thereby occludes the table. Thus, we argued that due to such contexts of
use, over can, by virtue of the process of reanalysis termed pragmatic strengthening
(as briefly introduced above), lead to the ‘occlusion’ reading becoming ‘detached’ from
the context in which it grounded, and reanalysed as a distinct sense-unit of over in
its own right.
A related idea that was important in the Principled Polysemy framework was the
notion of a functional element, an idea inspired by the work of Vandeloise (e.g., 1991,
1994) in his functional approach to spatial semantics. This notion related to the central
or core sense in a semantic polysemy network. Such lexical concepts we termed proto-
scenes. The proto-scene for over, what we termed the Above Sense as exemplified in (12),
constitutes an abstraction over spatio-geometric properties associated with the range
of spatial scenes in which a given preposition, such as over, is used.
However, as already noted, a large part, perhaps the majority, of uses of the proto-
scene of a given prepositional form relate to usages which are not purely or even wholly
spatio-geometric in nature (see Vandeloise 1991, 1994 and especially Herskovits 1986,
1988 as described above). Thus, Tyler and I argued that functional information forms
part of the semantic representation of any given proto-scene (see Evans and Tyler 2004a;
Tyler and Evans 2003 for details).
In sum, Principled Polysemy posits two kinds of lexical concept which popu-
late a prepositional polysemy network. The first kind, the proto-scene, is primarily
spatio-geometric in nature. Moreover, the proto-scene corresponds – for most of the
prepositions we surveyed – to the historically earliest lexical concept associated with
a given prepositional form (Tyler and Evans 2003). Nevertheless, proto-scenes include
a functional element, reflecting the way in which proto-scenes are ordinarily used.
That is, language users typically employ proto-scenes in ways which draw upon the
functional consequence of interacting with spatial scenes of certain kinds in humanly
relevant ways. Thus, linguistic knowledge associated with proto-scenes appears to
involve more than simply knowing the particular spatio-geometric properties encoded
by a particular form.
224 LANGUAGE, COGNITION AND SPACE

The second sort of lexical concept – the remainder of the senses in a prepositional
polysemy network – we hypothesised as being motivated by, and ultimately derived
from, the proto-scene. This said, we observed that the derivation is often complex
and indirect (see Tyler and Evans 2003 for detailed discussion). These derived lexical
concepts we referred to as sense-extensions. These ‘new’ lexical concepts, we argued, were
derived by virtue of the process of reanalysis (pragmatic strengthening) due to expe-
riential correlations of the sort described above for the development of the Occluding
Sense from the Proto-scene (i.e., the Above Sense).
One issue which Tyler and I largely side-stepped, in the version of Principled
Polysemy which appeared as Tyler and Evans (2003), concerned how best to account
for ‘common’ lexical concepts of different prepositions, such as the ‘state’ lexical concepts
for in, on and at, illustrated above in the examples in (1) to (3). The difficulty here is that
as the ‘state’ lexical concepts associated with in, at and on, for instance, are all identified
by a common label, this might be construed as suggesting that there is common semantic
representation. Yet, the ‘state’ lexical concepts appear, on the contrary, to be distinct
sense-units as evidenced by the distinct semantic arguments with which they each
collocate: their lexical profiles, in present terms. What is required is a theory of lexical
representation which has methodological tools for distinguishing between ostensibly
‘similar’ lexical concepts associated with different forms.
A further difficulty is that it is unclear, in Principled Polysemy, what the nature of
the functional relationship is holding between the lexical representation associated with
the proto-scene, and the diverse ‘functional’ lexical representations associated with the
range of derived senses we posited. That is, while Principled Polysemy posited a single
functional element associated with each proto-scene, it is not clear how this would
motivate the functional complexity apparent in the plethora of functionally diverse
extended senses, posited for each prepositional form.
Thus, while an important construct, there is good reason, therefore, to think that the
notion of a functional element associated with the proto-scene, as presented in Evans
and Tyler (2004b) and Tyler and Evans (2003) actually underestimates the functional
complexity that must be readily available to language users, as encoded by the range and
various combination of parameters associated with the distinct ‘state’ lexical concepts
across and within prepositions.
Ultimately, the difficulty for the Principled Polysemy framework is that while it
attempted to provide a detailed account of lexical representation, because of its primary
concern with detailing a rigorous methodology for establishing distinct sense-units,
it failed to work out the implications of the functional nature of spatial semantics for
lexical representation. 1
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 225

4 The Lexical Concepts and Cognitive Models (LCCM) approach

to lexical representation

In recent work (Evans 2006, 2009), I have begun to develop an approach to lexical
representation which is consistent with the context-dependent nature of the meanings
associated with words. Indeed, part of the focus of this particular research programme
is to develop an account of how lexical representations give rise to situated meaning
construction, and thus to provide a cognitively-realistic approach to meaning construc-
tion. While the issue of situated meaning construction is less relevant to the analysis
of how best to represent the ‘state’ lexical concepts in the present paper, and won’t be
addressed further, Evans (2006) constitutes an attempt to model lexical representation
that is relevant for any lexical class, including prepositions.
The starting point for the LCCM Theory account is the premise that linguistic
knowledge is usage-based. That is, I assume that the organisation of our language
system is intimately related to, and derives directly from, how language is actually
used (Croft 2000; Langacker 2000; Tomasello 2003). Through processes of abstraction
and schematisation (Langacker 2000), based on pattern-recognition and intention-
reading abilities (Tomasello 2003), language users derive linguistic units. These are
relatively well-entrenched mental routines consisting of conventional pairings of form
and semantic representation. The semantic representations conventionally associated
with a given unit of form, I refer to, as already noted, as a lexical concept.
While lexical concepts are mental representations, they underspecify the range
of situated meanings associated with a given form in an individual utterance. Thus,
I make a fundamental distinction between lexical concept as a mental unit, and
its context-dependent realisation in an utterance. This is akin to the distinction in
Phonological Theory between the abstract notion of a phoneme and the actual unit of
realised context-dependent sound, the allophone. My claim is that there is an essential
distinction between lexical representation and meaning. While meaning is a property
of the utterance, lexical representations are the mental abstractions which we infer must
be stored as part of the language user’s knowledge of language, in order to produce the
range of novel uses associated with situated instances of a particular word such as a
preposition. The meaning associated with an utterance I refer to as a conception. Thus,
conceptions are a function of language use.
There are a number of important properties associated with lexical concepts. I briefly
review some of the most relevant here (for detailed discussion see Evans 2009). Firstly,
and as noted above, linguistic units, as I use the term, are conventional pairings of form
and meaning. From this it follows that lexical concepts are form-specific. Secondly, as
mentioned above, although lexical concepts are form-specific, a single form can be
conventionally associated with a potentially large number of distinct lexical concepts
which are related to degrees as attested by the phenomenon of polysemy.2 That is, forms
are not lexical concept-specific. A consequence of this is that the lexical concepts which
share the same form can be modelled in terms of a semantic network (see Evans and
Green 2006: chapter 10 for discussion).
226 LANGUAGE, COGNITION AND SPACE

Thirdly, the definitional property of any given lexical concept is that it has a lexi-
cal profile, its unique ‘biometric’ identifier. A lexical profile is an extension of criteria
presented in Evans (2004a), and akin to the notion of an ‘ID tag’ (Atkins 1987) and
‘behavioural profile’ (Gries 2005). While a lexical concept associated with a particular
form can be provided with a semantic gloss, as in the case of lexical concepts associated
with over, an example of which I glossed as [above] or the lexical concepts associated
with in, at and on to be examined later which I preliminarily gloss as [state], whether a
particular usage of a form relates to one lexical concept rather than another is a matter of
examining the ‘selectional tendencies’ (the lexical profile) associated with a given usage.
While any given usage of a lexical concept will have its own unique collocational pat-
tern, general patterns can be established, and form part of the conventional knowledge
associated with a particular lexical concept.
Two sorts of information form a lexical concept’s lexical profile. The first relates to
semantic selectional tendencies. In Evans (2004a) this was referred to as the Concept
Elaboration Criterion. The second relates to formal or grammatical selectional tenden-
cies. In Evans (2004a) I referred to this as the Grammatical Criterion. Gries (2005) has
advocated the way in which corpus methodologies can be used to examine the lexical
profile associated with a specific lexical concept. For instance, each of the ‘state’ lexical
concepts associated with in, at and on have distinct lexical profiles. In the remainder of
this chapter I primarily rely on semantic selectional tendencies for adducing distinct
lexical concepts.
To provide a preliminary illustration of the construct of the lexical profile, I briefly
consider two lexical concepts, both of which I provisionally gloss as [state] – although
I revise this gloss later in the chapter – and which are conventionally encoded by the
English prepositional forms in and on. These are evidenced by the following examples:

(15) a. John is in trouble/danger

b. Jane is in love/awe
c. Fred is in shock
d. Jake is in a critical condition

(16) a. The guard is on duty

b. The blouse is on sale
c. The security forces are on red alert

While both in and on have ‘state’ lexical concepts conventionally associated with them,
the lexical profile for each is distinct. For instance, the [state] lexical concept associated
with on selects semantic arguments which relate to states which normally hold for a
limited period of time, and which contrast with salient (normative) states in which the
reverse holds. For instance, being ‘on duty’ contrasts with being off-duty, the normal
state of affairs. Equally, being ‘on sale’ is, in temporal terms, limited. Sales only occur for
limited periods of time at specific seasonal periods during the year (e.g., a winter sale).
Similarly, being ‘on red alert’ contrasts with the normal state of affairs in which a lesser
security status holds. Further, the states in question can be construed as volitional, in
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 227

the sense that to be ‘on duty/sale/red alert’ requires a volitional agent who decides that
a particular state will hold and takes the requisite steps in order to bring such a state
of affairs about.
In contrast, the semantic arguments selected for by the [state] lexical concept for
in relate to states which do not necessarily hold for a limited period of time, and do not
obviously contrast with a ‘normal’ state of affairs. Moreover, while states encoded by on
are in some sense volitional, states associated with in are, in some sense, non-volitional.
That is, we do not usually actively choose to be in love, shock or a critical condition,
nor can we, by a conscious act of will, normally bring such states about. That is, these
states are those we are affected, constrained and influenced by, rather than those which
are actively (in the sense of consciously) chosen.
The fourth and final property of lexical concepts that I review here concerns the
position that they have bipartite organisation. That is, lexical concepts encode linguistic
content and facilitate access to conceptual content. Linguistic content represents the
form that conceptual structure takes for direct encoding in language, and constitutes
what might be thought of as a ‘bundle’ of distinct knowledge types. There are a large
number of different properties encoded by linguistic content which serve to provide a
schematic or skeletal representation, which can be encoded in language (for a review
see Evans 2009: chapter 6). The one which is relevant for the present study relates to
the notion of parameterisation.
One way in which knowledge, in general terms, can be represented is in terms of
richly inflected nuances that serve to reflect the complexity of experience. An alternative
way is to ‘compress’ such fine distinctions into two, three or more, much broader, and
hence, more general distinctions. These I refer to as parameters. Linguistic content
serves to encode content by adopting the latter strategy, which is to say, to employ
parameterisation. Parameters are hence part of the ‘bundle’ of information that a lexical
concept encodes.
To illustrate this notion, consider the complex range of expressions that a language
user might employ, in English, in order to ‘locate’ themselves with respect to time,
thereby facilitating time-reference. Any one of the following could conceivably be
employed, depending upon context: today, January, 2008, the day after yesterday, the day
before tomorrow, this moment, now, this second, this minute, this hour, today, this week,
this month, this quarter, this year, this half century, this century, this period, the 8th day
of the month, this era, this millennium, and so on. A potentially unlimited set of finer
and finer distinctions can be made (e.g., 1 second ago, 2 seconds ago, 1 hour 4 minutes
and 3 second ago, 2 days ago, etc.), reflecting any manner of temporal distinction we
might care to make.
In contrast, paramaterisation functions by dividing all the possible distinctions relat-
ing to a given category, such as time-reference, into a small set of divisions: parameters.
Such parameters might distinguish between the past, for instance, and the non-past.
Indeed, this is the basis for the tense system in English, as illustrated by the following:

(17) a. He kicked the ball Past

b. He kicks the ball Non-past
228 LANGUAGE, COGNITION AND SPACE

English encodes just two parameters that relate to Time-reference: Past versus Non-
past, as exhibited by the examples in (17), and thus manifests a binary distinction.
Some languages, such as French, have three parameters: Past, Present and Future. Some
languages have more than three parameters, distinguishing additionally remote past
from recent past, for instance. The language with the most parameters for linguisti-
cally encoding time-reference is an African language: Bamileke-Dschang with eleven.
Crucially, parameters are encoded by specific lexical concepts, and thus form part of the
knowledge ‘bundle’ that constitutes a lexical concept. For instance, the parameter ‘past’
is encoded by the lexical concept associated with the –ed form in (17a). However, other
lexical concepts also include the parameter ‘past’ such as the lexical concepts associated
with the following forms: sang, lost, went, etc.
I argue, then, that a key feature of linguistic (as opposed) to conceptual content is
that it only encodes knowledge in parametric fashion. This is not to say that conceptual
content does not parameterise knowledge. Indeed, parameterisation is simply a highly
reductive form of abstraction: it serves to abstract across the complexity exhibited by
a particular category. The point, however, is that the parameters encoded by linguistic
content serves to ‘strip away’ most of the differences apparent in the original perceptual
experience, thereby reducing it to a highly limited number of parameters.
In addition to encoding linguistic content, a subset of lexical concepts – those
conventionally associated with open-class forms (see Evans 2009 for discussion of this),
serve as access sites to conceptual content. Conceptual content relates to non-linguistic
information to which lexical concepts potentially afford access. The potential body of
non-linguistic knowledge, what I also refer to as a lexical concept’s semantic potential, is
modelled in terms of a set of cognitive models. I refer to the body of cognitive models,
and their relationships as accessed by a given lexical concept, as the cognitive model
profile. A design feature of language is that it involves a bifurcation of lexical concept
types: those which are relatively more schematic in nature, such as those associated with
prepositional forms, the subject of the present study, and those which are relatively richer
in nature. As I am dealing with lexical concepts associated with closed-class forms in
this study, namely prepositions, I will have little more to say about cognitive models in
the remainder of this chapter.

5 Two factors in accounting for ‘state’ lexical concepts: lexical

profiles and parameters

In the Principled Polysemy framework the prototypical (i.e., spatial) sense with respect to
which a semantic network is structured is a proto-scene. As we saw earlier, proto-scenes
have a single functional element associated with them. In LCCM Theory in contrast,
lexical representations, and thus proto-scenes, are representationally more complex
than this, especially with respect to their functional properties. In this section I briefly
reconceptualise the nature of the core lexical concept associated with a prepositional
polysemy network in the light of LCCM Theory.
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 229

The prototypical semantic representation associated with a preposition, like the

other lexical concepts in the prepositional polysemy network, is a lexical concept.
As we saw in the previous section, lexical concepts have bipartite organisation: they
facilitate access to conceptual content and encode linguistic content. As prepositional
lexical concepts are associated with prepositions: closed-class forms, they constitute
closed-class lexical concepts. As such, while they encode linguistic content they do not
serve as access sites to conceptual content.
There are two aspects of linguistic content that will be relevant for the discussion
of the polysemy exhibited by the range of ‘state’ lexical concepts in this study. The first
concerns the lexical profile exhibited by lexical concepts, as manifested by distinct
collocational patterns in language use. As we saw earlier in the chapter, two sorts of
information form a lexical concept’s lexical profile: semantic selectional tendencies, and
formal or grammatical selectional tendencies. In this study I employ distinctions in the
semantic arguments which, I hypothesise, collocate with distinct ‘state’ lexical concepts
to uncover distinctions in lexical concepts both within and between prepositions.
The second aspect of linguistic content that will be relevant relates to parameterisa-
tion. One characteristic that serves to distinguish between lexical concepts, both across
prepositions and within a single preposition, relates to the parameters encoded. For
instance, the prototypical ‘spatial’ lexical concept associated with in, which I gloss as
[enclosure], encodes the parameter Containment, as evidenced by the example in
(18). In contrast, the [emotion] lexical concept – one of the ‘state’ lexical concepts
associated with in – encodes the parameter Psycho-somatic State, as evidenced in (19),
but not the Containment parameter.

(18) The kitten is in the box Parameter: Containment

(19) John is in love Parameter: Psycho-somatic state

That is, the [enclosure] lexical concept in (18) encodes a schematic dimension
abstracted from sensory-motor experience in which a TR is contained by the LM.
Notice that the relation encoded is highly schematic in nature; it says nothing about
whether there is contact or not between the TR and LM as in (20), nor as to whether
the TR represents part of the LM or not as in (21):

(20) a. The fly is in the jar (i.e., flying around)

b. The fly is in the jar (i.e., stationary on one interior surface)

(21) There’s a crack in the vase

Indeed, the precise spatio-geometric nature of the TR, LM and their relationship is
a function of the TR and LM and their possible forms of interaction, rather than the
abstract parameter encoded by the [enclosure] lexical concept associated with the
prepositional form in. This information derives from the semantic potential accessed
230 LANGUAGE, COGNITION AND SPACE

via the open-class lexical concepts, as mediated by compositional processes (see Evans
2009 for details).
In contrast, the [emotion] lexical concept associated with in encodes the parameter
Psycho-somatic state. This information is highly schematic in nature. That is, the param-
eter encoded does not determine which sorts of psycho-somatic states can collocate
with this lexical concept. This is a function of the lexical profile: knowledge relating
to the semantic selectional tendencies associated with this lexical concept, and hence
the range of psycho-somatic states which can co-occur with the [emotion] lexical
concept. Hence, while the parameters encoded by a lexical concept determine the pos-
sible range of semantic arguments that can co-occur, the lexical profile, which relates to
stored knowledge based on usage-patterns, provides information relating to the range
of permissible states which can co-occur with this lexical concept.

6 Functional consequences of parameters

I now consider how the ‘state’ lexical concepts arise from historically earlier spatial lexical
concepts, giving rise to the phenomenon of polysemy. Put another way, polysemy is a
consequence of new, or derived lexical concepts emerging, thereby exhibiting a semantic
relationship with a synchronically present – albeit diachronically antecedent – lexical
concept.
Based on arguments developed in Tyler and Evans (2001, 2003) I argue that the
spatio-geometric knowledge, encoded, in present terms, as abstract parameters by the
‘spatial’ lexical concepts associated with prepositional forms gives rise to the develop-
ment of non-spatial lexical concepts. In other words, ‘state’ lexical concepts emerge
by virtue of parameters such as that of Psycho-somatic state arising as a functional
consequence of spatio-geometric properties, in particular usage contexts. Hence, the
emergence of derived lexical concepts is a consequence of the functional consequences
of spatio-geometric parameters in a specific context of use. Such contexts of use Tyler
and I (2001, 2003) referred to as spatial scenes.
For instance, there are a large number of distinct sorts of spatial scenes that involve
the prototypical spatial lexical concept: [enclosure], associated with in, and which
hence encode the parameter Containment. This follows as different bounded land-
marks – a landmark which exhibits the structural properties interior, boundary and
exterior – have different functions, are employed for different ends and are viewed from
different vantage points. For instance, while a playpen, prison cell and a coffee cup all
restrict the containee to a specific location, they do so in service of different objectives,
respectively: safety, punishment and consumption. Hence, without understanding the
functional consequence of being located ‘in’ a bounded landmark such as a prison (cell),
the question in (22) would be uninterpretable:

(22) What are you in for?

 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 231

After all, in, here, does not relate directly to a given spatial relation, but rather to the
specific sets of knowledge systems relating to the ‘containment’ function of prison in
a particular society. Thus, in (22), being ‘in’ relates not purely to containment, a func-
tional consequence of the [enclosure] lexical concept, but rather, and in addition, to
punishment, a functional consequence of being contained in enclosures (i.e., bounded
landmarks) of a certain kind, i.e., prisons, which occupy a certain position, and fulfil a
specified role in the socio-cultural and legal institutions of a particular society.
Now consider a different sort of functional consequence associated with the [enclo-
sure] lexical concept for in. One consequence of certain sorts of bounded landmarks is
their utility in providing security. This is evident in the scenario involving a very small
child in a playpen for instance. But it is also true of bounded landmarks such as safes
used to safeguard valuable commodities such as money or jewels. Indeed, a functional
consequence of bounded landmarks of this sort is that the contents are occluded. This
of course assumes that the vantage point from which the bounded landmark is viewed
is exterior with respect to the volumetric interior of the bounded landmark in question,
here the safe. Thus, ‘containment’ or ‘location with surety’ is a functional consequence
of the spatial relation (i.e., the lexical concept) conventionally associated with in, i.e.,
of [enclosure].
The point is, then, that when in is employed in any given utterance, the conception
which derives will almost certainly always relate to a functional consequence attendant
on a specific sort of spatial scene, involving a containment relation, but will do so in
service of objectives and consequences specific to the sort of spatial scene in question.
Put another way, bounded landmarks are of many different kinds, a consequence of the
many different ways in which we interact with, and the complex range of functions to
which we put, bounded landmarks.
In terms of the phenomenon of polysemy, which is to say the emergence of derived
lexical concepts, it is precisely functional consequences of this sort which give rise to new
parameters. Such new parameters become conventionally associated with a lexical form,
and hence contribute to the formation of a new lexical concept. The occlusion afforded
by certain kinds of bounded landmarks, such as a jeweller’s safe, is a consequence of
placing valuables in a landmark that serves to protect the commodity in question.
Typically, such landmarks are made of materials that serve to occlude the contents,
a consequence – rather than the objective – of employing the types of materials used
for constructing the safe. This functional consequence has become abstracted from
such spatial scenes to give rise to a distinct parameter. This forms part of the linguistic
content encoded by a distinct lexical concept. Evidence for this comes from examples
of the following sort:

(23) The sun is in

This utterance relates to lack of visibility of the sun, rather than the sun, the TR, being
enclosed by a bounded LM of some sort. That is, the functional consequence of certain
sorts of containment relations has given rise to a distinct lexical concept which has a
Lack of Visibility parameter encoded as part of its linguistic content.
232 LANGUAGE, COGNITION AND SPACE

7 Lexical concepts for in

In this section I present an LCCM analysis of the ‘state’ lexical concepts associated with
in. That is, I argue that there is more than one distinct ‘state’ lexical concept conven-
tionally associated with the prepositional form in. I also show how these ‘state’ lexical
concepts relate to and are motivated by the functional consequences attendant upon
the range of spatial scenes which involve usages of in sanctioned by the [enclosure]
lexical concept.

7.1 ‘Spatial’ lexical concepts for in

As noted above, the central ‘spatial’ lexical concept associated with in I gloss as [enclo-
sure]. This lexical concept encodes the parameter Containment. This parameter con-
stitutes an abstraction across the spatio-geometric properties associated with bounded
landmarks, such as a box, as lexicalised by the example in (18). The key spatio-geometric
components associated with a LM such as a box is that it has the structural elements inte-
rior, boundary and exterior (see Tyler and Evans 2003: chapter 7 for detailed discussion).
There are a diverse range of complex conceptualisations across which the parameter
Containment is abstracted. This includes, at the very least, experiences relating to a
TR: the entity enclosed, and a bounded landmark which serves to enclose the TR.
Bounded landmarks themselves consist of many types even in everyday experience. For
instance, a bounded landmark includes an interior, which further subsumes an interior
surface, and the volumetric interior bounded by the interior surface. It also subsumes
a boundary, which can be rigid, as in a metal safe, or non-rigid, as in a plastic carrier
bag. The boundary also has other physical characteristics such as permeability and
degrees of opacity. Finally, the bounded landmark has, by definition, an exterior: that
region which constitutes the inverse of the volumetric interior. Accordingly, part of the
exterior includes the exterior surface.
As observed earlier, due to our interaction with enclosures, in is associated with a
number of functional consequences. That is, there are a number of identifiably distinct
sorts of functional categories associated with spatial scenes involving enclosure. These
include Location with Surety, Occlusion and Affecting conditions. Bounded landmarks
that are specialised for providing a Location with Surety function are known as ‘contain-
ers’. These can provide a support function by virtue of containing (i.e., holding and
restricting) the location of the TR. This was illustrated with the discussion of the light
bulb in the socket example earlier. Alternatively, containers can restrict access (and
escape), as in the case of prisons, and safes. The second functional category mentioned
relates to Occlusion. A consequence of certain bounded landmarks, due to the opacity
of the material which forms the boundary, is that the figure located on the volumetric
interior is occluded, and hence hidden from view. The third functional category, that of
Affecting conditions, relates to the fact that an enclosure provides a delimited environ-
ment which thereby affects the TR located on the volumetric interior. For instance, a
prisoner held in solitary confinement in a windowless sound-proofed room is thereby
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 233

subjected to a particular sensory environment, which is a direct consequence of the

nature of the bounded landmark in which s/he is located.
I suggest that it is these functional categories, which arise from the spatio-geometric
property of Enclosure, that serve to become abstracted as distinct parameters. Put
another way, abstracting across different sorts of sense-perceptory experiences, namely
the spatio-geometric properties associated with enclosures, gives rise to an Enclosure
parameter. Abstracting across re-occurring functional consequences of the spatio-
geometric properties associated with enclosure gives rise to further parameters notably
Location with Surety, Occlusion and Affecting Conditions. These parameters, which
arise from spatial scenes involving enclosure, are diagrammed in Figure 3.
Enclosure

Occlusion Spatial scenes Location

involving enclosure with surety

Affecting
conditions
Figure 3. Parameters deriving from spatial scenes involving enclosure

I suggest that the emergence of the parameters: Location with Surety, Occlusion
and Affecting Conditions, associated with the linguistic content encoded by in, can,
under certain conditions, give rise to new ‘state’ lexical concepts. While the parameter
Enclosure, entails, under certain conditions, all of the other parameters illustrated in
Figure 3, the other parameters do not necessarily entail the Enclosure parameter. For
this reason, as I shall argue, the Enclosure parameter can be seen to be primary; the
other parameters arise from spatial scenes in which Enclosure is a key attribute.
The means whereby new lexical concepts arise is due to a disjunction between the
various parameters. I illustrate this with the examples below which reveal the disjunction
between the Enclosure and Location with Surety parameters.
To do so, consider examples of the following kind:

(24) The toy is in the box

(25) a. The bulb is in the socket

b. The flower is in the vase
c. The umbrella is in his hand

The example in (24) is, I suggest, a consequence of the two parameters: Enclosure and
Location with Surety. That is, by virtue of being located in the interior portion of the
234 LANGUAGE, COGNITION AND SPACE

bounded landmark, the TR is thereby enclosed. Moreover, by virtue of being enclosed,

the TR is located with surety: if the box is moved, so also, is the TR – the toy – as a
direct consequence. This is what it means to say that Location with Surety is entailed
by Enclosure.
Evidence for thinking that the Location with Surety and Enclosure parameters
are, nevertheless, distinct units of knowledge encoded as part of a lexical concept’s
linguistic content comes from spatial scenes involving partial enclosure. In the
examples in (25), the TR is only partially enclosed by the bounded landmark: only
the base of a bulb is enclosed by the socket as illustrated in Figure 2, above, only the
stem, and not the whole flower, is enclosed by the vase (see Figure 4); and only the
umbrella handle is enclosed by the hand (see Figure 5). Indeed, the reason that the
form in can relate to spatial scenes involving partial, as well as full, enclosure is due
to the parameter of Location with Surety. It is precisely because the bounded LM
that partially encloses the TR serves to provide location with surety that the form in
is sanctioned in these instances.

Figure 4. The flower is in the vase

Figure 5. The umbrella is in his hand

On the basis of the examples in (24) and (25), there is no reason, however, to be con-
vinced that Enclosure and Location with Surety constitute distinct parameters, and
hence distinct knowledge units encoded as part of the linguistic content associated with
the [enclosure] lexical concept.
However, the example in (26) illustrates a crucial disjunction between the two.
While the TR, the bottle, is partially enclosed by the bounded LM, the cap, in exactly
the same way as the relationship between the bulb and the socket, this use of in in (26)
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 235

is semantically anomalous, as indicated by the hash sign. In the spatial scene described
by this example, the bottle is not located with surety by virtue of being partially enclosed
by the cap. That is, the bottle’s location is not determined by being partially enclosed by
the cap – although access to its contents are. Hence, in a situation where partial enclosure
applies, but location with surety does not, the [enclosure] lexical concept associated
with in cannot be applied. This reveals that in the absence of the Location with Surety
parameter, in cannot be applied to spatial scenes involving only partial enclosure.

(26) #The bottle is in the cap

The examples thus far considered reveal that the Enclosure parameter entails Location
with Surety. Hence, in spatial scenes in which there is no location with surety, yet
there is (partial) enclosure, as in the spatial scene to which (26) refers, the use of the
[enclosure] lexical concept cannot apply, as shown by the semantic unacceptability
of (26).
We must next examine whether the Location with Surety parameter can be employed
independently of the Enclosure parameter. If so, then we can posit that there is a distinct
lexical concept, which we can gloss as [location with surety], a lexical concept which
encodes the Location with Surety parameter as part of its linguistic content but does not
also feature the Enclosure parameter. Evidence for such a state of affairs is provided by
the following example, which relates to the spatial scene depicted in Figure 6.

(27) The pear is in the basket

Figure 6. The pear is in the basket

In this example, the pear (in the centre of the image) is not enclosed by the basket, as
it is supported by other fruit; although the supporting fruit are enclosed by the basket.
Yet, the form in can be applied to this spatial scene, as is evident in (27). I argue that
this is due to a [location with surety] lexical concept which sanctions this particular
usage. While the [enclosure] lexical concept apparent in (24) and (25) encodes the
Enclosure and Location with Surety parameters, the [location with surety] lexical
concept encodes the Location with Surety parameter but not the Enclosure parameter as
part of its linguistic content. This difference in linguistic content between the two lexical
concepts explains the difference in linguistic behaviour in the examples just considered.
The [enclosure] lexical concept requires full enclosure, or, partial enclosure plus
236 LANGUAGE, COGNITION AND SPACE

location with surety. However, in (27) neither full nor partial enclosure is apparent, yet
in is sanctioned. This follows as the independent, but semantically related (and hence
polysemous) [location with surety] lexical concept sanctions this use, I suggest.
Thus, we see that there are, plausibly, at least two ‘spatial’ lexical concepts associated with
in, [enclosure] and [location with surety], which encode different configurations
of parameters, and hence, subtly distinct linguistic content.

7.2 ‘State’ lexical concepts for in

I now turn to the ‘state’ lexical concepts, in order to see how these arise from the spatial
lexical concepts. Consider the following examples involving in.

(28) a. The cow is in milk

b. The girl is in love
c. John is in trouble/debt
d. He’s in banking [i.e., works in the banking industry]

While each relates to a ‘state’ of some kind, these examples in fact relate to slightly
different ‘states’: those that have a physical cause, as in (28a) – the state of being ‘in
milk’, which is a consequence of the physical production of milk – those that have a
psychological or emotional cause, as in (28b) – the state is a consequence of a subjective
state, which may (or may not) have physical, i.e., observable, manifestations – those
that have a social/inter-personal cause, as in (28c) – resulting from social/interpersonal
interactions which result in an externally-maintained state – and those that are a result
of a habitual professional activity, as in (28d). Put another way, each of these ‘states’
take distinct semantic arguments, relating a particular entity to quite different sorts
of states. In essence, I argue that these examples are sanctioned by four distinct ‘state’
lexical concepts for in. These distinct ‘state’ lexical concepts, as we shall see below, I
hypothesise to emerge from the functional category Affecting Conditions, which arises
from spatial scenes involving enclosure. I spell out the distinctions between the ‘state’
lexical concepts for in, below, with additional examples.
Physiological state (resulting in a ‘product’)
(29) a. The cow is in milk
b. The cow is in calf
c. The woman is in labour
Psycho-somatic state (i.e., subjective/internal state)
(30) a. John is in shock/pain (over the break-up of the relationship)
b. John is in love (with himself/the girl)
Socio-interpersonal state (i.e., externally-maintained state)
(31) a. The girl is in trouble (with the authorities)
b. John is in debt (to the tune of £1000/to the authorities)
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 237

Professional state (i.e., professional activity habitually engaged in)

(32) a. He is in banking
b. She is in insurance

The fact that in collocates with semantic arguments of the distinct kinds illustrated in
(29–32), relating to physiological, psycho-somatic, socio-interpersonal and professional
conditions or properties suggests that we are dealing with four distinct lexical concepts.
This follows as LCCM Theory claims that each distinct lexical concept has a unique
lexical profile.
In addition to evidence based on semantic selectional tendencies, the position
that there must be a number of distinct ‘state’ lexical concepts associated with in, along
the lines illustrated by the distinct examples in (29) to (32) inclusive can also be dem-
onstrated by virtue of ambiguities associated with an utterance of the following kind:

(33) She’s in milk

The utterance in (33) could potentially be interpreted as relating to a woman who is

nursing a baby, and thus lactating, or as relating to a woman who works in the dairy
industry. That is, given an appropriate extra-linguistic context, an example such as
this can be interpreted in at least two ways. The potential for divergent interpretations
is a consequence, in part, of our knowledge that in has a number of distinct lexical
concepts associated with it: what is relevant for this example is the distinction between
a [physiological state] lexical concept and a [professional state] lexical concept.
Moreover, ambiguities can be generated even when a relatively well entrenched example
is employed. For instance, even examples of the following kind:

(34) She is in labour

(35) He is in love

can be interpreted in alternate ways. For instance, (34) could be interpreted as relating
to childbirth or to a professional activity, e.g., the trade union movement. Similarly,
(35) could be interpreted as relating to an emotional state or a professional activity, e.g.,
marriage guidance counselling. The former reading is only possible by virtue of assuming
something akin to an [psycho-somatic state] lexical concept which is distinct from a
[professional state] lexical concept. That is, both lexical concepts must exist if ‘love’
can be interpreted in these ways in this example.

7.3 Derivation of the ‘state’ lexical concepts

I now consider how the ‘state’ lexical concepts for in exemplified in (29) to (32) inclusive
may have been extended from the prototypical [enclosure] lexical concept. I observed
above that in previous work with Andrea Tyler, Tyler and I argued that polysemy derives
238 LANGUAGE, COGNITION AND SPACE

from regular processes of semantic change, in which situated implicatures associated

with a particular context can become reanalysed as distinct semantic components,
in present terms, lexical concepts, which are associated with the relevant preposition
(Hopper and Traugott 1993; Traugott and Dasher 2004; cf. Levinson 2000). That is,
Tyler and I argued for a usage-based approach to language change, a position adopted
by LCCM Theory.
In terms of an LCCM account of the emergence of closed-class lexical concepts such
as the ‘state’ lexical concepts for in, the trajectory is as follows. Situated implicatures arise
in bridging contexts, as briefly discussed above. These are contexts in which a usage
sanctioned by the relevant ‘spatial’ lexical concept, such as the [enclosure] lexical
concept, also gives rise to a situated implicature, such as an affecting condition. If the
form is repeatedly used in such bridging contexts, the situated implicature may give
rise to the formation of a parameter: a highly abstract unit of knowledge, specialised
for being encoded as part of the linguistic content associated with a lexical concept, as
discussed earlier. I argue below that bridging contexts, involving the functional category
of Affecting Conditions, give rise to the formation of a number of related but distinct
‘state’ parameters, and hence lexical concepts.
In order to trace the development of the functional category Affecting Conditions,
we need to consider spatial scenes that might provide appropriate bridging contexts.
To illustrate, consider the following expressions:

(36) a. in the dust

b. in the sand
c. in the snow

While dust, sand and snow are physical entities which can ‘enclose’ they cannot, normally
fulfil the functions provided by, for instance, containers. That is, they do not typically
serve to locate with surety, exceptional circumstances such as quicksand and avalanches
excepted. For instance, dust, sand and snow, by virtue of enclosing, do not normally have
the structural attributes that allow an entity to be supported and thus transported (cf. a
bucket), nor do they normally restrict access in the way a prison cell does, for instance.
Nevertheless, these examples exhibit some of the spatio-geometric properties asso-
ciated with the [enclosure] lexical concept. This is a consequence of the properties
associated with these ‘bounded’ landmarks: they provide an affecting condition, an
environmental influence which affects our behaviour. For instance, they determine
the kinds of apparel we wear, and how we behave when we are exposed to the dust/
sand/snow, and so on. While examples such as sand, snow and dust can be construed
as enclosures with boundaries, there are other related examples of what we might refer
to as Prevailing Conditions which are much less clear-cut in terms of the nature of the
boundaries involved:

(37) a. the flag in the storm

b. the flag in the wind
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 239

I suggest that these instances of in are sanctioned by virtue of there existing a distinct
parameter Affecting conditions, which forms part of the linguistic content encoded by a
distinct [prevailing conditions] lexical concept. Clearly a storm and wind are much
less prototypically enclosures, and more saliently provide prevailing conditions which
thereby constitute an environment which affects us. As such, spatial scenes involv-
ing more prototypical enclosures have given rise to the functional category Affecting
Conditions, which has led to the formation of a distinct Affecting Conditions parameter
in semantic memory. The existence of a distinct [prevailing conditions] lexical
concept, as evidenced by examples in (37) provides suggestive evidence that such a
distinct Affecting Conditions parameter must exist, and has come to form the core a
distinct [affecting conditions] lexical concept.
I argue that there are a number of distinct ‘state’ lexical concepts associated with
in that encode the parameter of Affecting Conditions, rather than Enclosure – those
evidenced in (29)-(32). Indeed, these lexical concepts are what I have referred to as ‘state’
lexical concepts, as the states invoked all provide, in some sense, affecting conditions.
Moreover, all these ‘state’ lexical concepts are relatively, and to degrees, far removed
from the physical notion of enclosure from which they most likely originally evolved.
In essence, once an Affecting Conditions parameter becomes conventionalised, it can
be applied to distinct kinds of affecting conditions, even those that are non-spatial
in nature, such as states. This leads to the development of new lexical concepts, with
correspondingly new lexical profiles.
The first such ‘state’ lexical concept relates to the physical condition of an organism
which thus provides an affecting condition. Such physical conditions include good/ill
health, pregnancy, and any salient physical aspect of the organism’s condition which
affects and thus impacts on the organism’s functioning. This lexical concept I gloss
as [physiological state]. In addition to environmental and physical conditions,
affecting conditions can be caused by psycho-somatic states, such as grief, happiness
and sadness which are internal in nature. This ‘state’ gives rise to a [psycho-somatic
state] lexical concept associated with in. In addition, social interactions which give
rise to social or interpersonal relationships lead to conditions which may affect the
individual. Such extrinsic or socially-induced affecting conditions might include debts,
or other sorts of difficult situations which impose conditions on the behaviour of an
individual. This set of affecting conditions gives rise, I suggest, to what I gloss as the
[socio-interpersonal state] lexical concept associated with in. Finally, one’s habitual
professional activity provides an affecting condition by virtue of the physical and social
interactions that are attendant upon such activities. This provides an affecting condition
giving rise to a lexical concept glossed as [professional state] associated with in.
These are illustrated in Figure 7.
240 LANGUAGE, COGNITION AND SPACE

Enclosure

Occlusion Containment
Spatial scenes
involving enclosure

Affecting conditions

[PREVAILING [PHYSIOLOGICAL [PSYCHO-SOMATIC [SOCIO-INTERPERSONAL [PROFESSIONAL

CONDITIONS] STATE] STATE] STATE] STATE]

Figure 7. Parameters and their relationship with ‘state’ lexical concepts for in

8 Lexical concepts for on

In this section I deal, somewhat more briefly, with on.

8.1 The prototypical lexical concept for on: [CONTACT]

The spatial relation designated by on involves the relation of contact or proximity to

the surface of a LM, and so the functional consequence of being supported or upheld
by it. I gloss the prototypical ‘spatial’ lexical concept conventionally associated with on
as [contact]. This serves to encode the geometric parameter Contact and functional
parameter Support as part of its linguistic content. This lexical concept licenses an
example of the following sort:

(38) the apple on the table

Note that evidence that the parameters Contact and Support are both encoded by the
lexical concept [contact] comes from the fact that on can only felicitously be employed
to describe spatial scenes in which both parameters are apparent. For instance, if an
apple is held against a wall by someone, the utterance in (39) is semantically anomalous.
However, if the apple is affixed to the wall, for instance by glue, then (39) is entirely
appropriate.
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 241

(39) the apple on the wall

That is, while the apple is in contact with the wall in both scenarios, in the first scenario
it is the person, rather than the wall, that affords support, while it is the wall (and the
glue, which employs the wall as a means of affixing the apple) in the second. Hence,
the example in (39) applies when there is both physical contact between the TR and
the LM, and when the latter has a role in supporting the former.
Indeed, there are a number of distinct ‘support’ lexical concepts associated with
on which privilege the Support parameter, at the expense of the Contact parameter, as
illustrated by the following examples:
Body part which provides support
(40) a. on one’s feet/knees/legs/back
b. on tiptoe
c. on all fours

In the examples in (40), the use of on relates to that part of the body which provides
support, rather than being concerned with contact.
Means of conveyance
(41) a. on foot/horseback
b. on the bus

With respect to the example in (41b), it is worth pointing out, as Herskovits (1988) does,
that if children were playing on a stationary bus, for instance, that had been abandoned,
then it would not be appropriate to say: on the bus, but rather in would be more natural.
This supports the view that the [means of conveyance] lexical concept is a distinct
‘support’ lexical concept encoded by on.
Supporting pivot
(42) The Earth turns on its axis
Drug dependency/continuance

(43) a. Are you on heroin?

b. She’s on the pill
Psychological support
(44) You can count/rely on my vote
Rational/epistemic support
(45) on account of/on purpose
242 LANGUAGE, COGNITION AND SPACE

8.2 The [ACTIVE STATE] lexical concept for on

There is just one ‘state’ lexical concept for on, which I gloss as [active state]. This
lexical concept derives not from the functional category of Support. Rather, it pertains
to a functional category concerning ‘functionality’ or ‘activity’. That is, in many spatial
scenes, a consequence of contact is that the TR, as it comes into contact with a particular
surface, becomes functional. This category I refer to as Functional Actioning. Removing
contact precludes functional actioning. Such forms of contact, for instance, invoke
scenarios involving physical transmission, such as the very salient one of electricity.
Many times a day we plug-in or switch ‘on’ electrical appliances. It is by facilitating
contact between the appliance and the electrical circuit that an appliance is rendered
functional. A ‘switch’ provides a means of facilitating this contact, which is why we
employ the term ‘switch on’ in English. In other words, I suggest that the [active state]
lexical concept associated with on encodes a Functional Actioning parameter as part
of its linguistic content. It is this which makes it distinctive from the ‘spatial’ lexical
concepts of on discussed in the previous examples.
The [active state] lexical concept associated with on relates to adjectives or
nouns of action which involve a particular state which can be construed as ‘active’ or
‘functional’, as contrasted with a, perhaps, normative scenario in which the state does
not hold. In other words, states described by instances of on sanctioned by this lexical
concept are often temporally circumscribed and thus endure for a prescribed or limited
period of time. In this, the states referred to are quite distinct from those that in serves
to describe. Here, the notion of being ‘affected’, apparent with in, is almost entirely
absent. Consider some examples:

(46) a. on fire
b. on live (i.e., a sports game)
c. on tap (i.e., beer is available)
d. on sleep (as in an alarm clock on a particular mode)
e. on pause (as in a DVD player)
f. on sale
g. on loan
h. on alert
i. on best behaviour
j. on look-out
k. on the move
l. on the wane
m. on the run

Figure 8 depicts the parameter that underpins this lexical concept.

 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 243

Contact Spatial scenes Support

involving contact

Functional Actioning

[ACTIVE STATE]

Figure 8. Parameters and their relationship with ‘state’ lexical concept for on

9 The state senses of at

This section briefly examines the ‘state’ lexical concepts of at.

9.1 The prototypical lexical concept for at: [co-location]

The lexical concept which licenses spatial uses of at affords the most general expression of
localisation in space in English, expressing the relation between a TR and a point of space
that it is contiguous or proximal with. This lexical concept I gloss as [co-location].
Consequently, it is one of the most polysemous of all English prepositions. Indeed,
this lexical concept for at forms a contrast set (Tyler and Evans 2003) with the ‘place’
identifying lexical concepts associated with other prepositions. The [co-location]
lexical concept encodes the Co-location parameter, designating a highly abstract spatial
relation between a TR and a place, when the relation is not more precisely expressed
by ‘spatial’ lexical concepts associated with the following prepositional forms: near, by,
on, in, over, under, all of which, at times, can be encoded by at.
Perhaps the most salient functional category associated with at constitutes what I
will refer to as that of Practical Association. That is, a functional consequence of being
co-located with a particular LM is that the TR has some practical association with the
reference object. This is evidenced in the examples in (6) discussed earlier (e.g., at the
desk/bus-stop), and is particularly evident with examples such as the following:

(47) a. at school
b. at sea

In these examples, the activity associated with the school buildings or being out on the
sea is extremely salient.
244 LANGUAGE, COGNITION AND SPACE

9.2 The ‘state’ lexical concepts for at

There are three distinct lexical concepts associated with at that might be described as
relating to ‘states’. These are illustrated below:
State (or condition) of existence

(48) at rest/peace/ease/liberty
(e.g., He stood at ease, or He is at peace [=dead])
States relating to mutual relations

(49) at war/variance/strife/one/dagger’s drawn/loggerheads

(e.g., The EU is at war with the US over the imposition of steel tariffs)
States relating to external circumstances

(50) at peril/risk/hazard/expense/an advantage/a disadvantage

(e.g., The company is at risk of going under)

The ‘state’ lexical concepts for at appear to be motivated by the functional consequence
of close-proximity between two point-like entities giving rise to the formation of a
parameter: Practical Association.
In the case of the [state of existence] lexical concept, the practical association
resulting from the co-location is the state of existence which holds. That is, there is a
practical association which holds between a given entity and its state of existence.
The second lexical concept I gloss as [state of mutual relations], as evidenced
by (49). This lexical concept arises due to a salient practical association resulting from
co-location of two entities involving mutual relations. For instance, while warfare often
involves combatants who must be proximal to one another, the state of being ‘at war’
need not, as evidenced by the so-called ‘phoney war’ which held during 1939 when
the United Kingdom, France and Germany were officially ‘at war’, and yet no troops
engaged. Thus, the use of at to designate a state of mutual relations, independent of
spatio-geometric co-location, is due to the parameter of Practical Association being
invoked as part of the linguistic content encoded by this lexical concept. Put another
way, this lexical concept encodes a state of a particular kind, rather than the ‘spatial’
notion of proximity.
Finally, states pertaining to external circumstances may relate to evaluations con-
cerning circumstances associated with mutual relations. This is instantiated by the lexical
concept which I gloss as [states of external circumstances], as evidenced by the
examples in (50). The relationship between the parameter of Practical Association and
the ‘state’ lexical concepts is diagrammed in Figure 9.
 FROM THE SPATIAL TO THE NON-SPATIAL: THE ‘STATE’ LEXICAL CONCEPTS OF IN, ON AND AT 245

Co-location

spatial scenes involving

location

Practical
connection

[STATE OF EXISTENCE] [STATE OF MUTUAL RELATIONS] [STATE OF EXTERNAL

CIRCUMSTANCES]

Figure 9. Parameters and their relationship with ‘state’ lexical concepts for at

10 Conclusion: in vs. in vs. at

Having presented an analysis of i) distinct ‘state’ lexical concepts for in, on and at, and
ii) how these encode distinct parameters which relate to functional categories arising
from spatial scenes involving spatio-geometric relationships, I now return to one of the
observations with which I began this study. I observed that each of the ‘state’ lexical
concepts associated with in, on and at, as exemplified in (1)-(3), is minimally distinct in
that it is associated with distinct semantic arguments. Consequently the lexical concepts
exemplified in these examples relate to states of distinct kinds. The analysis presented
here has supported this initial assessment, and elaborated on it in three ways.
Firstly, the perspective offered here, particularly with respect to the construct of
the lexical concept, allows us to establish in a reasonably precise way the nature of
the distinction between the ‘state’ lexical concepts associated with in, on and at. That
is, given that lexical concepts are form-specific and moreover have distinct lexical
profiles – for instance they collocate with distinct kinds of semantic arguments – we
are able to establish that the ‘state’ lexical concepts (within and between) prepositional
forms are distinct.
Secondly, by taking seriously the functional nature of spatial relations, and the
formation of parameters: highly abstract knowledge structures specialised for being
directly encoded ‘in’ language, this allows us to understand the sorts of functional
motivations, and thus distinctions, between the ‘state’ lexical concepts among different
forms.
246 LANGUAGE, COGNITION AND SPACE

Thirdly, prepositions, particularly in and at have more than one so-called ‘state’
lexical concept associated with them. We have seen that the prototypical spatial lexical
concept associated with a given preposition is associated with a number of parameters,
including parameters derived from what I referred to as functional cognitive categories.
Providing an LCCM analysis gives us a way of establishing the sorts of distinctions that
exist between the ‘state’ lexical concepts associated with the same form. That is, we have
a means of understanding how these lexical concepts are distinct (based on a distinction
in parameters encoded) despite their conceptual similarity. We also have a means of
empirically verifying hypotheses as to distinctions in the underlying lexical concepts
which are assumed to sanction instances of use. This followed due to an examination of
semantic selectional tendencies, which relate to the theoretical construct of the lexical
profile in LCCM Theory: distinct lexical concepts are held to have a unique lexical profile
which forms part of the knowledge encoded by a given lexical concept.

Notes
1 This said, the framework developed in Tyler and Evans (2001, 2003) and Evans and
Tyler (2004a, 200b) remains important. Principled Polysemy, as articulated in those
publications, was and remains an important theoretical development in terms of what
it brings to descriptive accounts of spatial semantics. In particular, it sought, for good
reason, to address the sorts of methodological criticisms that had been levelled at the
early pioneering work of Brugman and Lakoff (Brugman [1981] 1988; Brugman and
Lakoff 1988; Lakoff 1987) in developing cognitive lexical semantics. While it doubtless
requires modification (see Evans 2004a), it nevertheless provides a relatively robust
set of methodologically constrained, and above all principled ‘decision principles’ (in
Sandra’s 1998 terms) for identifying and distinguishing between senses-units, and for, a
principled means of modelling lexical polysemy with respect to spatial relations. While
important developments in the use of psycholinguistic testing (see Sandra and Rice
1998; Cuyckens et al. 1997) and corpus-based techniques (see Gries 2005) have added
to the arsenal of cognitive lexical semanticists in this regard, empirical methods will
always require a theoretical framework in order to motivate the sorts of questions that
can be asked and to provide a lens for interpreting results, even though this may mean
modifying the theoretical framework. Indeed, this perspective is in fact compatible with
the desire to have more empirical methods in cognitive lexical semantics.
2 See Evans (2005) and Tyler and Evans (2001, 2003) for detailed discussion of polysemy.

References
Atkins, Beryl T. (1987) Semantic ID tags: Corpus evidence for dictionary senses.
Proceedings of the Third Annual Conference of the UW Centre for the New Oxford
English Dictionary 17–36.
Bennett, David (1975) Spatial and temporal uses of English prepositions. London:
Longman.
Brugman, Claudia ([1981] 1988) The Story of ‘over’: Polysemy, semantics and the
structure of the lexicon. New York: Garland.
Brugman, Claudia and George Lakoff (1988) Cognitive topology and lexical
networks. In S. Small, G. Cottrell and M. Tannenhaus (eds) Lexical ambiguity
resolution 477–507. San Mateo, CA: Morgan Kaufman.
Coventry, Kenny and Simon Garrod (2004) Saying, seeing and acting. The psychologi-
cal semantics of spatial prepositions. Hove: Psychology Press.
Croft, William (2000) Explaining language change: An evolutionary approach.
London: Longman.
Cuyckens, Hubert, Dominiek Sandra and Sally Rice (1997) Towards an empirical
lexical semantics. In B. Smieja and M. Tasch (eds) Human Contact Through
Language and Linguistics 35–54. Frankfurt: Peter Lang.
Deane, Paul (2005) Multimodal spatial representation: On the semantic unity of
over. In B. Hampe (ed.) From perception to meaning: Image schemas in cognitive
linguistics 235–284. Berlin: Mouton de Gruyter.
Evans, Nicholas and David Wilkins (2000) In the mind’s ear: The semantic extensions
of perception verbs in Australian languages. Language 76(3): 546–592.
Evans, Vyvyan (2004a) The structure of time: Language, meaning and temporal
cognition. Amsterdam: John Benjamins.
Evans, Vyvyan (2004b) How we conceptualise time. Essays in arts and sciences 33(2):
13–33.
Evans, Vyvyan (2005) The meaning of time: Polysemy, the lexicon and conceptual
structure. Journal of Linguistics 41(1): 44–75.
Evans, Vyvyan (2006) Lexical concepts, cognitive models and meaning-construction.
Cognitive Linguistics 17(4): 491–534.
Evans, Vyvyan (2009) How words mean. Oxford: Oxford University Press.
Evans, Vyvyan, Benjamin K. Bergen and Jörg Zinken (2007) The cognitive linguistics
enterprise: An overview. In V. Evans, B. Bergen and J. Zinken (eds) The cognitive
linguistics reader. London: Equinox Publishing Co.
Evans, Vyvyan and Melanie Green (2006) Cognitive linguistics: An introduction.
Edinburgh: Edinburgh University Press.
Evans, Vyvyan and Andrea Tyler (2004a) Rethinking English ‘prepositions of
movement’: The case of to and through. In H. Cuyckens, W. De Mulder and T.
Mortelmans (eds) Adpositions of movement (Belgian Journal of Linguistics 18)
Amsterdam: John Benjamins.
Evans, Vyvyan and Andrea Tyler (2004b) Spatial experience, lexical structure and
motivation: The case of in. In G. Radden and K. Panther (eds) Studies in linguistic
motivation 157–192. Berlin: Mouton de Gruyter.
Feist, Michele (This volume) Inside in and on: Typological and psycholinguistic
perspectives.
Grady, Joseph (1997) Foundations of meaning: Primary metaphors and primary
scenes. Doctoral Thesis, Linguistics dept., UC, Berkeley.
Gries, Stefan Th. (2005) The many senses of run. In S. Gries and A. Stefanowitsch
(eds) Corpora in cognitive linguistics. Berlin: Mouton de Gruyter.
Herskovits, Annette (1986) Language and spatial cognition. Cambridge: Cambridge
University Press.
Herskovits, Annette (1988) Spatial expressions and the plasticity of meaning. In B.
Rudzka-Ostyn (ed.) Topics in cognitive linguistics. Amsterdam: John Benjamins,
271–298.
Hopper, Paul and Elizabeth Closs Traugott (1993) Grammaticalization. Cambridge:
Cambridge University Press.
Johnson, Mark (1987) The body in the mind: The bodily basis of meaning, imagination and
reason. Chicago: University of Chicago Press.
Johnson, Mark (2007) The meaning of the body: Aesthetics of human understanding.
Chicago: University of Chicago Press.
Lakoff, George (1987) Women, fire and dangerous things: What categories reveal about the
mind. Chicago: University of Chicago Press.
Lakoff, George and Mark Johnson (1999) Philosophy in the flesh: The embodied mind and
its challenge to Western thought. New York, NY: Basic Books.
Langacker, Ronald (2000) A dynamic usage-basked model. In M. Barlow and S. Kemmer
(eds) Usage-based models of language 1–64. Stanford, CA: CSLI Publications.
Levinson, Stephen (2000) Presumptive meanings: The theory of generalized conversational
implicature. Cambridge, MA: MIT Press.
Miller, George and Philip Johnson-Laird (1976) Language and perception. Harvard:
Harvard University Press.
Sandra, Dominiek (1998) What linguists can and can’t tell you about the human mind: A
reply to Croft. Cognitive Linguistics 9(4): 361–478.
Sandra, Dominiek and Sally Rice (1995) Network analyses of prepositional meaning:
Mirroring whose mind – the linguist’s or the language user’s? Cognitive Linguistics
6(1): 89–130.
Tomasello, Michael (2003) Constructing a language: A usage-based theory of language
acquisition. Harvard: Harvard University Press.
Traugott, Elizabeth Closs and Richard Dasher (2002) Regularity in semantic change.
Cambridge: Cambridge University Press.
Tyler, Andrea and Vyvyan Evans (2001) Reconsidering prepositional polysemy networks:
The case of over. Language 77(4): 724–765.
Tyler, Andrea and Vyvyan Evans (2003) The semantics of English prepositions: Spatial
scenes, and the polysemy of English prepositions. Cambridge: Cambridge University
Press.
Vandeloise, Claude (1991) Spatial prepositions: A case study from French. (Translated by
Anna R. K. Bosch.) Chicago: University of Chicago Press.
Vandeloise, Claude (1994) Methodology and analyses of the preposition in. Cognitive
Linguistics 5(2): 157–184.
 Part V
Spatial representation in specific languages

249
10 Static topological relations in Basque
Iraide Ibarretxe-Antuñano

1 Space in language and cognition

Space is one of the most studied areas not only from the point of view of linguistic
description, that is, the description of the linguistic devices that languages have to
express and describe space, but also from the perspective of cognition, how space is
understood and computed in our brains. In recent years, a major focus of analysis in
the domain of space has been the relationship between language and cognition (Landau
and Jakendoff, 1993; Levinson, 2003; Majid et al., 2004).
In this paper, I will offer a description of the linguistic means available in Basque for
the lexicalisation of space as well as their usage, that is, which of these devices is used
more often by Basque speakers when they talk about space. More concretely, I will deal
with those devices used for the description of topological relations, what Levinson et
al. (2003: 486) call basic locative constructions, i.e. answers to ‘where’ questions. Based
on empirical data elicited by means of the Topological Relation Picture Series, stimuli
developed at the Max Planck Institute for Psycholinguistics (Pederson, Wilkins, and
Bowerman 1993), I will concentrate on the type of spatial information that is crucial
for Basque speakers in the description of space, and also on certain features that seem
to influence how Basque speakers conceptualise space.

2 Topological relational markers in Basque

One of the main difficulties that a linguist faces when s/he starts to analyse how space
is described in a given language is the wide range of semantic and morpho-syntactic
elements and mechanisms that are more or less directly involved in its codification. Space
is expressed not only by means of nouns, verbs, adverbs, adpositions, cases… but also by
combinations of these elements, that is to say, spatial description is not always localised
on one single lexical item but distributed alongside several words (Sinha and Kuteva,
1995). For example, the Basque ablative means ‘through’ only if it is accompanied by a
transversable ground (e.g. door), otherwise it means ‘source’. The situation becomes even
more complicated if one tries to describe languages where some elements do not exactly
fit into traditional linguistic categories such as the ‘category of associated-motion’ in
Arrernte (Koch, 1984; Wilkins, 1991, 2004), and almost impossible when some elements
are classified differently depending on the linguistic framework one works in as is the
case for the distinction between case and postposition in structuralism, generativism,
and functionalism (cf. Agud, 1980 and Blake, 2001).

251
252 LANGUAGE, COGNITION AND SPACE

In order to avoid this terminology problem, I adopt the cover term topological
relational marker proposed by Levinson et al. (2003: 486) and use it for all the ‘various
form classes involved in coding topological relations’. In this section, I analyse some of
the most commonly used topological relational markers in Basque: spatial cases, spatial
nouns, and motion verbs.

Spational cases in Basque

There are five spatial cases in Basque (see Ibarretxe-Antuñano, 2001, 2004a, for a detailed
account):

• The locative case (-n) is one of the most productive cases in the Basque case
system. Its basic meaning is ‘location’ in space (‘in’, ‘on’, ‘at’), as in kale-eta-n
[street-pl-loc]1 ‘in the streets’, mahai-an [table-loc.sg] ‘on the table’, etxe-an
[house-loc.sg] ‘at home’. Sometimes it also expresses motion (‘into’), as in
geltoki-an sartu [station-loc.sg enter] ‘to go into the station’.

• The ablative case (-ti(k)) is usually defined as expressing the ‘source of

motion’. For example, etxe-tik [house-abl.sg] means ‘from the house’. In
specific contexts, the ablative can also convey the meaning ‘through’ as in
leiho-tik [window-abl.sg] ‘through the window’.

• The allative case (-ra(t)) expresses the ‘goal of motion’ in the domain of
space, as in etxe-ra [house-all.sg] ‘to the house’.

• The goal, destinative or terminative allative (-raino) conveys the meaning

‘up to’ in the spatial domain, as in etxe-raino [house-ter.sg] ‘up to the
house’. It indicates a telic motion event, that is, the trajector reaches its final
destination.

• The directional allative (-rantz, -runtz, -rontz) indicates the notion of

‘towards’ in the spatial domain as in etxe-rantz [house-dir.sg] ‘towards home’.
This spatial case profiles the directionality of the motion event. The trajec-
tor moves towards a specific destination but it is not specified whether the
trajector reaches or not the place towards which it moves.

Within the Basque case system, spatial cases form a special group, not only because
they share a common reference to space, but also because they behave morphologically
differently from the other Basque cases. Their main properties are the following: (i) they
are of direct relevance to the distinction between animate and inanimate head nouns,
(ii) they lack the article -a in the definite singular form, and (iii) they have the infix –(e)
ta in non-singular inanimate NPs.
 STATIC TOPOLOGICAL RELATIONS IN BASQUE 253

Apart from the five spatial cases, two other cases have been used for spatial descrip-
tion in our elicited data. The adnominal marker, also called locative genitive (-ko) as in
Euskal Herri-ko-a [basque country-adn.sg-abs.sg] ‘from the Basque Country’, and the
dative (-i) as in Lakioa kandela-ri lotuta dago [ribbon.abs.sg candle-dat.sg tie.pple is.3sg]
‘the ribbon is tied to the candle’. The dative argument when used with verbs of motion
such as etorri ‘come’, joan ‘go’, and ibili ‘walk’ usually refers to a goal as in na-tor-ki-zu
[1sg.abs-come.stem-dative-2sg.dat] ‘I am coming to you’.

Spatial nouns

Basque has more than thirty or more spatial nouns2 that specify even more finely spatial
relations between figure and ground. Table 1 summarises some of the most widely used.
All these spatial nouns follow the same structure: the ground they belong to usually
takes the genitive case, which can be dropped, and, to a lesser extent, the absolutive,
ablative, dative or instrumental cases, and, the spatial noun in turn takes any of the five
spatial cases. For example, mahai-aren gain-etik [table-gen top-abl] ‘from the top of the
table’ or mendi-tik zehar [mountain-abl through] ‘through / over the mountain’. In our
elicited data, almost all spatial nouns are inflected in the locative case.
Table 1. Spatial nouns in Basque

Case for Spatial

Meaning Example
Ground noun
(Gen) Aitzin ‘front’ (eastern) Eliza(ren) aitzin-era ‘to the front of the church’
(Gen) Albo ‘side’ Zuhaitz(aren) albo-an ‘next to the tree’
Ikastola(ren) aldamen-erantz ‘towards near the
(Gen) Aldamen ‘side’
school’
(Gen) Alde ‘side’ Etxe(aren) alde-tik ‘from near the house’
(Gen) Arte ‘space between, among’ Arrok(en) arte-tik ‘from between the rocks’
Abl At ‘outside’ Etxe-tik at ‘outside from the house’
Abl, Instr Ate ‘door’ Eliza-tik ate-an ‘outside the church’
(Gen) Atze ‘back’ Etxe(aren) atze-tik ‘from the back of the house’
(Gen) Aurre ‘front’ Eliza(ren) aurre-an ‘in front of the church’
(Gen) Azpi ‘bottom, lower part’ Mahai(aren) azpian ‘under the table’
(Gen) Barne ‘interior, inside’ Etxe(aren) barne-tik ‘from inside the house’
‘interior, inside; bottom, lower Eliza(ren) barren-ean inside the house’
(Gen), Loc Barren
part’ Oihan-ean barren-a ‘through the forest’
(Gen) Barru ‘interior, inside’ Etxe(aren) barru-ra ‘to the interior of the house’
(Gen), Loc, Mendi-an behe-ra ‘to the lower part along the
Behe ‘bottom, ground, lower part’
Abl, Ø mountain’
(Gen), All Buru ‘centre’ ‘extremity’ Kale(aren) buru-an ‘at the end of the street’
(Gen) Erdi ‘middle, centre’ Eliza(ren) erdi-tik ‘from the middle of the church’
(Gen) Gain ‘top, upper part’ Mahai(aren) gain-era ‘to the top of the table’
Loc Gaindi ‘through’ Mendi-an gaindi ‘through the mountain’
(Gen) Gibel ‘back’ (eastern) Etxe(aren) gibel-etik ‘from the back of the house’
254 LANGUAGE, COGNITION AND SPACE

Case for Spatial

Meaning Example
Ground noun
(Gen), Loc,
Goi ‘top’ Etxe-tik go-ra ‘from the house to the top’
Abl, Ø
(Gen) Inguru ‘vicinity’ Eliza(ren) inguru-an ‘in the vicinity of the church’
Abl, Instr Kanpo ‘outside, exterior’ Etxe-tik kanpo ‘outside the house’
(Gen), Dat Kontra ‘against’ Horma-ri kontra ‘against the wall’
Abl Landa ‘field’ Hiri-tik landa ‘outside the city’
(Gen) Ondo ‘side’ Ikastola(ren) ondo-raino ‘up to near the school’
(Gen) Oste ‘back’ Eliza(ren) oste-an ‘at the back of the church’
(Gen) Pare ‘opposite side’ Etxe(aren) pare-an ‘across from the house’
(Gen) Pe ‘lower part, below, under’ Zuhaitz bat(en) pe-an ‘below the tree’
(Gen) Saihets ‘side’ Ama(ren) saihets-ean ‘next to the mother’
Loc, Abl Zehar ‘through, across’ Mendi-an zehar ‘through the mountain’

Motion verbs

In Basque, there are more than 2000 different types of motion verbs (Ibarretxe-Antuñano,
in prep.). A possible explanation for this rich repertoire, which contradicts the general
prediction about the number of this type of verbs in verb-framed languages (Slobin,
1996), can be found in the various morphological strategies that Basque employs to
create its motion verb lexicon (Ibarretxe-Antuñano, in press). Let us briefly explain
some of these strategies:
Simple verbs. These can be classified into four classes according to their perfec-
tive participle3 (Hualde and Ortiz de Urbina, 2003: ch. 3.5; Trask, 1997: 103):
(i) verbs in –i such as iritsi ‘arrive’, (ii) verbs in –tu such as jarraitu ‘follow’. This
is the only suffix that can be used in borrowings from other languages such as
ailegatu ‘arrive’ and bueltatu ‘return’, (iii) verbs in –n such as joan ‘go’, and (iv)
verbs with no suffix such as igo ‘ascend’.
Derived verbs. There are two categories: (i) verbs formed from nouns and
adjectives plus the past participle suffix in –tu (-du after a nasal or lateral) as in
zuzen-du [straight-suf] ‘head, set off ’, and (ii) verbs formed from (spatial) nouns
inflected in the allative singular plus the past participle suffix –tu as in lurre-ra-tu
[ground-all-suf] ‘go down’, aurre-ra-tu [front-all-suf] ‘go forward’. In these cases
the verb always means ‘go/bring to (spatial) noun’ (Hualde, 2003a: 347).
Compound verbs.4 There are two categories: (i) uninflected spatial noun plus the
verb egin ‘make, do’ as in alde egin [side make] ‘leave’, and (ii) inflected spatial
noun, usually in the allative or directional allative, plus the verb egin ‘make, do’
as in eskuma-rantz egin [right-dir.all make] ‘go right’, behe-ra egin [below-all
make] ‘go down’.
 STATIC TOPOLOGICAL RELATIONS IN BASQUE 255

The possibility of using these strategies for conveying motion verbs implies that the
lexicon is very rich. For instance, if we wanted to say ‘go out’ in Basque, the lexicon would
give us the opportunity to choose among four different possibilities: atera, irten, kanpo-ra
egin, and kanpo-ra-tu, plus a construction with the spatial noun (kanpo ‘outside’) with
the allative and the verb joan ‘go’, i.e. kanpo-ra joan. This means that for the same motion
description we can use a wide variety of choices: compound verbs, derived verbs, pairs
of synonyms which belong to different verb classes, such as iritsi and heldu ‘arrive’, even
loans from other surrounding languages such as ailegatu and arribatu ‘arrive’ from
Spanish llegar and French arriver respectively.
In the specific subcase of locative verbs, Basque is one of those languages that only
offers a small set of locative/posture verbs (Ameka and Levinson, 2007). The static verb
par excellence is egon ‘static be’. In standard Basque, there is a distinction between egon
‘static be’ as in etxe-an dago [house-loc is.3sg] ‘s/he at home’ and izan ‘existential be’ as
in hizkuntzalari-a da [linguist-abs is.3sg], but in eastern dialects there is only one basic
copular verb, izan, which covers both functions (etxe-an / hizkuntzalari-a da). Basque
also has a set of posture verbs such as eseri ‘sit’, zutitu, jagi ‘stand’, zintzilikatu, eskegi
‘hang’, etzan ‘lie down’, makurtu ‘crouch’… These verbs are mostly used for the descrip-
tion of an active change of posture as in eseri da [sit-perf aux.2sg] ‘s/he sits down’ and
Mikel-ek soka zuhaitz bat-etik zintzilikatu du [Mikel-erg rope.abs tree one-abl hang.perf
aux.3sg.abs.3sg.erg] ‘Michael hung the rope from the tree’. For the description of static
postures, these verbs are always used in the participle form (-ta, -rik) with the verb egon
as in eseri-ta dago [sit-pple is.3sg] ‘he’s sitting’ and soka zuhaitz bat-etik zitzilikatu-ta
dago [rope.abs tree one-abl hang.pple is.3sg].

3 Topological relational markers and their usage

Data elicitation

The elicitation tool I used in this study is called the Topological Relation Picture Series and
was developed at the Max Planck Institute for Psycholinguistics in Nijmegen (Pederson,
Wilkins, Bowerman, 1993; see also Bowerman, 1996). This tool is a booklet that consists
of seventy-one line-drawings that depict different topological spatial relations equivalent
to English prepositions in, on, at, under, near, in the middle of…and such like. Each
drawing has a figure coloured yellow and marked with an arrow, and a ground object
in black and white as shown in Figure 1.

Figure 1. Example of a drawing from the Topological Relation Picture Series

256 LANGUAGE, COGNITION AND SPACE

The procedure is very simple. The researcher asks the informant to answer a question
of the form: ‘Where is the [Figure] (with respect to the [Ground])?’. Informants provide
descriptions of the drawings, including the spatial relational marker that would most
naturally be used to describe the relation depicted. If informants provide more than
one answer these are also noted down as second or third choices. Twenty-six Basque
native speakers participated in this study.

Which topological relational markers does Basque use more frequently?

The topological relational marker par excellence in Basque is the locative case –n. The
locative is usually found with the verb egon ‘static be’ (1), with other spatial nouns (2):
and with participles (3).

(1) Sagarra ontzi-an dago [2]5

apple.abs bowl-loc is.3sg

(2) Sagarra ontzi-aren barru-an dago [2]

apple.abs bowl-gen inside-loc is.3sg

(3) Tirita hanka-n jarri-ta dago [35]

bandaid.abs leg-loc put.pple is.3sg

The Basque locative is a good example of what Feist (2004, 2008) calls a ‘General spatial
term’. Similarly to Turkish –da, Ewe le and Indonesian –di, the Basque locative occurs
in all spatial descriptions, with or without more specific terms, and with no specific
information about the location of the figure. The locative only lexicalises the semantic
component location, i.e. position in space, leaving for the other elements of the spatial
description (semantic content of figure, ground, verbs…) the details about the exact
spatial configuration. For example, in boat on water [11] the locative only tells us that the
figure boat is located in the area of interaction of the ground water. It is only thanks to
the characteristics of boats and water and their canonical relationship that the inference
that the boat is on the water and not inside the water arises.
As far as its usage is concerned, the locative (alone, with no other spatial noun) is
the most widely used topological relational marker in our data. Informants employ the
locative case as their first choice in 38 drawings (54%) and as their second choice in
26 drawings (37%). There are only seven drawings (10%) where the locative case is not
used on its own, although it does appear in the spatial noun.
Spatial nouns are also used quite frequently for the description of these drawings;
more concretely, they are the first choice in 29 drawings (40%) and appear as second and
third choice in virtually all the pictures. In our data, spatial nouns fulfil two different
functions. On the one hand, they cover spatial scenes that cannot be or are not usually
described by the locative case as in cloud over mountain [36]. On the other, they give
more specific information about the spatial scene than the locative case. Here, several
 STATIC TOPOLOGICAL RELATIONS IN BASQUE 257

subcases can be distinguished. Spatial nouns are used when the informant wants to
be more specific about the scene s/he is describing. For example, in drawings such as
book on self [8] and hat on head [5], informants tend to use the spatial noun gain-ean
[top-loc] ‘on top of ’, even though the locative could be enough for the description of such
spatial configurations. The canonical positions of these figures – books and hats – with
respect to their grounds do not really allow for other type of interpretation than ‘be on
top’. Spatial nouns are also used when the locative is too uninformative; that is to say,
when the locative offers only very general and ambiguous spatial information that gives
rise to different spatial configurations. This is the case of drawings like owl in tree hole
[67] and lamp over table [13], where informants prefer to mention as their first choices
spatial nouns such as barru-an [inside-loc] and gain-ean [top-loc] respectively. Finally,
they are also used for the description of a common spatial type scene. These are cases
where the spatial relationship between figure and ground is prototypical, that is, they
are located in a way that is expected by the informant due to the characteristics of both
elements. Drawings such as cup on table [1], cat under table [31] are good examples of
this subcase. Here, informants consistently choose a spatial noun like gain-ean [top-loc]
and azpi-an [below-loc] instead of the locative alone.
With respect to the utilisation of motion verbs, the most widely used verb is egon
‘static be’. The semantic information of this locative verb is very poor and general, and
thanks to the tolerance of Basque for verb omissions (Ibarretxe-Antuñano, 2004b, in
press), the verb egon is often elided. As mentioned in the previous section, Basque also
has several posture verbs, but these are mainly used as adverbial participles accompa-
nying the verb egon. Adverbial participles such as itsatsi-ta, inke-ta, ‘stuck’, eskegi-ta,
zintzil-ik ‘hung’, eutsi-rik ‘clung’, ipini-ta, jarri-ta ‘put’, and lotu-ta ‘tied, joined’ are
basically used to specify the information in the locative (or ablative) and/or spatial noun.
For instance, in drawings like coat on hook [9] and clothes on line [37], the majority of
informants – 22 (84,6%) in the former and 21 (80,7%) in the latter – mention ‘hung’ in
their descriptions. It is interesting to point out that in a couple of drawings, informants
seem to interpret and describe the scenes dynamically by means of motion verbs such
as ingururatu ‘go around’ in fence around the house [15], and zeharkatu ‘cross’ in arrow
through apple [30]. These dynamic descriptions give rise to a very important question:
Can location be conceptualised both as dynamic and static? I will come back to this
issue later in the discussion.

4 Spatial information

In this section, I will discuss the type of spatial information that Basque topological rela-
tional markers pay attention to. Authors such as Herskovits (1986), Jackendoff (1993),
Miller and Johnson-Laird (1976), Vandeloise (1991) have proposed different labels for
spatial components. Here I will follow the terms commonly used in other work related
to the Topological Relational Picture Series produced by the Language and Cognition
group at the MPI (see Bowerman, 1996; Bowerman and Choi, 2001; Levinson et al.,
2003). Spatial components are written in small caps. For each spatial component, I will
258 LANGUAGE, COGNITION AND SPACE

mention the topological relational markers (TRM) related to them and the drawings
that better exemplify these spatial configurations. The complete distribution of these
relational markers and the pictures – extensional maps – can be found in the Appendix.
Here Venn diagrams are used to show how Basque topological relational markers group
certain scenes together. Six are the semantic components covered by Basque TRMs in
our elicited data:

location
The locative case is the prototypical vehicle for the expression of location. As said in
the previous section, the locative case by itself does not offer any further specification
about the spatial configuration. This is inferred from the figure and ground, which can be
one-, two-, or three-dimensional. It can be used in all sorts of situations with or without
containment, with or without contact-and-support as shown in the description of
drawings such as rabbit in cage [54], boat on water [11], and ribbon around candle [4].

verticality
This spatial component is present in the spatial nouns gain-ean [top-loc] and azpi-an
[bottom-loc]. Contact-and-support information is not relevant for these spatial
nouns since they are used both in contact-and-support situations as in the drawing cup
on table [1] and in non-contact-and-support cases as in lamp over table [13].

containment (Enclosure)
Barru-an [inside/interior-loc] is the spatial noun that pays attention to this spatial
component. The distinction between partial and total inclusion is not relevant for
this spatial noun as its appearance in the drawings box in bag [14] and apple in bowl [2]
demonstrates. A three-dimensional containment is not necessary for barruan; this
spatial noun appears both in a two-dimensional scene such as dog in basket [47] and
in a three-dimensional scene such as apple in bowl [2]. Although both barruan and
barnean refer to the interior of a given ground, the conceptualisation of interior does
not seem to be exactly the same. Barru is usually applied to grounds that happen to
have an interior as part of their intrinsic configuration as in the case of dog in basket
[47]. Barne, on the other hand, only occurs in situations where the ground itself is the
one that delimits or creates an interior as in the drawing house in fence [60].

horizontality
This spatial element is present in the spatial nouns atze-an [back-loc] and aurre-an
[front-loc] as represented in drawings such as tree in front of church [49] and boy behind
the chair [64]. It is important to point out that these spatial nouns not only involve a
topological concept – horizontality – but also projective meanings, i.e. they specify
an angle in relation to a ground and project a search-domain for the referent from
that ground. Atzean and aurrean fit into the intrinsic frame of reference (FOR). This
coordinate system is object-centred and its coordinates are determined by the conceptual
characteristics of the ground object: its shape, canonical orientation and so on (see
Levinson, 2003: ch.2, for a complete discussion of frames of reference).
 STATIC TOPOLOGICAL RELATIONS IN BASQUE 259

distance
This is represented in our data by spatial nouns like ondo-an [side-loc], albo-an [side-
loc], and alde bat-ean [side one-loc]. All of them denote proximity as in the drawing
boy next to fire [38]. This spatial element can be complemented with relative frame of
reference information. The relative coordinate system ‘presupposes a viewpoint, and a
figure and ground distinct from the viewpoint […] and utilizes coordinates fixed on the
viewpoint to assign directions to figure and ground’ (Levinson, 2003: 43). For example,
informants use eskubi-tara [right-all] in the drawing dog near kennel [6] and ezkerr-
etara [left-all] in boy next to fire (38). Although more research is needed, the elicited
data suggest that aldean is preferred in cases where the figure is not just located near
the ground but also constitutes an integral part of it as in drawings like tree on hillside
[17] and strap on purse [66], whereas ondoan and alboan do not necessarily entail this
identity connection.

point-to-point attachment
This spatial information is lexicalised by means of the ablative case together with a
posture verb as adverbial participle as in –tik zintzilik / eskegita [-abl hang.pple], and
-tik eutsirik [-abl hold.pple]. Drawings like apple on twig [27] and lamp on ceiling [63]
are good examples of this spatial configuration.

5 Conceptualisation of space

In this section, I would like to touch on two issues that seem to play a role in how
Basque conceptualises space: the opposition between dynamicity and stativity in spatial
configurations and the role of agentivity.
All the drawings used as an elicitation tool in this study depict static topological
scenes; that is, they are static representations of a figure located at some position in
relation to a given ground. No information is given about the procedure (movement,
change of position…) followed by the figure in order to reach that location or the
agent that caused that state of affairs. However, it seems that informants consistently
conceptualise some of these scenes as dynamic and some others with an implicit agent.
Speakers have other options to describe these scenes statically, but they choose to
include this type of information. At this stage, I can only say that these two elements
are present in the conceptualisation of space in Basque but I would like to argue that
dynamicity and agentivity are two intrinsic components of space in Basque. Further
research will tell us whether these results are true or just a coincidence. Let us briefly
look at some examples.

Dynamicity vs. stativity

If all these scenes are static, it is only natural to expect static descriptions of these draw-
ings. As discussed in previous sections, the typical formula in Basque is: (spatial noun)
+ locative case + (adverbial participial of posture verb) + egon ‘static be’, where elements
260 LANGUAGE, COGNITION AND SPACE

in brackets can be omitted. This formula is employed in the description of most of the
drawings, but there are some scenes where speakers used more dynamic descriptions.
This dynamicity is often obtained by the use of a motion verb instead of the static
verb. A good example is the picture arrow through apple [30], where speakers (39,3%)
tend to use a motion verb such as zeharkatu and gurutzatu ‘cross’ instead of the static
description sagarran sartuta dago [apple.loc enter.pple is.3sg]. The choice of a specific
spatial case seems to be relevant too. In the drawing insects on wall [52], some informants
prefer the ablative case, which implies more dynamicity, over the locative. A possible
explanation is that these speakers distinguish between things that are statically on the
wall and cannot move, and things like the insects that happen to be on the wall – attached
to the wall – but that can change places.
Spatial nouns are also a good example for the distinction between dynamicity and
stativity because some seem to be intrinsically dynamic such as zehar ‘through, across’
and barren ‘interior, inside; lower part’, and some others intrinsically static such as gain
‘top, upper part’. As mentioned above, this is an area that requires further research but
I would like to bring forward a case that seems to support this distinction between
static and dynamic spatial nouns: the pair barru and barren. Both spatial nouns refer
to ‘interior, inside’, but I would like to argue that their conceptualisation is different on
the basis of dynamicity. Compare these two examples.

(4) Kaia barru-an

port inside-loc

(5) Kaia barren-ean

port inside-loc

Whereas barru-an in (4) only entails location at that port, barren-ean means not only
that the boat is located at the port but also that the boat moved there. The result is the
same, the boat is at the port, but their conceptualisation is different as represented in
Figure 2.

Result: inside (4) barru

(5) barren

Figure 2. Stativity and dynamicity in barru and barren

 STATIC TOPOLOGICAL RELATIONS IN BASQUE 261

In other words, if barru-an is used, the final state – location at the port – is profiled. It is
assumed that the boat was at point a at time x – although this is information is neither
present nor relevant for barru-an – but what barru-an tells us is that the boat is at the
port at this time. If barren-ean is used, there is a double profiling: the final state – location
at the port – and the process followed to reach that final destination – the boat moves
from point a to the port.
In the elicited data, there is a very illustrative example that can shed some more
light on the dynamic conceptualisation of barren. One of the informants uttered the
following sentence for the description of insects on wall [52]:

(6) Xomorrok paret-etik barren-a dabiltz

bug.abs.pl wall.abl.pl inside-all move.3.pl

In (6), this informant uses the spatial noun barren with the allative case, which is usually
translated as ‘through’. Here, the idea is that the insects are located on the wall, got there
because they moved from some other place, and are still moving or creeping along.

Agentivity

All the pictures in the data reproduce static topological relations where the figure is
located in relation to some ground. In most cases, informants describe these pictures
without mentioning the agent that causes that figure to be located at that position.
Although one can assume that in drawing [5] hat on head, for example, the man put on
his hat or that in drawing [31] cat under table the cat himself moved to that position,
this information is neither represented in the drawings nor relevant for the speakers,
and therefore, it is only natural that informants omit this type of information. There
is, however, a group of drawings where informants do mention an implicit agent. For
example, in picture [20] balloon on stick, more than half of the speakers (55%) use the
dative instead of the locative as illustrated in (7) and (8).

(7) Globoa makila-ri lotuta dago

balloon.abs stick-dat tie.pple is.3sg

(8) Globoa makilaren punta-n dago

balloon.abs stick-gen tip-loc is.3sg

In these cases, the dative implicitly shows that somebody had to tie the balloon to
the stick, whereas the locative only tells us that the balloon is located there. The
use of the dative in some descriptions can be explained if we take into account the
relationship between figure and ground. Although more research is needed, I would
like to argue that the dative is used in those cases where the topological relationship
is not natural, either because the figure is not in a typical position or because the
figure cannot be located at the ground on its own without the help of an external
262 LANGUAGE, COGNITION AND SPACE

agent. This would help us to understand why apple on tree [27] is described with the
locative (zuhaitz-ean [tree-loc]) or the ablative (zuhaitz-etik zintzilik [tree-abl hang.
pple]) and balloon on stick [20], ribbon around candle [4], and clothespin on line [33]
with the dative (makila-ri lotuta [stick-dat tie.pple], kandela-ri lotuta [candle-dat tie.
pple], soka-ri lotuta [rope-dat tie.pple]).

6 Conclusions

In this paper, I have offered an overview of the main linguistic means used in Basque
for the lexicalization of space. Three are the main types of topological relational mark-
ers: spatial cases, spatial nouns, and motion verbs. Based on elicited data from the
Topological Relation Picture Series, I have shown that the structure that Basque speakers
mostly prefer for the description of static topological relations is the locative case and
the static verb egon ‘be’. Therefore, Basque can be classified as one of those languages
with a general spatial term and a small set of locative/posture verbs – mainly used as
adverbial participles. Speakers opt for spatial nouns in cases when the locative case
cannot be utilised for the description of a spatial configuration or when it is semantically
too uninformative and vague. To a lesser extent, informants also use the dative case
and path motion verbs. This usage is restricted to very specific spatial scenes where an
agent is implicitly needed and where the location is conceptualised as dynamic. With
regard to the spatial information provided by these topological relational markers,
I have found six main spatial components: location, verticality, containment,
horizontality (with intrinsic FOR information), distance-proximity (with relative
FOR information), point-to-point attachment.
There is still a lot of work to do in order to get to grips on the conceptualisation of
space in Basque. In the last part of this paper, I have pointed out that elements such as
dynamicity and agentivity are to be taken into account since they seem to be present in
the linguistic characterisations of some of the space scenes used as stimuli. The spatial
noun barren-ean has the same meaning as barru-an ‘inside’, but its conceptualisation is
different, barren-ean profiles not only final location at some point but also the dynamic
procedure followed by the figure in reaching that destination. Another area that deserves
more attention is the contrastive study among topological relational markers with a
similar function. Spatial nouns like ondo, alde, and albo are all used for ‘near’ but we
need to know whether they can be employed in the same contexts or whether there are
any subtle differences among them. Another interesting issue for further research is the
study of hierarchical relations between spatial nouns. Levinson et al (2003: 489) have
found that languages with large sets of spatial adpositions are bound to be arranged
in taxonomic trees where ‘subordinate terms are more specific: they have, if one likes,
additional features missing from their superordinate or more general terms’. Basque
is one of those languages with numerous spatial resources, and hyponymic chains
like locative ‘location’ → barru ‘any interior, inside part’ → barne ‘interior, inside part
delimited by ground’ are worth investigating.
 STATIC TOPOLOGICAL RELATIONS IN BASQUE 263

Notes
1 List of abbreviated morphemes: abl Ablative; abs Absolutive; adn Adnominal; all Alla-
tive; ben Benefactive; dat Dative; det Determiner; dir Directional Allative; erg Ergative;
gen Genitive; ger Gerund; ter Terminative Allative; ind Indicative; indf Indefi nite;
inst Instrumental; loc Locative; perf Perfective; pl Plural; pple Participle; sg Singular.
Morphemes are separated with a hyphen only in those cases when they are relevant for
the discussion, otherwise they will be written together.
2 There is a great deal of discussion about the categorial status of these space elements in
Basque grammars. I use the general theory-free term of spatial noun to cover what other
authors call locative nouns (de Rijk, 1990), postpositions (Euskaltzaindia, 1991; Hualde
y Ortiz de Urbina, 2003), adverbs (Bostak Bat, 1996) and locative cases (Laka, n.d.).
3 The citation form of Basque verbs in most dialects is the perfective participle. Basque
is a language in which the majority of finite verb forms are largely analytical – also
called periphrastic in the Basque grammar tradition – and as such, verb expressions are
formed by a participle, which carries the lexical meaning and aspect information, and
an auxiliary, which contains information about tense, mood and argument structure
(cf. Hualde and Ortiz de Urbina, 2003: chapter 3.5, for a more detailed description of
verbs in Basque).
4 Syntactically these are called ‘complex predicates’ in contrast with simple verbs and
derived verbs which are ‘simplex predicates’ (Etxepare, 2003).
5 Examples taken from the data will always show the number of the corresponding draw-
ing in square brackets.

References
Agud, A. (1980) Historia y Teoría de los Casos. Madrid: Gredos.
Ameka, F. K. and Levinson, S. C. (2007) Introduction: The typology and semantics of
locative predicates. Posturals, positionals, and other beasts. Linguistics 45(5–6):
847–871.
Blake, B.J. (2001) Case. Cambridge: Cambridge University Press.
Bostak Bat. (1996) Diccionario Hiru Mila Hiztegia. Bilbao: Elkar.
Bowerman, M. (1996) Learning how to structure space for language: A cross-
linguistic perspective. In P. Bloom, Peterson, M., Nadel, L. and Garrett, M. (eds)
Language and Space 385–436. Cambridge, MA: MIT Press.
Bowerman, M. and Choi, S. (2001) Shaping meanings for language: Universal
and language-specific in the acquisition of spatial semantic categories. In
M. Bowerman and Levinson, S. (eds) Language Acquisition and Conceptual
Development 475–511. Cambridge: Cambridge University Press.
Etxepare, R. (2003) Valency and argument structure in the Basque verb. In J. I.
Hualde and Ortiz de Urbina, J. (eds) A Grammar of Basque 363–425. Amsterdam
and Philadelphia: John Benjamins.
Euskaltzaindia. (1991) Euskal Gramatika: Lehen Urratsak I-II. Iruñea: Euskaltzaindia.
Feist, M.I. (2004) Talking about space: A cross-linguistic perspective. In K.D. Forbus,
Getner D. and Regier, T. (eds) Proceedings of the 26th Annual Meeting of the
Cognitive Science Society 375–380. Mahwah, NJ: Lawrence Erlbaum Associates.
264 LANGUAGE, COGNITION AND SPACE

Feist, M.I. (2008) Space between languages. Cognitive Science 32(7): 1177–1199.
Herskovits, A. (1986) Language and Spatial Cognition: An Interdisciplinary Study of
the Prepositions in English. Cambridge: Cambridge University Press.
Hualde, J.I. (2003) Derivation. In J. I. Hualde and Ortiz de Urbina, J. (eds) A
Grammar of Basque 328–351. Amsterdam and Philadelphia: John Benjamins.
Hualde, J.I. and Ortiz de Urbina, J. (2003) A Grammar of Basque. Amsterdam and
Philadelphia: John Benjamins.
Ibarretxe-Antuñano, I. (2001) An overview of Basque locational cases: Old descrip-
tions, new approaches. International Computer Science Institute Technical Report
No. 01–006. University of California, Berkeley.
Ibarretxe-Antuñano, I. (2004a) Polysemy in Basque locational cases. Belgian Journal
of Linguistics 18: 271–298.
Ibarretxe-Antuñano, I. (2004b) Dicotomías frente a continuos en la lexicalización de
los eventos de movimiento. Revista de Lingüística Española 34(2): 481–510.
Ibarretxe-Antuñano, I. (in press) Basque: Going beyond verb-framed typology.
Linguistic Typology.
Ibarretxe-Antuñano, I. (In preparation) Towards a typological classification of
motion verbs in Basque: Their structure and meaning. Unpublished Manuscript.
Universidad de Zaragoza.
Jackendoff, J. (1993) Semantics and Cognition. Cambridge, MA: MIT Press.
Koch, H. (1984) The category of ‘associated motion’ in Kaytej. Language in Central
Australia 1: 23–34.
Laka, I. (n.d.) A Brief Grammar of Basque. Retrieved on June 2005 from http://www.
ehu.es/grammar/index.htm.
Landau, B. and Jackendoff, R. (1993) ‘What’ and ‘where’ in spatial language and
spatial cognition. Behavioral and Brain Sciences 16: 217–238.
Levinson, S.C. (2003) Space in Language and Cognition. Explorations in Cognitive
Diversity. Cambridge: Cambridge University Press.
Levinson, S.C., Meira, S. and Language and Cognition Group (MPI) (2003) ‘Natural
concepts’ in the spatial topological domain – Adpositional meanings in crosslin-
guistic perspective: An exercise in semantic typology. Language 79(3): 486–516.
Majid, A., Bowerman, M., Kita, S., Haun, D.b.M. and Levinson, S.C. (2004) Can
language restructure cognition? The case for space. Trends in Cognitive Sciences
II(3): 108–114.
Miller, G. and Johnson-Laird, P.N. (1979) Language and Perception. Cambridge, MA:
Harvard University Press.
Pederson, E., Wilkins, D. and Bowerman, M. (1993) Static topological relations. MPI
Fieldmanual.
Pederson, E., Danzinger,E., Wilkins, D., Levinson, S.C., Kita, S. and Senft, G. (1998)
Semantic typology and spatial conceptualisation. Language 74(3): 557–589.
Rijk, R.P.G.de (1999) Location nouns in Standard Basque. Anuario del Seminario de
Filología Vasca ‘Julio de Urquijo’ XXVI(1): 3–20.
Sinha, C. and Kuteva, T. (1995) Distributed spatial semantics. Nordic Journal of
Linguistics 18: 167–199.
 STATIC TOPOLOGICAL RELATIONS IN BASQUE 265

Slobin, D. I. (1996) Two ways to travel: Verbs of motion in English and Spanish. In
M. Shibatani and Thompson, S.A. (eds) Grammatical Constructions. Their Form
and Meaning 195–219. Oxford: Clarendon Press.
Trask, R.L. (1997) The History of Basque. London: Routledge.
Vandeloise, C. (1991) Spatial Prepositions: A Case Study from French. Chicago:
Chicago University Press.
Wilkins, D. (1991) The semantics, pragmatics and diachronic development of ‘associ-
ated motion’ in Mparntwe Arrernte. Buffalo Papers in Linguistics 1: 207–257.
Wilkins, D. (2004) The verbalization of motion events in Arrernte. In S. Strömqvist
and Verhoeven, L. (eds) Relating Events in Narrative: Typological and Contextual
Perspectives 143–157. Hillsdale, NJ: Lawrence Erlbaum.

Appendix: Basque topological relational markers and their distribution

-n (LOCATIVE) erdian barruan barnean

60
19

7 10 41 35 2 54 47

39 14 67
3 25 12 71
32

62

46 21 8 29 1
11 22 40 59

34
61

70
65 5 36 23
18 56 13
26

alde baten/batera aurrean

38

66
gainean
48
68
17 49 43
6

A B L A T I V E kanpo(an...) ondoan alboan

15

69

52
50 42 51
64 atzean

DATIVE
9 4

45
57
20
44 33 55

inguruan
58

63 30
37 27

kontra
24 53
31 16

azpian
11 Taking the Principled Polysemy Model of
spatial particles beyond English: the case of
Russian za
Darya Shakhova and Andrea Tyler

1 Introduction

Motivated by the goals of providing a replicable methodology and a theoretically

grounded model of the polysemy networks of English spatial particles, Tyler and Evans
(e.g. 2001, 2003) developed a model of semantic extension, termed the Principled
Polysemy model. Relying primarily on established principles of language processing,
such as embodied experience, taking multiple perspectives on a scene, and pragmatic
inferencing, the model offered both a method for determining the central sense of a
preposition and a more comprehensive accounting of the meaning extension mecha-
nisms involved in the polysemy networks of English spatial particles.
The model emphasizes universal properties of human cognition, such as knowledge
of real world force dynamics, leading Tyler and Evans (2003) to hypothesize that the
model is likely to be applicable to many languages. However, Tyler and Evans’ analyses
have been based almost solely on English prepositions and the hypothesis concerning
universal application is yet to be tested. A primary purpose of this paper is to begin
to test the universality of the Principled Polysemy model by applying it to one of the
most highly polysemous prepositions in Russian, za. A second purpose is to investigate
how the model might be flexibly augmented when applied to a language whose system
of spatial referencing includes a complex system of case marking, which is lacking in
English.
Tyler and Evans (2001, 2003) noted that, in spite of many years of research, scholars
still disagreed about the appropriate representation of the central meaning of preposi-
tions, how to distinguish independent senses, and how to systematically account for
the many extended meanings associated with each preposition. Their model aimed at
providing a replicable methodology for determining the central sense and distinguishing
independent, extended senses.
Under their analysis, the many meanings associated with a single spatial particle are
represented as a complex category organized around a central sense, or the proto-scene.
The meaning of a spatial particle is thus represented as a motivated, systematic polysemy
network. Tyler and Evans (2003) posited that the central sense of a preposition represents
a particular spatial relation between a focus element, the trajector (TR), a background
or locating element, the landmark (LM), and a functional element (Vandeloise, 1991);
the functional element represents the humanly meaningful consequences of the TR

267
268 LANGUAGE, COGNITION AND SPACE

and LM being in a particular spatial configuration and is understood to be an integral

part of the central sense.
The Principled Polysemy Model accounts for semantic extension through applica-
tion of several established cognitive and linguistic principles such as: taking multiple
perspectives on a scene, including highlighting subparts of a spatial scene; experiential
correlation; knowledge of force dynamics; attention to the role of context in establishing
an interpretation; and distributed meaning, i.e. appropriately attributing the overall
interpretation of a preposition (or any lexical item) within an utterance to the meaning
provided by all the elements of the utterance (including the rich social-cognitive context
in which the utterance occurs).
Tyler and Evans provided a systematic account of many of the most commonly
occurring prepositions in English. However, languages vary widely in how they represent
spatial relations and in the patterns of polysemy associated with those spatial elements.
For instance, Japanese employs a combination of particles and special nouns to indicate
spatial relations. Languages such as Finnish and German employ a combination of
spatial particles and case marking. In order to test the universality of the Principled
Polysemy model, analyses of additional languages, with different grammatical mecha-
nisms for representing spatial relations, is required. The purpose of the present work
is to begin to test the universality of the Principled Polysemy model by applying the
model to the Russian spatial particle, za.1 Za is a particularly good test of the model
because it is highly polysemous. In addition, the analysis of za entails considering how
the Principled Polysemy model applies to a language whose system of spatial referencing
includes a complex system of case marking, which is lacking in English. In this area of the
analysis, we have been greatly influenced by Janda’s (1993, 2000) work on Russian case.

2 The present study

2.1 The data

All the proposed senses are supported by attested uses of za. Our examples come from
the following sources: (1) Ozhegov’s Russian Language Dictionary (1984); (2) online
dictionaries: Dal’ Dictionary (1998), Dictionary ‘Obschaĭa Leksika’; (3) grammar books:
Poltoratsky and Zarechnak (1961), Pulkina and Zakhava-Nekrasova (2000); (4) on-line
corpora: online corpus of the works of Dostoevsky, the National Corpus of the Russian
Language. Approximately 1500 sentences in which za occurred were translated by
Shakhova, a native speaker of Russian and English.

2.2 The problem

The Russian preposition za is associated with a broad range of meanings, e.g. standard
Russian-English dictionaries list as many as 21 different senses including the following
English prepositions: behind, over, outside, beyond, after, for, at, by, and near. To begin to
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 269

get a sense of the challenge of representing the meaning of za, consider a small sample
of its uses and their English translations:

(1) Behind:
Za domom 275a budet vystroen novyĭ supermarket.
[Za building-INST 275a will be built new supermarket-NOM]
The new supermarket will be built behind Building 275a.

(2) Over:
On zhivët za rekoĭ, na ĭugo-vostoke.
[He-NOM lives za river-INST, at south-east-PREP]
He lives over the river, in the south-east.

(3) For:
Fuks zaplatil za svoë osvobozhdenie 15000 dollarov.
[Fuks-NOM paid za his release-ACC 15000 dollars-GEN]
Fuks paid 15,000 dollars for his release.

(4) At:
Sima poslushno sela za pianino.
[Sima-NOM obediently sat down za piano-ACC]
Sima obediently sat down at the piano.

In addition to having a wide range of senses, za interacts with case in a complex way.
Some senses occur with both Instrumental and Accusative Case, others occur with
only one or the other.
To date, no unified analysis has been offered for the many, seemingly unrelated
meanings associated with za nor its interaction with case. The grammar books do offer a
few general rules, but for the most part, the meanings are considered as an arbitrary list.
Dictionaries often attempt to represent the many meanings of za by defining a particular
use in terms of another Russian preposition. The definitions are often misleading in
that they usually fail to provide key elements of contextualized interpretation of za.
Ozhegov’s Russian Language Dictionary, one of the most widely used reference volumes
for Russian language, defines za as okolo, ‘near’ or vokrug ‘around’ as in:

(5) sidet’ za stolom = sidet’ okolo/vokrug stola

‘to sit near/around the table’ (Ozhegov 1984)

However, while za denotes a spatial scene where the TR is positioned in close proximity
to the LM (here the table) and therefore conforms to the notion ‘near’, in this context za
also evokes an additional understanding that the TR is purposefully sitting proximal to
the table. The same is true of a spatial scene involving multiple TRs positioned ‘around’
the table. Thus the full interpretation of sidet’ za stolom includes the notion that the TR
is sitting in order to use the table and prompts for an implicature of the TR’s legs being
270 LANGUAGE, COGNITION AND SPACE

under the table. Such a fine-grained configuration is not captured by the preposition
okolo ‘near’ or vokrug ‘around’. The analysis presented here argues that applying the
Principled Polysemy model, in conjunction with Janda’s analysis of Russian case, allows
us to represent the range of meanings exhibited by za, including fine-grained interpreta-
tions illustrated above, as a systematic, motivated polysemy network.

3 The analysis

3.1 Establishing the proto-scene

Key to the successful analysis of any polysemy network is establishing the appropriate
central sense, or in Tyler and Evans’ terminology, the appropriate proto-scene, from
which all other senses are held to derive. Tyler and Evans (2003) suggest several steps
for establishing the proto-scene. Here we consider only three:
1) Examining the spatial configuration of LM and TR in multiple sentences in
which the spatial particle is used.
2) Examining sentences that use contrasting spatial particles (members of a con-
trast set). Tyler and Evans established that a contrast set can involve aspects of
the spatial configuration of the TR and LM, as in the contrast between English
over versus under. Or a contrast set can involve the variations in the functional
element, as in English over, whose functional element involves proximity or
mutual influence between the TR and LM, versus above, whose functional ele-
ment involves distance or lack of mutual influence between the TR and LM.
3) Frequency in the polysemy network. By this Tyler and Evans mean that, in the
case competing analyses of the central sense, the majority of independent senses
should be derivable from the proto-scene.

3.2 Case marking

Before examining the proto-scene, a few words about Russian case are in order. The
Russian case system is rather complex and scholars have offered many analyses of
just what the various cases mean. We have found Janda’s (1993, 2000) analysis most
convincing. Using a Cognitive Linguistic approach she carried an extensive analysis
of Instrumental and Accusative case, the two cases with which za occurs. Under her
analysis Instrumental primarily denotes stable physical configurations or ‘setting’. Thus,
it tends to contribute a sense of a static scene. In contrast, Accusative generally occurs
in situations depicting motion. Janda’s unique contribution here is the discovery that
the most prototypical meaning of the Accusative involves a destination (which may
involve an extended dimensionality) or the endpoint of an action.
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 271

3.3 The proto-scene

The proto-scene posited for the central sense of za involves a spatial configuration in
which the LM is horizontally oriented away from the TR. In other words, the LM is
conceptualized as having asymmetrical front and back, and the TR is conceptualized as
being in back of the LM. In this respect, za is similar to the English preposition behind,
as the following examples illustrate:

(6) The oriented LM:

(a) Rĭadom so mnoĭu sidel Vanĭa, a za nim Marusĭa.
[Next to I-INST was sitting Vanĭa-NOM, and za him-INST Marusĭa-NOM]
Vania was sitting next to me with Marusia (TR) behind him (LM).
(b) Za domom 275a budet vystroen novyĭ supermarket.
[Za building-INST 275a will be built new supermarket-NOM]
The new supermarket (TR) will be built behind Building 275a (LM).

In each of these examples, the LM is clearly inherently oriented and the TR’s position is
understood as being at the back of or ‘behind’ the LM. A common property of human
perception is to also assign orientation to objects based on gravity, resemblance to
human beings, behavior, or function. The next example serves as an illustration:
(c) Kogda prishël Tom, malysh sprĭatalsĭa za zerkalo.
[When came Tom-NOM kid-NOM hid za mirror–ACC]
When Tom came, the kid hid behind the mirror.

Here, the LM, the mirror, is functionally oriented, with the reflecting surface interpreted
as the front.
While the landmark’s orientation is salient in the proto-scene, the trajector’s orienta-
tion is not specified. Consider the following examples:

(7) (a) Gde-to za nimi vozvyshalsĭa Ėl’brus.

[Somewhere za they-INST was rising Elbrus-NOM]
Elbrus was rising somewhere behind them.
(TR is unoriented.)
(b) Na urokakh, Kolĭa obychno sidit za Sëmoĭ.
[At classes-PREP Kolĭa-NOM usually sits za Sëma–INST]
In classes, Kolĭa sits behind Sëma.
(The TR, can be variably oriented, e.g., Kolĭa can be turned either toward or away
from Sëma.)
272 LANGUAGE, COGNITION AND SPACE

These two examples show that the particular orientation of the TR is inconsequential
for use of za. However, were we to switch the orientation of the LM, a new preposition,
like 'pered ‘in front of ’ would be required.
One of the criteria posited by Tyler and Evans (2003) for deriving the proto-scene
of a spatial particle is examining how this particle interacts with members of a contrast
set. A contrast set is a minimal pair of spatial particles that have complementary func-
tions in dividing up the conceptual space along a particular dimension. For example,
up and down divide the space along the vertical dimension, while in front of and behind
divide up space along the horizontal dimension. Tyler and Evans hypothesize that the
meaning component used to differentiate members of a contrast set is likely to be key
to establishing the primary sense. In terms of spatial configuration, za most clearly
contrasts with pered ‘in front of ’ as in:

(8) (a) Za mnoĭ stoĭal stol.

[Za I-INST stood table-NOM]
The table stood behind me.
(b) Peredo mnoĭ stoĭal stol.
[Peredo me-INST stood table-NOM]
The table stood in front of me.

Unlike za, pered is associated with a relatively limited cluster of senses. Its most basic spa-
tial meaning is a configuration in which a horizontally oriented landmark is ’facing’ the
trajector. Switching between za and pered in Russian results in switching the orientation
of the landmark from facing the trajector (pered) to facing away from the trajector (za).
Za also forms a minimal pair with pozadi in terms of distance along the horizontal
axis. Za indicates a proximal relationship while pozadi indicates a distal one (analogous
to the distinction between English over versus above which divide spatial relations on
the vertical axis).

(9) (a) On stoĭal za mnoĭ i sheptal mne v ukho.

[He-NOM stood za I-INST and whispered me-DAT in ear-ACC]
He stood behind me and whispered in my ear.
(b) ?On stoĭal pozadi menĭa i sheptal mne v ukho.
[He-NOM stood pozadi I-INST and whispered I-DAT in ear-ACC]
He stood behind me and whispered in my ear.
(c) Magda sela obedat' za stol.
[Magda-NOM sat to dine za table-ACC]
Magda sat down at the table to have dinner.
(d) ?Magda sela obedat’ pozadi stola.
[Magda-NOM sat to dine pozadi table-ACC]
Magda sat down at the table to have dinner.
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 273

In these pairs the version containing pozadi strikes native speakers as odd because the
contexts require proximity between the TR and the LM.
Considering all this evidence, we conclude that za’s central scene involves an ori-
ented LM (represented in the diagrams by a ‘nose’; in diagram 1, the ‘nose’ is on the
LM and is pointing away from the TR) and a neutral (i.e. non-oriented) TR which is
positioned at the back of and proximal to the oriented LM:

Diagram 1: The proto-scene for za

Oriented LM facing away from a proximal TR; vantage point is off-stage.

The functional element arising from this configuration is one of proximity or mutual
influence or potential interaction between the TR and LM, analogous to the functional
element posited for English over. Positing this functional element accounts for the
notion of the purposefulness of the TR being positioned proximal to the LM, hence
the interpretation that if a person is located za the table, she is positioned such that she
can purposefully interact with the table.

3.4 The extended network

Tyler and Evans (2003) posited a basic principle for determining whether an extended
sense exists independently. That principle was an independent, extended sense had to
involve either a different spatial configuration than represented by the proto-scene or
an independent non-spatial meaning. In either case, the extended meaning could not
be inferred from the proto-scene as it occurred in the context of the utterance. For
instance, Tyler and Evans posited an independent on-the-other-side sense for English
over on the basis of sentences such as:

(10) The boathouse is over the river from Rosslyn, under the Key Bridge (linguist list, Oct. 2005).

They noted that unless the listener knew that over had an independent on-the-other-side
sense, it would be impossible to appropriately interpret this sentence. In contrast, in a
sentence such as:

(11) The plane flew over the desert.

The sense of motion derives from our understanding of the verb fly, which denotes
motion, and our knowledge of planes. Contra several analyses, e.g. Dewell (1994),
274 LANGUAGE, COGNITION AND SPACE

Tyler and Evans argued that with such sentences any sense of motion is derived from
contextual inferences and should not be taken as evidence of a separate sense which
contains the information +motion.
This analysis largely derives from the principle of distributed meaning. Following
Sandra and Rice (1995), Tyler and Evans (2003) noted that many cognitive analyses
of prepositions had fallen into what they termed the ‘polysemy fallacy’. Essentially, the
polysemy fallacy resulted in positing overly many senses. The polysemy fallacy arises
from assigning aspects of the overall interpretation of an utterance, including regularly
occurring implicatures, as part of the meaning of the preposition. Such analyses fail
to appropriately determine which aspects of the interpretation are contributed by the
various elements in the utterance. Similar to the flying plane example, Tyler and Evans
argued that in an utterance such as, The cat jumped over the wall, the interpretation of the
TR being in motion and following an arc-shaped trajectory comes from an integration
of the central meaning of over (a TR located higher than, but proximal to a LM), the
meaning of the verb jump, and application of our basic understanding of force dynamics,
rather than a special sense of over that includes a trajectory and a TR in motion.
The principle of distributed meaning is particularly important in accurately analyz-
ing Russian prepositions, as Russian prepositions combine with nouns (in LM position)
which occur in various cases. Case marking is in part governed by the preposition
itself. In the case of za, which combines with both Accusative and Instrumental case,
the case marking appears to contribute to the exact interpretation of the scene being
prompted for, for instance, whether the TR in the scene being depicted is interpreted
as being static or involving motion. Consider the following sentences in which the case
on the LM noun varies:

(12) On zhivët za rekoĭ, na ĭugo-vostoke.

[He-NOM lives za river-INST, at south-east-PREP]
He lives over (on the other side of) the river, in the south-east.

In this sentence, the LM carries Instrumental case and the interpretation involves a
static scene. Our understanding of the scene is that there is no TR in motion and no
trajectory involved. Use of Accusative case is unacceptable.

(13) Potom ĭa uekhal za granitsu.

[Then I-NOM left za border-ACC]
Then, I moved abroad.

In contrast, in example (13), the LM is marked with Accusative case. The interpretation
is that the TR is in motion and therefore a trajectory is involved. Use of Instrumental
case is unacceptable. In both these examples, case alone does not establish the scene
as static or involving motion and destination. Certainly the verbs and our background
knowledge contribute to the interpretation, but case is consistent with these meanings.
Under certain analyses, e.g. Lakoff (1987), these two instances of za would constitute
two separate senses, on the grounds that one involves motion while the other does
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 275

not. Under such an analysis, the assumption is that movement (or non-movement) is
part of the meaning of a particular sense of the preposition. Following the principle of
distributed meaning, we posit that these two instances of za constitute one independent
sense which means ‘beyond or on-the-other side’. The exact interpretation of +/- move-
ment is not part of the semantics of the preposition, but rather provided by case marking
and other elements of the sentence (such as the verb uehkal ‘left’, which prompts for
motion, and zhivët ‘lives’, which prompts for a static scene). Thus, in our analysis of the
polysemy network for za, we do not posit separate senses based on case even though
there are differences in the exact understanding of the spatial scene being depicted when
Accusative occurs versus Instrumental.
As mentioned in the introduction, the Principled Polysemy model also identified
a set of cognitive mechanisms by which the central sense of a preposition could be
extended to create independent, distinct senses. These mechanisms are all independ-
ently established in the literature. They include (1) multiple ways of viewing a scene
(Langacker, 1987); (2) knowledge of real world force dynamics (Talmy, 2000); (3) making
pragmatic inferences based on the linguistic prompts and background knowledge (Grice,
1975; Wilson and Anderson, 1986). Tyler and Evans (2003) argued that in most cases
an extended sense could be traced back to an utterance in which the proto-scene (or an
established sense derived from the proto-scene), in conjunction with the context, created
a novel interpretation of the preposition. After repeated uses, such a contextualized use
of a preposition could be established as an independent sense in the network. Once
the meaning was established in the network, the context which originally gave rise to
the new sense would no longer be needed in order for the speakers to interpret the
preposition. The many meanings associated with a preposition were thus represented
as a motivated polysemy network.
Our analysis of the 1500 naturally occurring examples of za revealed that, in addi-
tion to the proto-scene, five independent senses occur with both Instrumental and
Accusative. We term these senses the Shared Network. The following sentences illustrate
each of these senses:

(14) Behind-Deictic orientation:

(a) Misha sprĭatalsĭa za kustikom naprotiv kamysheĭ i
[Misha-NOM hid za (small) bush-INST opposite rushes-GEN and
stal zhdat’.
waited]
Misha hid behind a small bush opposite the rushes and waited.
(b) Alekseĭ s siloĭ shvyrnul v gruzoviki limonku-ACC i
[Alekseĭ-NOM forcefully hurled at trucks-ACC grenade-ACC and
pryglnul za kuchu khvorosta.
leaped za pile-ACC brushwood-GEN]
Alekseĭ forcefully hurled a grenade at the trucks and leaped behind the pile of
brushwood.
276 LANGUAGE, COGNITION AND SPACE

(15) Functional:
(a) Kak i prezhde, vechera oni korotali za chteniem
[As before, evenings-ACC they-NOM whiled away za reading-INST
vslukh.
aloud]
As before, they whiled away their evenings at reading aloud.
(b) On na sekundu pripodnĭal golovu i snova prinĭalsĭa
[He-NOM for second-ACC raised head-ACC and again began/set to
za chtenie.
za reading-ACC]
He raised his head for a second and then went back to reading.

(16) Beyond/on-the-other-side ‘over’:

(a) Ĭa odin provozhal bol’nuĭu starukhu za reku.
[I-NOM alone accompanied ailing old woman-ACC za river-ACC]
I alone accompanied the ailing old woman to the other side of the river.
(b) Tam, za rekoĭ… uzhe zagoralis’ pervye zvëzdy.
[There, za river-INST… already were lighting up first stars-NOM]
There, over the river, first stars were already lighting up.

(17) Focus of attention:2

(a) Mama volnuetsĭa za devochku.
[Mom-NOM is worried za girl-ACC]
Mom is worried for the girl.
(b) Moĭa mat’ obeshchala pereekhat' k nam i smotret' za
[My mother-NOM promised to move in with we-DAT and look za
det’mi.
children-INST]
My mother promised to move in with us and look after the children.

(18) Purpose:
(a) Stol’ ozhestochënnaĭa bo’rba za golosa razvernulas’ eshchë i v
[Such fierce battle-NOM za votes-ACC unfolded also in
predvidenii nizkoĭ ĭavki.
anticipation-PREP low turn-out-GEN]
Such fierce battle for votes unfolded in anticipation of low turn-out.
(b) My zabespokoilis' i poslali za doktorom Brusesom.
[We-NOM started to worry and sent za doctor-INST Bruses-INST]
We started to worry and sent for doctor Bruses.
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 277

Focus of Attention Purpose

Functional Beyond/On-the-Other-Side
Following

Behind-Deictic In-Tandem

Proto-Scene

Diagram 2: The Shared Network

We found a number of additional senses that occur in conjunction with one of the
cases but not the other. In these instances, the data indicate that distinct senses arise as
a result of an extended sense of za as it combines with particular aspects of meaning
associated with either Instrumental or Accusative case.

3.5 Accusative network

Sentences illustrating the senses which occur only with Accusative case:

(19) Exchange:
Fuks zaplatil za svoë osvobozhdenie 15000 dollarov.
[Fuks-NOM paid za his release-ACC 15000 dollars-GEN]
Fuks paid 15000 dollars for his release.

(20) Substitution:
Luchshe gluptsa prinĭat’ za umnogo, chem umnogo za gluptsa.
[Better fool-GEN to take za sage-ACC, then sage-GEN za fool-ACC]
Better to take a fool for a sage, than a sage for a fool.

(21) Contact:
Militsioner vzĭal starushku za ruku i perevël cherez
[Militiaman-NOM took old lady-ACC za hand-ACC and walked across
dorogu.
road-ACC]
The militiaman took the old lady by the hand and walked her across the road.
278 LANGUAGE, COGNITION AND SPACE

(22) Support:
Gosduma progolosovala za prinĭatie zakonoproėkta.
[State Duma-NOM voted za approval-ACC bill-GEN]
The State Duma voted for approval of the bill.

(23) Cause:
Neozhidanno polkovnik razozlilsĭa na sebĭa za svoĭu
[Unexpectedly colonel-NOM became angry at himself-ACC za his
sentimental’nost’.
sentimentality-ACC]
Unexpectedly, the colonel became angry with himself for his sentimentality.

Substitution

Exchange Cause

Focus of Attention Purpose Contact

Functional Beyond/On-the-Other-Side
Following

Behind-Deictic Cluster In-Tandem Cluster

Support

Proto-Scene

Diagram 3 The Accusative Network

3.6 Instrumental Network

We also found many examples of senses which occur only with Instrumental case:

(24) Covering:
Za shumom voln nichego nel’zĭa bylo rasslyshat’.
[Za sound-INST waves-GEN nothing impossible was to hear]
It was impossible to hear anything behind the sound of waves.
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 279

(25) Obstacle:
Za neimeniem sredstv izdatel’stvo ‘Prosveshchenie’
[Za lack-INST resources-GEN publisher-NOM ‘Prosveshchenie’-NOM
prekratilo izdanie zamechatel’noĭ knigi.
stopped publication-ACC wonderful book-GEN]
For lack of financial resources, the publishing company ‘Prosveshchenie’ stopped its
publication of a wonderful book.

(26) Following:
Za mnoĭ, moĭ chitatel’, i tol’ko za mnoĭ, i ĭa
[Za me-INS, my reader-NOM, and only za me-INS, and I-NOM
pokazhu tebe takuĭu lĭubov’!
will show you-DAT such love-ACC]
Follow me, my dear reader, follow me, and I will show you such love!

(27) Sequence:
Odin za drugim na ėkrane krutĭatsĭa fil’my pro
[One za another-INST on screen-PREP play films-NOM about
zalozhnikov.
hostages-PREP]
One after another, films about hostages play on the screen.

(28) Possession:
Brat’ĭa dali za nevestoĭ ogromnoe pridanoe…
[Brothers-NOM gave za bride-INST large dowry]
The brothers gave the bride a large dowry…
Focus of Attention Purpose
Sequence

Obstacle
Functional Beyond
Following

Covering

Behind-Deictic Cluster In-Tandem Cluster

Possession

Proto-Scene

Diagram 4: The Instrumental Network

280 LANGUAGE, COGNITION AND SPACE

3.7 Motivation for specific senses

Because of space limitations, we cannot provide a detailed justification for all the spatial
scenes in the network; thus, we will look at only five. However, we note that our analysis
revealed striking similarities between the spatial scenes for senses of za often translated
by English prepositions such as over and the scenes posited for either the proto-scenes
or extended meanings of these English prepositions. We will illustrate this in our discus-
sion of the ‘on-the-other-side’ sense of za. There are also notable similarities between
the ‘covering’ sense (often expressed by ‘over’ as in She placed her hands over her eyes);
the ‘purpose’, ‘cause’, and ‘exchange’ senses (often translated as ‘for’), the ‘in-tandem’
sense (often translated as ‘after’ or ‘behind’), and the focus of attention sense (often
translated as at).
Key to a systematic analysis of the majority of the senses involved in the network
is recognizing two major clusters of senses, the ‘Behind-Deictic Center’ cluster and the
‘In-tandem’ cluster. A cluster involves a key extension from the proto-scene, which in
turn forms the basis for other extended senses. Tyler and Evans also found clusters of
senses within the polysemy networks of English prepositions.
The first main extension we will consider is the Behind-Deictic Center cluster, as
represented by the following diagram:

Diagram 5: Behind-Deictic Center

Note that the spatial scene represented in this diagram looks very like the proto-scene
except that the vantage point has shifted from off-stage to on-stage (as represented by the
eye to the left of the LM). The Deictic Orientation scene represents a natural extension of
the proto-scene in that humans in their everyday lived experience are constantly viewing
the same spatial configurations from varying vantage points. Indeed, Clark (1973) has
argued that the human perceptual system crucially relies on humans constantly shifting
perspective. Langacker (1987, 1991) has identified changing perspectives on spatial
scenes as a key cognitive process that has multiple manifestations in language.
An important consequence of this particular shift in vantage point from off-stage
to on-stage and in front of the LM is a shift in how orientation of the LM is established.
This is essentially a shift from intrinsic orientation in the proto-scene that emanates
from the nature of the LM itself to a deictic orientation imposed by the ‘vantage point’.
Although the LM is still understood as being oriented, the mechanism by which orienta-
tion is assigned is different. Determining the front and back of a LM by means of deictic
orientation has the effect of de-emphasizing whatever intrinsic orientation that might
be inherent in the LM. This de-emphasis is represented in the diagram by the ‘nose’ of
the LM appearing in broken lines rather than solid lines. The following is an example
from our corpus illustrating the ‘Behind-Deictic’ sense:
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 281

(29) Za zaborom roslo derevo.

[Za fence-INST was growing tree-NOM]
The tree was growing behind the fence.

Note that this sentence could be uttered by someone standing inside the area enclosed by
the fence, looking out, or standing outside the enclosed area, looking in. This ambiguity
of the scene being depicted clearly shows that the LM, the fence, does not have an
intrinsic orientation, in the sense that a human is intrinsically oriented with a front and
back; nor does the fence have an intrinsic functional orientation as a mirror or house
would have. What is understood as ‘in front of ’ the fence or ‘behind’ the fence depends
on the viewer’s perspective. In other words, in the scene described here, the orientation
of the LM is assigned by the viewer’s vantage point. (For additional discussion of similar
shifts in perspective, see Zinken, this volume).
Now we turn to an examination of one of the extended senses within the ‘Behind-
Deictic Orientation’ cluster – the ‘Beyond’ sense.

Diagram 6 The Beyond sense

As with the other extensions, the change in this spatial scene and the one it is linked
to, that is the Deictic Center sense, is incremental. Here the change involves a shift in
interpretation of the TR. In both the proto-scene and the Deictic Center scene, the TR
is neutral, that is, it is not highlighted or given particular salience. In the spatial scene
associated with the Beyond sense, the interpretation of the TR and its location in relation
to the LM are highlighted. Such highlighting represents a shift in perspective (Langacker,
1987) or a shift in the conceptualization of the scene. In the Beyond scene the viewer is
particularly focused on the location of the TR that is highlighted. Highlighted status is
represented in the diagram by the dotted line ringing the TR. The following sentences
illustrate this use of za.

(30) On zhivët za rekoĭ, na ĭugo-vostoke.

[He-NOM lives za river-INST, at south-east-PREP]
He lives over the river, in the south-east (beyond/on the other side of).

In this sentence, the LM is the river. The speaker is standing on one side of the river and
focusing on the location of the TR (‘he’ is a metonymy for ‘his’ house) on the opposite
side of the river. There also seems to be a shift in conceptualization of the LM. In the
proto-scene the LM serves a neutral locating function, as in:

(31) Za mal’chikami vozvyshalsĭa staryĭ dub.

[Za boys-INST was towering old oak-tree-NOM]
Behind the boys towered an old oak tree.
282 LANGUAGE, COGNITION AND SPACE

In the Beyond scene, the LM is conceptualized as a barrier or boundary between the

on-stage viewer and the TR. The TR is conceptualized as being at a distance from the
viewer, i.e. on the other side of the barrier.
In English, either over or on-the-other-side can be used to appropriately convey
the interpretation of sentence 30. This scene is strikingly like the spatial scene for the
on-the-other-side sense of over in English, with an on-stage perspective point and a
LM conceptualized as a barrier or boundary between the on-stage viewer and the TR
as in the sentence,

(32) Arlington is over the river from Georgetown.

Notice that in sentence (30) the LM, river, is marked with Instrumental case. Following
Janda’s analysis, we hypothesize that Instrumental indicates a static scene in which no
trajectory occurs.
Now consider a second sentence illustrating the Beyond sense:

(33) Potom ĭa uekhal za granitsu.

[Then I-NOM left za border-ACC]
Then, I moved abroad. (Then, I left to the other side of/ over/beyond the border.)

Here we infer that the TR, ĭa/I, was on one side of the border and crossed it, ending up on
the other side of the border. We also understand that the purpose of the movement was
to reach a destination. Note that this movement/destination interpretation is prompted
for by the verb and Accusative case on granitsu, the LM.

3.8 The In-tandem cluster

The second main extension revealed by the data is what we term the ‘In-tandem’ sense,
as represented by diagram 7:

Diagram 7 The In-tandem scene

Note again, that this spatial scene looks very similar to the proto-scene. The only dif-
ference is that the TR is oriented, as well as the LM. This is represented by the ‘nose’ on
the TR, which is fully darkened to represent the importance of the TR being oriented.
In our real world, lived experiences, we often observe two entities, often animate,
aligned in what Hill (1978) calls an ‘in-tandem’ configuration. In such a configuration,
both the LM and the TR are interpreted as being oriented in the same direction, one
behind the other. An unavoidable consequence of both entities being oriented is the
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 283

potential introduction of a second vantage point, that of the TR, if the TR is animate. This
is an important distinction from the proto-scene. The following sentence exemplifies
the In-tandem configuration in which both the LM and TR are inherently oriented and
an on-stage perspective point residing with the TR exists:

(34) Na kontserte mne bylo plokho vidno, potomu chto ĭa sidela

[At concert-PREP I-DAT was poorly visible because I-NOM was sitting
za vysokim muzhchinoĭ.
za tall man-INST]
At the concert, I couldn’t see much, because I was sitting behind a tall man.

Here the speaker, the TR, clearly has a perspective point. However, not all instances of
this sense require the TR to have a perspective point.

(35) Prĭamo za muzhchinoĭ sidel bolshoĭ plĭushevyĭ mishka.

[Directly za man-INST was sitting big teddy bear-NOM]
Sitting directly behind the man, was a large teddy bear.

Here both the TR and LM are oriented, but the TR has no vantage point.

(36) …drug za drugom, v odin rĭad, opustiv golovy, shli

[one za other-INST, in one line-ACC, heads-ACC down, walked
desĭatki sobak.
dozens-NOM dogs-GEN]
Heads down, dozens of dogs walked in one line, one behind the other.

Here both the TR and LM are oriented, but the vantage point is off-stage.
Recall we noted Tyler and Evans (2003) argued that if a particular interpretation
was derivable from integrating the proto-scene, the other elements in the sentence and
our knowledge of the world, then it should not be considered an independent sense in
the polysemy network. What that analysis overlooked is that there can be instances of
a spatial scene that have particular properties, which, although they could be inferred
from context, occur so frequently that they may become entrenched in memory and
thus become part of the polysemy network. We suggest that the In-tandem sense is
such a case. Consider the sentence: I couldn’t see because I was sitting behind a tall
man. With our basic knowledge of the human body and the proto-scene for za we
can infer that both the LM and TR are oriented and that the LM and TR are aligned
such that the TR is posterior to the LM. Thus the interpretation of the scene could be
derived through on-line processing and inferencing. But it is important to note that this
configuration does add two components to the basic proto-scene, namely that the TR
must be oriented and the LM and TR must both be aligned such that they are facing in
the same direction. As we saw earlier, these are not requirements for the proto-scene.
284 LANGUAGE, COGNITION AND SPACE

Hill (1978) noted that in-tandem alignment is ubiquitous and hypothesized that it has
an important cognitive status. Moreover, the Russian data indicate that the In-tandem
scene forms the basis for a wide range of za’s extended meanings. It seems questionable
to argue that there is no independent, entrenched representation of the In-tandem sense
in the polysemy network if it forms the basis of many established meanings. Thus we
posit the In-tandem sense as an independent sense in the network and designate it as
the foundational scene in the In-tandem cluster.

3.9 The Following sense

If the two in-tandem entities are in motion, the one we experience as first encountered
is interpreted as leading, the entity we experience as second encountered is interpreted
as following the first.

(37) Muzhchina shël za kolonnoĭ demonstrantov.

[Man-NOM was walking za line–INST demonstrators-GEN]
The man was following a line of demonstrators.

Interpreting the LM and TR as being in a leading/following relationship assigns a degree

of intentionality to the spatial-physical relationship. This sentence might be translated
at some ‘literal’ level as ‘The man was walking behind a line of demonstrators’ but this
translation loses the native speaker’s interpretation that the man didn’t just happen to be
taking a stroll and inadvertently ended up taking the same route as the demonstrators.
Rather native speakers interpret this sentence to mean that the man was purposefully
walking behind or following the demonstrators. We represent this notion of intentional-
ity on the part of the TR by the eye in the TR’s head, which emphasizes the animacy
and viewpoint of the TR. Another distinguishing characteristic of the Following sense
is that the LM and TR must be in motion. We designate this by representing both the
LM and TR as walking.

Diagram 8 The Following sense

3.10 The Following sense and case

Let’s return to sentence 37. Note that in this sentence kolonnoĭ ‘line’ (lit. ‘column’) occurs
in the Instrumental case. At first blush, this may seem strange as it is often the case that
if the TR is in motion, Accusative case is used (as in examples 6c and 13). However, in
our rather extensive corpus, we found no instances of the Following sense occurring
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 285

with Accusative case. We hypothesize that this is so because while in this scene the
TR may be close to the LM and moving along with the LM, the TR doesn’t actually
reach the LM. Moreover, the LM seems not to be conceptualized as a destination. For
example, there is nothing in sentence 37 that implies the man is trying to catch up with
the demonstrators. We hypothesize that in the Following scenes, even though the LM
and TR are in motion, there is no change in the TR’s position vis-a-vis the LM. In other
words, this is a stable scene, in which the LM is not conceptualized as a destination.
This analysis is consistent with Janda’s analysis of the central meanings of Accusative
and Instrumental case. By Janda’s analysis, the prototypical meaning of Instrumental
is scene setting, while the prototypical meaning of Accusative case indicates a destina-
tion or end point of motion, not that the noun marked with Accusative is in motion.
According to Janda, this interpretation holds for the meaning of Accusative case with all
Russian prepositions as well as case marking of the direct objects (which are analyzed
as representing the end point of the motion, or in Langacker’s terms, the energy sink).
Thus, since the LM is in Following sense does not represent a destination, the LM is
marked with Instrumental case. A similar analysis applies to the In-tandem sense.

3.11 The Purpose sense

Even though the In-tandem sense does not co-occur with the Accusative, a number of
senses that derive from the In-tandem scene can co-occur with the Accusative. By our
analysis this is possible because once a scene is entrenched in memory it is subject to
re-analysis. This includes being viewed from different perspectives, which can potentially
give rise to re-interpretations, new implicatures and eventually semantic extensions. For
instance, the stable LM-TR relationship prompted for by the Following sense can be
reinterpreted such that the LM is understood as a goal. As a goal, the LM is privileged
or highlighted within the scene. This shift in conceptualization reflects the common
experience of humans being in a following situation and having the additional desire to
reach the LM. Thus, the LM is no longer simply a neutral locater for the TR. Consider
the English sentence The hunter followed the fox. An inference that fits our schema
for fox hunting includes the notion that capture of the fox (which necessarily entails
physically reaching the fox) is likely the hunter’s goal. Because we understand from our
lived experience that following often includes the goal of reaching the LM, a Purpose
sense has been extended from the Following sense.

Diagram 9: The Purpose sense

The diagram representing the spatial scene we posit for the Purpose sense is similar to
the Following scene. There are two main differences. The LM is reconceptualized as a
286 LANGUAGE, COGNITION AND SPACE

goal to be reached by the TR. This privileging is indicated by dotted lines surrounding
the LM. Second, the LM is no longer explicitly oriented. Key elements are an intentional
TR (who has a particular vantage point on the scene), the interaction between the TR
and LM (a continuation of the functional element in the proto-scene) and the LM
(re)conceptualized as a goal. Importantly, once the LM is reconceptualized as a goal
and the Purpose sense becomes entrenched in the network, the exact qualities of the
original scene, in this case the orientation of the LM, can drop away. Again, in our lived
experience we often intentionally move towards entities that we conceptualize as goals
that are not necessarily inherently oriented. The Purpose sense co-occurring with the
Accusative is exemplified below:

(38) Shkoly sorevnovalis’ za luchshuĭu uspevaemost’.

[Schools-NOM competed za best results-ACC]
Schools competed for best results.

Here it seems clear that the schools’ purpose in competing is to obtain the best results.
The data from our corpus show that the interaction of case with the Purpose sense
is rather complex. The Purpose sense can co-occur with either the Instrumental or the
Accusative. However, unlike examples we saw in the Deictic Center cluster, whether
or not the TR is in motion does not explain the distribution of case. In the following
sentence, the sentence-level TRs are in motion, but ‘milk’ appears in the Instrumental:

(39) My s mamoĭ poshli v magazin za molokom.

[We-NOM with mom-INST went to store-ACC za milk-INST]
Mom and I went to the store for milk.

We hypothesize the choice of case in the Purpose sense has largely to do with the
underlying schema prompted for by the verb. The verb ‘go’ prompts for a Path schema
with a TR moving along the Path. In sentence 39, we find an overtly articulated physical
destination, ‘the store’, which is marked with the Accusative case, as we would expect
from Janda’s analysis. The TR following za articulates the reason for moving to the par-
ticular physical location. The physical destination does not need to be overtly articulated
in order for the concept of a destination to be available:

(40) Mama poshla za molokom.

[Mom-NOM went za milk-INST]
Mom went to get milk.

Here the physical destination is not overtly articulated, but a physical destination can be
easily deduced from our background knowledge about how and where we obtain milk.
Importantly, the notion of physical destination is coherent with the semantics of the
verb ‘to go’. ‘To go’ is a member of a class of verbs that can be called ‘directional’, which
prompt for scenes that involve destinations. Following Tyler and Evans’ (2003) account,
destination is part of the general Path schema. They argue that Path is ‘a consequence
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 287

of an endpoint or goal being related to a starting point or locational source by virtue of

a series of contiguous points. That is, the concept of path requires a particular spatial
goal…’ (Tyler and Evans, 2003:218) This understanding of Path would predict that
‘directional verbs’ will co-occur with endpoints marked by Accusative.
Now re-consider example 38, reproduced below for ease of argumentation:

(38) Shkoly sorevnovalis’ za luchshuĭu uspevaemost’.

[Schools-NOM competed za best results-ACC]
Schools competed for best results.

This example employs a ‘non-directional’ verb, ‘compete’, that does not prompt for
a scene involving the prototypical Path schema and a particular spatial destination.
Although we clearly understand that the competition took place somewhere, a path
with a particular physical destination does not seem to be part of the scene. Rather the
reason for or purpose of the activity seems to also be the desired endpoint of the activity.
In other words, non-directional verbs seem to prompt for a different schema in which
purpose involves some of the attributes of destination or endpoint. With directional
verbs, destination and purpose are clearly distinguished.
This distinction is illustrated in the following minimal pairs:

(41) Ehat’ na Olimpiĭskie Igry za zolotoĭ medal’ĭu.

Go to Olympic Games-ACC za gold medal-INST]
Go to the Olympic Games for the gold medal.

Here our understanding of the scene involves a Path schema with France as the physical
destination. The purpose for going to the Olympic Games is to obtain the gold medal,
which is marked with Instrumental case.

(42) Borot’sĭa na Olimpiĭskikh Igrakh za zolotuĭu medal’.

[Fight at Olympic Games-PREP za gold medal-ACC]
Fight for the gold medal at the Olympic Games

Here our understanding of the scene does not involve a path with a physical destination.
The purpose for competing at the Olympic Games is to win the gold medal; winning
the gold medal is the desired endpoint of the activity. These examples demonstrate that
if a Path schema is prompted for, as in sentences 39 and 41, the physical destination is
coded by Accusative case and the purpose is coded by Instrumental case. In contrast,
if a Path schema (with a spatial destination) is not prompted for, as in sentences 38 and
42, the purpose is interpreted as the desired endpoint of the activity and is coded by
Accusative case.
A consequence of this entrenched distinction in interpretation of destination or
endpoint with directional verbs versus non-directional verbs is that a purpose phrase
that is marked with Instrumental case tends to be physical or concrete, while a purpose
marked with Accusative tends to be more abstract. This explains the oddity of the
following examples:
288 LANGUAGE, COGNITION AND SPACE

(43) (a) Poĭti za molokom

[Go for milk-INST]
(b) ?*Poyti za obrazovaniem
[?*Go for education-INST]

In sentence 43a, we see the familiar directional verb ‘go’ plus purpose in which ‘milk’
(the purpose for going) is marked with Instrumental. However, in sentence 43b, even
though the directional verb ‘go’ plus the Purpose sense co-occurs with the expected
Instrumental case, native speakers judge the sentence as odd. We believe the sense of
oddness stems from the non-physical TR, ‘education’, being marked with Instrumental
case. We find a similar pattern in sentence 44.

(44) (a) Borot’sĭa za obrazovanie

[Fight for education-ACC]
(b) ?*Borot’sĭa za moloko
[?*Fight for milk-ACC]

In sentence 44a, the non-directional verb ‘fight’ co-occurs with a Purpose phrase whose
TR is non-physical and marked with the expected Accusative case. In contrast, in 44b,
even though ‘fight’ plus the Purpose sense occurs with the typical Accusative case, the
sentence sounds odd to native speakers. We hypothesis that this oddity arises from the
physical TR, ‘milk’, being marked with the Accusative case. (We do note that certain
contexts can be created in which these questionable sentences sound less odd).
To conclude our discussion of case and the Purpose sense, we can represent the
case distribution for the Purpose sense of za with the following patterns:

(45) Instrumental:
[N + destination verb + (destination -ACC) + za (purpose) + N (physical)-INST]
Zvonarëva + returned + (to Memphis) + za + victory-INST
Zvonarëva returned to Memphis for victory.
Zvonarëva vernulas’ v Memfis za zolotoĭ medal’ĭu
[Zvonarëva-NOM returned in Memphis-ACC za gold medal-INST]

(46) Accusative:
N + non-destination verb + (location) + za (purpose) + N (non-physical)-ACC]
Zvonarëva + borolas + (v Memfise) + za + victory-ACC
Zvonarëva fought for victory in Memphis.
Zvonarëva borolas’ v Memfise za zolotuyu medal’.
[Zvonarëva-NOM fought in Memphis-PREP za gold medal-ACC]
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 289

4 Conclusion

In this chapter we have demonstrated that Tyler and Evans’ (2003) Principled Polysemy
model can be successfully extended to languages other than English. By applying the
basic principles laid out in the Principled Polysemy model, we have been able to provide
a systematic, motivated analysis for the highly polysemous Russian preposition, za.
The analysis revealed that while the proto-scene for za appears to bear similarities to
that of English behind, many of the spatial scenes associated with za’s extended senses
are quite different. Indeed, a number of the extended senses associated with za that are
standardly translated into various English prepositions, such as over and for, represent
spatial scenes that are very similar to the spatial scenes associated with the extended
senses of these English prepositions. For instance, the spatial scene associated with the
on-the-other-side sense of za, one of the extended senses in the Deictic Center cluster,
is very similar to the spatial scene Tyler and Evans posit for the on-the-other-side sense
of over, one of the extended senses from the ABC trajectory cluster. This is one of the
contexts in which za is regularly translated as ‘over’. This is consistent with Tyler and
Evans’ predictions.
Moreover, the analysis has shed light on some puzzling aspects of the distribu-
tion of Instrumental and Accusative case in Russian. It has been common to associate
Accusative case with motion and Instrumental case with lack of motion. However, Janda
has recently offered a more refined analysis in which Instrumental case is represented
as prototypically linked with scene setting and Accusative with destination. Drawing
on Janda’s insights, we argued that the Following sense, in which za always co-occurs
with Instrumental case even though the participants are in motion, does not entail
the TR reaching the LM and therefore does not involve the notion of destination in its
interpretation.
The distribution of case with the Purpose sense also challenges the simple associa-
tion of Accusative with motion and Instrumental with lack of motion. Analysis of our
data revealed that with directional verbs, such as ‘go’, which clearly involve a Path
schema containing a beginning and a particular spatial destination, Accusative is used
to mark the spatial destination; the TR in the Purpose phrase is consistently marked
with Instrumental case and seems to be quite separate from the spatial destination. In
contrast, with non-directional verbs, such as ‘compete’ and ‘fight’, which do not evoke
a prototypical Path schema with a spatial destination, the TR in the Purpose clause is
marked with Accusative. We hypothesize that with these non-directional verbs there is
a conceptual coalescence of purpose and the end of the action. This analysis provides
support for Tyler and Evans’ (2003) distinction between Path and the trajectory followed
by the TR in a specific motion.
Finally, the analysis of za’s polysemy network forced us to reconsider the strong
claim made by Tyler and Evans that if a non-proto-scenic interpretation is derivable
from context, then it should not be considered an independent sense. We argued that
certain spatial scenes, although derivable from the proto-scene and context, may occur
so frequently that they are entrenched in memory. Once they are entrenched in memory,
they are free for re-analysis and can form the basis for further extended meanings. We
believe that za’s In-Tandem sense represents such a case.
290 LANGUAGE, COGNITION AND SPACE

Notes
1 Za is both a preposition and a verbal prefi x. In this paper we only address its uses as a
preposition. We believe that the verbal prefi x meanings associated with za are related to
its prepositional meanings, but that analysis goes beyond the scope of this paper.
2 This sense is part of what we called the ‘in tandem’ cluster, following Hill (1978). Hill
noted that a salient, frequently occurring orientation for humans involves two indi-
viduals facing the same direction and lined up one behind the other. He termed this
spatio-physical arrangement ‘in tandem’.

References
Clark, H. (1973) Space, time, semantics and the child. In T. E. Moore (ed.) Cognitive
Development and the Acquisition of Language 27–64. New York: Academic Press.
Dal’, V. I. (1998) Tolkoviy Slovar Zhivogo Velikorusskogo Yazika. Moscow: Tsitadel.
(Original published in 1881.) Retrieved on 10 February 2003 from http://www.
rusword.org/rus/dal_in.php
Dewell, R. (1994) Over again: Image-schema transformations in semantic analysis.
Cognitive Linguistics 5(4): 351–380.
Dostoevsky’s online corpus: Slovar’ Konkordans Publicistiki Dostoevskogo. Retrieved
on 20 December 2002 from http://mmedia3.soros.karelia.ru:8080/~dost_voc/
Grice, H. P. (1975) Logic and conversation. In P. Cole and J. Morgan (eds) Syntax and
Semantics 3: Speech Acts 41–58. New York: Academic Press.
Hill, C.A. (1978) Linguistic representation of spatial and temporal orientation. In
Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society
524–538. Berkeley, CA: UC Berkeley Press.
Janda, L. A. (1993) A Geography of Case Semantics: The Czech Dative and the Russian
Instrumental. Berlin/New York: Mouton de Gruyter.
Janda, L. A. (2000) A cognitive model of the Russian accusative case. In R. K.
Potapova, V.D. Solov’ev and V.N. Poljakov (eds) Trudy meždunarodnoj konferen-
cii Kognitivnoe modelirovanie 4(Part I): 20–43. Moscow: MISIS.
Lakoff, G. (1987) Women, Fire and Dangerous Things. Chicago: University of Chicago
Press.
Langacker, R. (1987) Foundations of Cognitive Grammar Vol. I: Theoretical
Prerequisites. Stanford, CA: Stanford University Press.
Langacker, R. (1991) Foundations of Cognitive Grammar Vol. II: Descriptive
Application. Stanford, CA: Stanford University Press.
National Corpus of the Russian Language. Retrieved in February-March 2006 from
http://ruscorpora.ru
Ozhegov, S.I., Shvedova, N.Y. (1984) Slovar’ Russkogo Yazika. Moscow: Russkiy Yazik.
Poltoratsky, M. and Zarechnak, M. (1961) Russkiy Yazik: Pervaya Kniga 2.
Milwaukee: The Bruce Publishing Company.
Pulkina, I.M. and Zakhava-Nekrasova, E.B. (2000) Russkiy Yazik: Prakticheskaya
Grammatika S Uprazhneniyami. Uchebnik (dlia govoryashih na angliyskom
yazike). (Eighth edition.) New York, Moscow: Russkiy Yazik.
 TAKING THE PRINCIPLED POLYSEMY MODEL OF SPATIAL PARTICLES BEYOND ENGLISH: THE CASE OF RUSSIAN ZA 291

Sandra, D. and Rice, B. (1995) Network analysis of prepositional meaning: Mirroring

whose mind – the linguist’s or the language user’s? Cognitive Linguistics 6(1):
89–130.
Slovar’: Obschaya Leksika Russkogo Yazyka. Retrieved on 5 March 2003 from http://
lingvo.yandex.ru/cgi-bin/lingvo.pl?text=%E7%E0
Talmy, L. (2000) Toward a Cognitive Semantics Vol. I: Concept Structuring Systems.
The MIT Press.
Talmy, L. (2000) Toward a Cognitive Semantics Vol. II: Typology and Process in
Concept Structuring. The MIT Press.
Tyler, A. and Evans, V. (2001) Reconsidering prepositional polysemy networks: The
case of ‘over’. Language 77(4): 724–765.
Tyler, A. and Evans, V. (2003) The Semantics of English Prepositions: Spatial Scenes,
Embodied Meaning and Cognition. Cambridge: Cambridge University Press.
Vandeloise, C. (1991) L’espace en francais. Paris: Editions du Seuil. English transla-
tion by A. Bosch: Spatial Prepositions. Chicago: University of Chicago Press.
Wilson, P. and Anderson, R. (1986) What they don’t know will hurt them: The role of
prior knowledge in comprehension. In J. Orasunu (ed.) Reading Comprehension:
From Research to Practice 31–48. Hillsdale, NJ: Lawrence Erlbaum Associates.
Zinken, Jörg (this volume) Temporal frames of reference. In V. Evans and P. Chilton
(eds) Language, Cognition and Space: The State of the Art and New Directions.
London: Equinox.
12 Frames of reference, effects of motion, and
lexical meanings of Japanese front/back
terms
Kazuko Shinohara and Yoshihiro Matsunaka

1 Introduction*

Spatial cognition is often said to play the central and fundamental role in our think-
ing. Spatial concepts and how they are expressed have been discussed for many years
in a wide variety of disciplines including philosophy, physics, cognitive science, and
anthropology. Of course, these topics have also attracted linguists, who have long noticed
that spatial concepts are related to a vast range of linguistic phenomena. For example,
words that express spatial relations are among the most basic elements in language, as
instantiated by adpositions, conjunctions, and so forth. These functional elements in
language are often the result of the process known as grammaticalization. Moreover,
spatial concepts also serve as the source domains of widespread metaphors, such as
metaphors of time, state, emotion, and life. It is not a simple task, however, to clarify how
spatial cognition and lexemes denoting space are related. Naturally, while innumerable
previous studies have addressed this issue, there remain a lot of unsolved problems. For
example, one such problem involves two contradicting positions, one that claims that
frames of reference are an extra-linguistic matter (Levinson 2003), and the other that
rejects the possibility that frames of reference may apply at the linguistic level (Svorou
1994; Carlson-Radvansky and Irwin 1993).
In addition to the difficulty in specifying the relationship between spatial cognition
and lexical meanings, there is an issue of how much specification concerning space
should be attributed to spatial lexemes. Some researchers describe lexical meanings of
space in terms of rich and detailed information. For example, Lakoff (1987) presents
a highly rich description of the image-schematic networks of ‘over’. Other researchers
like Tyler and Evans (2003) avoid assigning such specified information to each lexeme
but attribute much of spatial relations associated with spatial expressions to contextual
information and encyclopedic knowledge.
In this study, we aim to consider these issues by examining the uses and meanings
of three Japanese spatial lexemes mae (front), ushiro (back), and saki (front/ahead).
Our analysis and empirical data will support Levinson’s position and Tyler and Evans’s
position stated above. In the remainder of this paper, we first review relevant previous
studies and present our goal in Section 2. Then we examine unmarked uses of these
lexemes in Section 3. In Section 4, our experiment on these lexemes is reported. Finally,
Section 5 concludes this study.

293
294 LANGUAGE, COGNITION AND SPACE

2 Previous studies and issues

Two lines of research are critical for the present study. One is Levinson’s (1996, 2003)
framework of spatial frames of reference, and the other is the theory of lexical meanings
by Tyler and Evans (2003) and Evans (2004). We will first introduce these theories and
ideas, and then present the issues we are addressing.

2.1 Frames of reference

Many researchers (Clark 1973; Talmy 1983, 2000; Vandeloise 1991; Svorou 1994; Levelt
1996; Levinson 1996, 2003, among others) describe the notion of frames of reference
as playing one of the most fundamental roles in the study of spatial cognition and its
linguistic expression. Though this notion has been defined and classified in various
ways in different disciplines, the shared view seems to be that cognition of the spatial
relationships of objects involves at least the following three elements.

(i) a referent, trajector, or a figure (the object to be located)

(ii) a relatum, landmark, or a ground (the object relative to which the referent is
located)
(iii) a perspective system or a frame of reference (the system that determines the
relation of a referent to a relatum)

Levelt (1996: 78) uses the terms ‘referent’, ‘relatum’ and ‘perspective system’, which
correspond to ‘figure’, ‘ground’ and ‘frame of reference’ respectively in Talmy’s (1978,
1983) and Levinson’s (1996) terminology. Langacker (1987) calls the first two elements
‘trajector’ (TR) and ‘landmark’ (LM). In the main part of this paper, the terms ‘figure’,
‘ground’, and ‘frame of reference’ will be consistently used to avoid confusion. In the
expression ‘X is in front of Y’, for example, X is the figure, Y is the ground, and the frame
that determines the spatial relation of X to Y is the frame of reference.
Scholars differ in their ideas on what kinds of frames of reference are necessary and
sufficient. Some posit two types of frames of reference, e.g., egocentric frame of reference
versus allocentric frame of reference in developmental psychology, or deictic frame of
reference versus intrinsic frame of reference in linguistics. Others posit three types of
frames of reference, e.g., viewer-centered frame of reference versus object-centered
frame of reference versus environment-centered frame of reference in psycholinguistics.
(These classifications are reviewed by Levinson 2003: 26.) Among these different sub-
divisions of frames of reference, this paper follows Levinson’s three-way classification
of linguistic frames of reference: the intrinsic, the relative, and the absolute frames of
reference. Figure 1 illustrates the three frames of reference.
 FRAMES OF REFERENCE, EFFECTS OF MOTION, AND LEXICAL MEANINGS OF JAPANESE FRONT/BACK TERMS 295

North

Right
Front

Back
Left
Observer
Figure 1. Frames of reference

In the intrinsic frame of reference, the cat (figure) is said to be ‘in front of the truck
(ground)’. This relation is based on the coordinate system (or particularly the front/
back axis) determined by the intrinsic properties of the truck including its functional
aspects. For example, the side of the truck that faces the default direction of motion
may be regarded as the ‘front’ of the truck. The other three directions, ‘back’, ‘right’
and ‘left’ are derived from ‘front’. In this frame, the figure object (cat) is located in the
direction of ‘front’ of the ground object (truck) which is determined in this way. Note
that the intrinsic front/back asymmetry of the figure object (cat) does not determine
the orientation of the ground object in this case. Though the cat is looking in a different
direction from that of the truck, the cat is still referred to as being ‘in front of ’ the truck
when the intrinsic frame of reference is employed.
In the relative frame of reference, the cat is described as being ‘to the left of the truck’.
This relation comes from the observer’s viewpoint. That is, the observer’s coordinate
system (front, back, right, and left) provides the ground object with these directions.
The truck thus obtains its front, back, right, and left based on the observer’s viewpoint,
which makes it possible to say ‘The cat is to the left of the truck’. The observer’s left is
regarded as the left side of the truck in this case, though the same side of the truck may
be the ‘front’ side in the intrinsic frame of reference.
In the absolute frame of reference, the coordinate system of the truck is determined
by the configuration of the outside world, which is non-relative or non-intrinsic. The
expression ‘The cat is north of the truck’ is based on the earth’s magnetic field and car-
dinal orientations derived from it, which are determined independent of the observer’s
viewpoint or the intrinsic orientation of the truck.
In the intrinsic frame, the ground object itself provides the coordinate system. In
the relative and absolute frames, the coordinate system of some other object is projected
onto the ground object. In case of the relative frame, the observer is the source of the
coordinate system of the ground object, while in the absolute frame, the earth is the
source of the coordinate system of the ground object. Talmy (2000) classifies the latter
296 LANGUAGE, COGNITION AND SPACE

two cases together and states that both of them have Secondary Reference Objects
while the intrinsic frame has a Primary Reference Object. In the present study, we do
not refer to the distinction between Primary and Secondary Reference Objects, since
we only deal with the relative frame of reference.
Among these three frames of reference, the relative frame has three subtypes of
projecting the coordinate system on the viewpoint of the observer onto the ground
object. They are reflection analysis, translation analysis, and rotation analysis (Levinson
2003: 86–88). Figure 2 illustrates them.

Coordinate 2

C
B
Coordinate 1

A
D
Left
Front

Back
Right

Figure 2. Subtypes of the relative frame of reference

In reflection analysis, the polar A of Coordinate 2 in Figure 2 is said to be front of the

tree; B is back of the tree; C is left of the tree; D is right of the tree. The coordinate system
on the viewpoint is projected onto the ground (the tree) in such a way that the front/
bask axis is reversed but the right/left axis is not reversed. ‘The cat is in front of the
tree’ means that the cat is between the observer and the tree. In this analysis, the dog is
said to be ‘to the right of the tree’. In other researchers’ works, reflection analysis uses
different terms. Clark (1973) uses the term ‘canonical encounter’, Hill (1978) adopts
‘mirror-image’ strategy, and Moore (2000) calls this ‘ego-opposed’ strategy. English,
Japanese, and perhaps many other languages have this type of projection system.
The second subtype of the relative frame is the translation analysis. In this case, A
is said to be back of the tree; B is front of the tree; C is left of the tree; D is right of the
tree. In this subtype, the coordinate system on the viewpoint is translated without any
reversal or rotation. If this frame is employed, the ‘front’ side of the tree is the farther
side of the tree from the observer, and the space between the observer and the tree is
regarded as ‘behind’ the tree. Thus, speakers who take this frame will describe the scene
in Figure 2 as ‘The cat is behind the tree’. As for the right/left axis, this frame results
in the same expressions as reflection analysis. Hill (1978) calls this frame ‘in-tandem’
strategy, and Moore (2000) adopts the term ‘ego-aligned’ strategy because the ground
object that has no intrinsic front/back axis is construed as if it stood looking in the
 FRAMES OF REFERENCE, EFFECTS OF MOTION, AND LEXICAL MEANINGS OF JAPANESE FRONT/BACK TERMS 297

same direction as the observer. Hausa, an African language of the Chad family, is said
to have this type of frame of reference as the dominant, unmarked frame for FRONT/
BACK terms (Hill 1975, 1978, 1982, Levinson 2003).
The third subtype, the rotation analysis, is a rare case in world languages. In this
frame, A is front of the tree; B is back of the tree; C is right of the tree; D is left of the
tree. The coordinate system on the viewpoint is projected onto the ground object after
being rotated 180 degrees. Thus, the right/left axis and the front/back axis are both
reversed. In Figure 2, the cat between the observer and the tree is said to be ‘in front of
the tree’ and the dog is said to be ‘to the left of the tree, while in English the dog may
be to the right of the tree. It is said that Tamil, a Dravidian language, uses this system
(Levinson 2003: 88).

2.2 Frames of reference in Japanese

In Japanese, it is observed that all three frames of reference, the absolute, the intrinsic,
and the relative frames of reference, are available. However, the absolute frame of refer-
ence is relatively limited in use, except in some cases in rural dialects (Inoue 1998, 2002,
2005, Kataoka 2003). The absolute frame of reference is mostly used to refer to distal
geographical places. For example, when describing maps or directing ways to places
in the distance, the absolute frame of reference tends to be used. To describe things in
more proximal space, the intrinsic and the relative frames are likely to be used by most
Japanese, especially city-dwellers. Things that we see around us and manipulate or use
in daily life are mostly described in the intrinsic or the relative frame of reference. It is
not that we cannot describe proximal spatial relations in terms of the absolute frame of
reference, but this is limited to certain dialects, and it requires a considerable amount of
effort for speakers of standard Japanese to do this. This was confirmed by the authors’
in-class research. It took students more time and induced more mistakes to use the
absolute frame to describe the relative position of pens, books, and other objects on a
desk, than to use the relative frame to do the same thing. (Inoue (2005) reports, however,
that there are dialectal variations concerning the use of these reference frames: in some
rural regions in Japan the absolute reference frame tends to be used for proximal spatial
relations or even body parts such as teeth.)
As for the three subtypes within the relative frame of reference, most of the
relevant previous studies report that the reflection analysis is dominant in adult
Japanese (Inoue 1998; Imai and Ishizaki 1999; Imai, Nakanishi, Miyashita, Kidachi
and Ishizaki 1999; Odate, Shinohara and Matsunaka 2003; Shinohara, Matsunaka
and Odate 2003; Shinohara, Odate and Matsunaka 2003; Yoshida 2003; Shinohara
and Matsunaka 2004; Shinohara, Kojima and Matsunaka 2004a, b; Matsunaka and
Shinohara 2005, etc). Some have also shown that children tend to use translation
analysis for mae (front) and ushiro (back) more than adults do (Odate et al. 2003;
Yoshida 2003). All of these points hold in English ‘front’ and ‘back’ as well (Clark
1973, Harris and Strommen 1972, Hill 1978, 1982, Levinson 1996, 2003, etc.).
298 LANGUAGE, COGNITION AND SPACE

2.3 Lexical specification of frames of reference

Several issues have been raised concerning frames of reference in linguistics. One of
them is at which level they are coded. Levinson states this question as follows.

In psycholinguistic discussions about frames of reference, there seems to be some

unclarity, or sometimes overt disagreement, about at which level – perceptual,
conceptual or linguistic – such frames of reference apply. … [W]e need to distinguish
in discussions of frames of reference between at least three levels, perceptual,
conceptual and linguistic, and we need to consider the possibility that we may
utilize distinct frames of reference at each level. (Levinson, 2003: 33–34)

Some researchers deny the possibility that frames of reference may apply at the linguistic
level. Svorou (1994: 23) states that ‘typically RFs are not coded linguistically in spatial
expression’ (RF stands for ‘reference frame’). Carlson-Radvansky and Irwin, in their
studies on the spatial term ‘above’, find that frames of reference are not linguistically
coded (1993: 242). However, Levinson argues against these views.

[I]n most languages there are many subtle details of the use of expressions that
generally mark which frame of reference they are being used with – thus at the truck’s
front or in the front of the truck can only have an intrinsic reading, not a relative
one – so this cannot be treated as an extralinguistic matter. (Levinson, 2003: 108)

In addition to this Levinson’s analysis, previous studies on Japanese spatial lexemes like
mae (front), saki (front/ahead), temae (front), and ushiro (back) provide more evidence
of linguistically coded frames of reference. Imai et al. (1999), through an experimental
study of mae and ushiro, show that 97% of decisions concerning the front/back axis
of objects without an intrinsic axis are based on reflection analysis. Matsunaka and
Shinohara (2004, 2005) state that mae, saki and temae exhibit some restriction in the
choice of frames of reference, which renders it difficult to shift freely to other frames
of reference or subtypes. These restrictions cannot be explained if we assume that
frames of reference reside only in pre-linguistic cognition or perception, independent of
linguistic coding. We must assume, instead, that lexical items can at least in some cases
determine which frames of reference they can relate themselves to. Moreover, Shinohara
et al. (2004a) examine two Japanese spatial terms (mae and saki) denoting the frontal
concept, and show that the unmarked usage of mae is based on the reflection analysis
while that of saki is based on the translation analysis. This, they conclude, indicates that
at least some information about frames of reference is included in each of these words.
Thus, contra Svorou’s and Carlson-Radvansky and Irwin’s view that spatial frames of
reference cannot be settled at the linguistic level, evidence presented by Levinson and
previous studies on Japanese spatial lexemes show that spatial frames of reference are
linguistically-coded and included in lexical meanings at least to some extent.
 FRAMES OF REFERENCE, EFFECTS OF MOTION, AND LEXICAL MEANINGS OF JAPANESE FRONT/BACK TERMS 299

2.4 Meanings of spatial lexemes

As we have seen, Levinson’s view that spatial frames of reference are at least to some
extent coded at the linguistic level seems adequate for Japanese spatial lexemes such as
mae, saki, and temae. However, it does not necessarily lead to the idea that the senses of
spatial lexemes should be as rich as they can be. Contrary to this, scholars like Tyler and
Evans (2003: 17–18) argue that semantic properties of lexemes should be described in a
simple manner, avoiding excessively specified information. They suggest that meanings
that can be obtained by elaborating lexical meanings using contexts should be excluded
from semantic description of the lexeme. For example, Tyler and Evans’s (2003) descrip-
tion of English spatial lexeme ‘over’ is simpler than Lakoff ’s (1987) well-known analysis
of ‘over’. They state their idea as follows.

In essence, by attempting to build too much redundancy into the lexical

representation, Lakoff ’s model vastly inflates the number of proposed distinct
meanings associated with a spatial particle such as ‘over’. An implicit consequence
of this representation is that discourse and sentential context, which is utilized in
the conceptual processes of inferencing and meaning construction, is reduced in
importance, as much of the information arising from inferencing and meaning
construction is actually built into the lexical representation. (Ibid.: 42)

Thus, they argue that the general inference system elaborates the meanings of lexemes,
for example, ‘over’, in terms of the context in which the expressions are used. In their
analysis, the detailed shape of the landmark, verticality of the landmark, multiplex
nature of the trajector(s), coverage, contact between trajector and landmark, etc. are not
included in the semantic network of the lexeme ‘over’ but these kinds of information
are claimed to be obtained through elaboration.
The present study deals with slightly different aspects of meaning shifts of spatial
lexemes, but we take a similar standpoint as Tyler and Evans’ in that we intend to claim
that the meaning shifts we are looking at are induced by contextual information and
thus they should not be included in the senses of each spatial lexeme.

2.5 Issues and the goal of this study

In this study, we will consider the issue of spatial frames of reference specified by spatial
lexemes, and the issue of semantic shifts induced by context. We will demonstrate that
three Japanese front/back terms, mae, ushiro, and saki, exhibit interesting tendencies
that seem to support Levinson’s position that frames of reference can be settled at
linguistic level, and the position that lexically determined meanings of these terms,
especially for relative frames of reference (Levinson 2003), may be rather simple,
but perceptual spatial context can affect the uses of these terms to produce varying
300 LANGUAGE, COGNITION AND SPACE

construals of spatial relations of objects that have no front/back axis. We support this
argument by demonstrating the following two points: (1) specifications of frames of
reference are included in the lexical meanings of mae, ushiro and saki; (2) the effects
of motion on the uses of these terms indicate that their meanings are not so rich as
to include concrete, specific spatial regions or positions. We present evidence for the
former argument from our previous studies, and for the latter claim, we employ an
experimental method to demonstrate that different conditions of motion can add to
the basic, unmarked orientation of the ground object that these terms prompt. Visual-
perceptual contexts such as the motion of the observer or of the objects can add extra
information about orientation of the ground object whereby interaction between lexical
senses and contextual information can take place. Such context can induce shifts in
what these lexemes mean in each case. However, we regard these meaning shifts as not
included in the lexical senses. Thus, we argue that these spatial lexemes are rich enough
to prompt the unmarked reference frame to refer to, but that they are simple enough
in meaning so as to not designate concrete spatial regions or positions.

3 Basic meanings and frames of reference of mae, ushiro, and

saki

In this section we describe basic, unmarked usage of the Japanese spatial terms mae
(front), ushiro (back), and saki (front/ahead). (All of these three terms are related to
the FRONT/BACK concepts, rather than other spatial axes or directions like RIGHT/
LEFT, NORTH/SOUTH, UP/DOWN, etc.) Then we will show that they have different
specifications of frames of reference. This is especially clear when an object that has
neither intrinsic directions nor an asymmetrical shape, such as a block, a cylinder, or
a ball, is the ground object.
Mae and ushiro originally derived from bodily meanings. Mae is related to the word
me (eye), and ushiro is related to the word shiri (hip or buttock). They are unmarked
words for FRONT and BACK in Japanese. Saki basically means a tip or a sharp point
of a stick-like object (e.g., the saki of a pencil is its pointed end).
The critical question for these unmarked uses of the three spatial terms is which
of the frames of reference is employed for each term. In Section 2, we described three
different frames of reference, the intrinsic, the relative, and the absolute frames. In fact,
mae, ushiro, and saki can all be used for the intrinsic frame and the relative frame, but
not for the absolute frame. When the ground object has an intrinsic front/back axis, like
a truck, a car, a house, etc., mae can mean their frontal part, side, or region, and ushiro
can mean their back part, side, or region, based on their intrinsic front/back axis. If
the ground object has a gradually narrowing shape and a sharp tip, then saki can mean
the tip itself, or the direction of that tip, or the region in that direction. These uses are
based on the intrinsic frame of reference.
However, if the ground object has no such intrinsic axes or directions, then these
terms are interpreted based on the relative frame. That is, the viewer’s front/back axis
is projected onto the ground object. In this case, one of the three subtypes (reflection,
translation or rotation) of the relative frame of reference (see Figure 2) is employed.
 FRAMES OF REFERENCE, EFFECTS OF MOTION, AND LEXICAL MEANINGS OF JAPANESE FRONT/BACK TERMS 301

Figure 3 shows the unmarked uses of these words in the relative frame. Both the figure
object (ball) and the ground object (block) lack an intrinsic front/back axis. That is, the
orientation of the ground object cannot be determined by its shape or function. Moreover,
the situation described in Figure 3 is assumed to be static, that is, neither of the objects
are moving. Even in such cases, the lexemes mae, ushiro, and saki can be used to designate
front/back relation. These uses are understood and shared by the speakers of standard
Japanese. If you say, in Japanese, a sentence that means ‘There is a ball in mae of the block’,
the ball is normally in area A in Figure 3. If you say ‘There is a ball in ushiro of the block,
the ball is in area B. If you say ‘There is a ball in saki of the block’, the ball is in area B.

ushiro, saki

mae

Viewer

Figure 3. Unmarked uses of mae, ushiro and saki

The unmarked uses of mae and ushiro are based on reflection analysis: the nearer side
to the viewer from the ground object is referred to as mae (front), and the farther side is
referred to as ushiro (back). The unmarked use of saki is based on translation analysis: the
farther side of the ground object is referred to as saki (front). Figure 4 and 5 illustrates
these frames for mae, ushiro, and saki.

FRONT BACK
mae of the block ushiro of the block
Ground object
Viewer

Figure 4. Reflection Frame

FRONT
saki of the block

Viewer
Ground object

Figure 5. Translation Frame

302 LANGUAGE, COGNITION AND SPACE

As shown above, the basic, unmarked uses of mae, ushiro, and saki are crucially related
to certain frames of reference. Previous studies show that 80% to 97% of the uses of
these words are based on these unmarked patterns (Imai et al. 1999; Imai and Ishizaki
1999; Odate et al. 2003; Shinohara, Matsunaka and Odate 2003; Shinohara, Kojima
and Matsunaka 2004a). This indicates that these spatial lexemes have, as part of their
lexical properties, at least some specification of the unmarked frame of reference to
be referred to (Matsunaka and Shinohara 2004, 2005). Without such specification,
it would be impossible for most native speakers to share the same judgment about
the spatial usage of these terms. It is even more obvious when we consider the fact
that both mae and saki mean FRONT but the regions they actually refer to are the
opposite. Their difference seems to reside in the different frames of reference they
are associated with. We cannot say that in Japanese the frontal direction is generally
based on the reflection frame, since saki is a strong counterexample to this claim. Each
of these Japanese spatial lexemes seems to have, as its meaning, specification of the
frame of reference to be referred to in unmarked cases. This is, as we have stated in
the foregoing, inconsonant with Svorou’s (1994) and Carlson-Radvansky and Irwin’s
(1993) argument.
What we have argued in this section concerns the Japanese FRONT/BACK terms
mae, ushiro, and saki, and therefore, it cannot be directly applied to similar spatial
terms in other languages. However, some implication for semantics of spatial terms in
language may be derived from our analysis. Tyler and Evans (2003) describe meanings
of the English phrase ‘in front of ’ as follows.

As we have just seen in our discussion of ‘in front of ’, the Priority Sense and the
proto-scene involve essentially the same relationship between the TR and LM. In
both senses, the LM is oriented towards the TR. (Ibid.:164)

In their analysis, ‘orientation’ is treated as an essential part of the meaning of ‘in front
of ’. This argument seems quite convincing. We would like, however, to point out that
their analysis of the meanings of ‘in front of ’ does not include the case where both the
figure and the ground objects are symmetrical in terms of the front/back axis (that is,
lack an intrinsic front/back axis derived from physical shape, default direction of motion,
or functional properties like accessibility). The only example they give for symmetrical
objects is the case where two bottles without intrinsic front/back axis are moving in line
on a conveyor-belt. The concept of FRONT, however, is obtainable even in cases without
such motion. As we have seen in this section, static objects that have no intrinsic front/
back axis, such as a block and a ball, can be construed in terms of FRONT relation. By
including this instance, Tyler and Evans’s analysis would be more exhaustive. It should
be emphasized that this does not mean that their analysis is wrong. Rather, the FRONT/
BACK conceptualization of objects that have no intrinsic axis, no motion, nor privileged
accessibility may occupy a peripheral position in the radial category structure of frontal
terms. We will discuss this later.
 FRAMES OF REFERENCE, EFFECTS OF MOTION, AND LEXICAL MEANINGS OF JAPANESE FRONT/BACK TERMS 303

In this section, we have examined basic, unmarked uses of mae, ushiro, and saki in
the relative frame, and have shown that the first two lexemes select the reflection frame
as the unmarked frame to refer to, but saki selects the translation frame as its unmarked
frame of reference when the objects have no intrinsic front/back axis. It has been made
clear that each lexeme has its own specification of frame of reference for unmarked uses.

4 The effect of motion on the meanings of mae, ushiro, and saki

In the previous section, we have shown that Japanese spatial lexemes mae, ushiro, and
saki designate, at least to some extent, specification of the frames of reference they are
associated with. In this section, we will further show that these spatial lexemes can
sometimes shift their frames of reference when motion is involved in perceptual context.
We do this by way of experimentation. By using three-dimensional computer graphic
images to create a sense of virtual reality, we made the viewer feel as if she were moving
toward the objects on the screen or as if the objects were drawing nearer to the viewer.
In the following, we will describe the method of our experiment, show the data we have
obtained, and then discuss what the results mean.

4.1 Method of experiment**

In our experiment, the stimuli consisted of twelve pairs of three-dimensional computer

graphic images and sentences. In each image, two objects (a green ball and a red block,
both of which have no intrinsic front/back or right/left axes) were located along the
viewer’s frontal axis, so that the viewer sees these two objects aligned with her orienta-
tion. Figure 6 shows the spatial configuration of the objects shown on the screen. The
numerals in Figure 6 represent the ratios of the distance between the objects and the
viewer. Figure 7 is a rough image of what the participants see just before the objects
start moving. Two conditions for motion were set: the viewer-in-motion condition
and the objects-in-motion condition. (Under the viewer-in-motion condition, the
viewer feels as if she were moving toward the two objects in the screen. Under the
objects-in-motion condition, the viewer feels as if the two objects on the screen were
moving toward her.)

18.0

2.0

4.0
Figure 6. Design of the stimuli
304 LANGUAGE, COGNITION AND SPACE

Figure 7. Computer graphic image

Each graphic image was accompanied by a Japanese sentence describing the spatial
relation of the objects, which has one of the following two sentence structures.
(a) midori-no kyuu-wa akai rippoutai-no mae [ushiro / saki]-ni aru.
green-Gen. ball-Nom. red block-Gen. front [back / front]-Loc. be
‘The green ball is in front [back / ahead] of the red block.’
(b) akai rippoutai-wa midori-no kyuu-no mae [ushiro / saki]-ni aru.
red block-Nom. green-Gen. ball-Gen. front [back / front]-Loc. be
‘The red block is in front [back / ahead] of the green ball.’

In each sentence, one of the three Japanese spatial terms mae, ushiro, or saki was used,
so that the sentences were all complete ones. Thus, we prepared twelve stimuli in total.
Two motion conditions {viewer-in-motion, objects-in-motion}, two sentence patterns
{the green ball as the subject, the red block as the subject}, three spatial lexemes {mae,
ushiro, saki} were fully crossed.
Thirty native speakers of Japanese participated in this experiment. They were
instructed to (1) read the Japanese sentence shown on the screen, (2) press the key
when they understood the sentence (then a three-dimensional computer graphic image
appeared on the screen), (3) look at the motion image, and (4) rate how the sentence
matched the image in a 4-point scale (where -2=complete mismatch, +2=complete
match), by pressing a key on the keyboard. When the subject pressed a key for rating,
the next sentence appeared on the screen and the process (1) to (4) was repeated. The
twelve stimuli were presented in a random order for each subject.

4.2 Results

The data obtained were categorized into two classes, positive responses (+2 and +1)
and negative responses (-2 and -1). The numbers of responses for each category were
counted and the total numbers for each condition were statistically analyzed using the
Chi-square test.
First, we considered the use of mae (front). The arrangement of the two objects, as
shown in Figure 7, matches the sentence ‘the ball is in mae of the block’ in the unmarked,
 FRAMES OF REFERENCE, EFFECTS OF MOTION, AND LEXICAL MEANINGS OF JAPANESE FRONT/BACK TERMS 305

normal interpretation, but does not match the sentence ‘the block is in mae of the ball’
(See Section 3). Hence, it is expected that, for the sentence ‘the block is in mae of the
ball’, negative responses will dominate. We examined this case, that is, the case where
negative responses are expected to dominate. If motion conditions did not affect how
this spatial arrangement is perceived and expressed using mae, this expectation (the
dominance of negative responses) would be satisfied equally under the viewer-in-motion
condition and the object-in-motion condition. However, as shown in Figure 8, the
viewer-in-motion condition received significantly more positive responses than the
object-in-motion condition (Chi-square(1)=7.72, p< .001). That is, the farther side of
the ground object, which is not normally thought of as mae (front) of the ground object,
was judged as mae of the ground object more frequently under the viewer-in-motion
condition than under the object-in-motion condition.

25

20

15
Positive
Negative
10

0
Viewer-in motion Object-in-motion

Figure 8. The block is in mae of the ball.

Next, we examined the use of ushiro (back). In Figure 7, the block is normally said
to be in ushiro of the ball but the ball is not said to be in ushiro of the block in static
situation. Consequently, it is expected that the sentence ‘the ball is in ushiro of the
block’ will receive dominantly negative responses. If motion conditions did not affect
how this spatial arrangement is perceived and expressed using ushiro, this expectation
(dominance of negative responses) would be satisfied equally for the viewer-in-motion
condition and the object-in-motion condition. However, as Figure 9 shows, the viewer-
in-motion condition received significantly more positive responses than the object-in-
motion condition (Chi-square(1)=6.66, p< .01). That is, the nearer side, which is not
normally regarded as ushiro (back) of the ground object, was judged as ushiro of the
ground object more frequently under the viewer-in-motion condition than under the
object-in-motion condition.
306 LANGUAGE, COGNITION AND SPACE

30

25

20

Positive
15
Negative
10

0
Viewer-in motion Object-in-motion

Figure 9. The ball is in ushiro of the block

In the third test, saki exhibits quite different results. In Figure 7, the block is normally
said to be in saki (front/ahead) of the ball but the ball is not in saki of the block in static
situation. Therefore, it is expected that the sentence ‘the ball is in saki of the block’ will
receive dominantly negative responses. If motion condition did not affect how this
spatial arrangement is perceived and expressed using saki, this expectation (dominance
of negative responses) would be satisfied equally for the viewer-in-motion condition
and the object-in-motion condition. However, the object-in-motion condition received
significantly more positive responses than the viewer-in-motion condition as Figure 10
shows (Chi-square(1)=6.98, p< .01). That is, the nearer side of the ground object was
judged as saki (front/ahead) of the ground object more frequently under the object-in-
motion condition than under the viewer-in-motion condition.
30

25

20

Positive
15
Negative
10

0
Viewer-in-motion Object-in-motion

Figure 10. The ball is in saki of the block

 FRAMES OF REFERENCE, EFFECTS OF MOTION, AND LEXICAL MEANINGS OF JAPANESE FRONT/BACK TERMS 307

Thus, we obtained different effects of motion for the three Japanese spatial lexemes.
Table 1 summarizes these results.
Table 1. Summary of the results

Motion condition that received more positive responses for non-standard expressions

mae Viewer-in-motion
ushiro Viewer-in-motion
saki Object-in-motion

4.3 Discussion

As shown in the previous section, the two motion conditions, the viewer-in-motion
condition and the object-in-motion condition, affect the subjects’ responses. The viewer-
in-motion condition, compared with the object-in-motion condition, induced greater
positive judgments for non-standard, unusual uses of mae and ushiro, while the object-
in-motion condition, compared with the viewer-in-motion condition, induced greater
positive responses for those of saki (see Table 1). Why do these motion conditions affect
the responses in such different ways?
A likely answer to this question may be that the direction of motion perceived in
visual context is projected onto the ground object. For example, if the viewer is moving
toward the objects, the direction of this frontward motion may be projected onto
the ground object. An illustration of this is given in Figure 11. The viewer is moving
in the forward direction, toward the ground object (block). This is indicated by the
black arrow. As already explained in Section 3, the unmarked, dominant construal
of an object’s orientation is based on the reflection frame when the term mae (front)
or ushiro (back) is used. Hence, the mae side of the block is normally the nearer
side to the viewer (indicated by the bright arrow in Figure 11; this is the unmarked
orientation of mae) and ushiro is the opposite side. Onto this block, the direction of
the viewer’s frontward motion is projected (indicated by the broken arrow and the
dark arrow on the right side of the block). Thus, the block obtains the same direction
as the viewer’s motion. This projection, it seems, does not override the unmarked
orientation of the object. Since our data indicate that half of the subjects responded
positively to the sentence that designates the nearer side of the block as mae of the
block (see Figure 8), the unmarked construal of mae of the ground object is not
totally cancelled by the projection of the direction of viewer’s motion onto the ground
object. Rather, it suggests that the ground object obtains the projected direction in
addition to the unmarked one.
308 LANGUAGE, COGNITION AND SPACE

FRONT mae BACK ushiro

(FRONT mae)

Ground object
Viewer

Figure 11. The effect of viewer’s motion on the ground object

The above case is based on the viewer-in-motion condition. How about, then, the object-
in-motion condition? The effect of the object-in-motion condition is quite different.
Figure 12 illustrates the different effect. In this case, there is no viewer’s motion, but
instead, the ground object moves toward the viewer. This motion defines the moving
object’s front side (indicated by the black arrow in Figure 12). As stated above, the
unmarked, dominant construal of the ground object’s orientation is defined by the
reflection frame when the term mae or ushiro is used (indicated by the bright arrow).
Hence, in this case, the frontal direction given by the block’s motion toward the viewer
coincides with the frontal direction given by the reflection frame. Thus, the motion
of the block does not add a new direction but just reinforces the unmarked frontal
orientation of the block. This explains why the farther side of the block was judged not
as mae but as ushiro of the block by most of the subjects under the object-in-motion
condition (Figure 8, 9).

FRONT mae BACK ushiro

Viewer Ground object

Figure 12. The effect of object’s motion on the ground object

Thus, our data for mae and ushiro can be explained in terms of the effect of motion on
the orientation of the ground object. The viewer-in-motion condition affects positively
for non-standard expressions of mae and ushiro because the viewer’s motion can give
the opposite direction to the ground object.
The same mechanism seems to work for saki, but the apparent effects look different
because the unmarked orientation defined by saki is based on a different frame, not the
reflection but the translation frame. As the bright arrow in Figure 13 illustrates, saki of
the block is the farther side of the block from the viewer’s point of view. This is because
the unmarked orientation of the object is based on the translation frame in the case of
 FRAMES OF REFERENCE, EFFECTS OF MOTION, AND LEXICAL MEANINGS OF JAPANESE FRONT/BACK TERMS 309

saki. The black arrow in Figure 13 indicates the direction of the viewer’s motion that is
projected onto the block. The block, then, obtains the same direction as the viewer’s, as
indicated by the dark arrow. As the Figure 13 shows, this projected direction and the
unmarked construal of saki of the block coincide. Thus, the viewer’s motion does not
add a new direction but just reinforces the unmarked construal of saki. This explains
why most subjects responded negatively to the sentence that designates the nearer side
of the block as saki under the viewer-in-motion condition.
Under the object-in-motion condition, however, the direction of motion of the
object and the direction designated by the unmarked use of saki oppose each other as
illustrated in Figure 14. (The bright arrow indicates the unmarked orientation of the
ground object in the case of saki.) Since a moving object obtains a front axis defined
by the direction of motion, the ground object, the block, obtains the direction toward
the viewer as its front. (The black arrow indicates this direction.) Thus, both the farther
side and the nearer side of the block can be the saki (front) of the block. This explains
why about half of the subjects responded positively to the nearer side of the block being
called saki of the block (Figure 10).

FRONT saki

Ground object
Viewer

Figure 13. The effect of viewer’s motion on the ground object

FRONT saki

Viewer Ground object

Figure 14. The effect of object’s motion on the ground object

In this way, the result of our experiment can be explained if we assume that (1) a moving
object obtains frontal direction defined by the direction of that motion, (2) the direc-
tion of the viewer’s motion can be projected onto the objects being observed, and (3)
mae, ushiro, and saki lexically designate certain frames of reference that determine the
unmarked frontal orientation of the ground object (as discussed in Section 3).
310 LANGUAGE, COGNITION AND SPACE

4.4 Implication for lexical meanings

In Section 3, we have demonstrated that the three Japanese spatial lexemes mae, ushiro,
and saki, include specification of frames of reference (either the reflection frame or the
translation frame within the relative frame of reference) for the cases where the ground
object has no intrinsic front/back axis. The question is, then, whether these lexemes are
rich enough to include specific, concrete spatial regions or positions as their senses.
The results of our experiment indicate that what these lexemes denote may not be such
concrete information about spatial regions or positions but may be only the basic setting
of front/back orientation of the ground object.
Support for this argument comes from the effect of motion on the interpretation of
the spatial relations of the ball and the block. As discussed in the previous section, the
results of our experiment can be explained if we assume that the motion of the viewer
or of the objects can affect the orientation of the reference object ((1) and (2) in Section
4.3.), and that each of the three lexemes designates a certain frame of reference ((3) in
Section 4.3). We also assume that frames of reference determine the orientation (frontal
direction) of the ground object.
If, however, these lexemes denote specific, concrete regions or positions relative
to the ground object, our results cannot be explained in such a simple manner. This is
because the concepts of REGION and POSITION may be of a quite different kind than
MOTION. Though it seems reasonable to assume that the concept of MOTION includes
the conceptual element of DIRECTION, and thus it seems natural that motion can affect
direction, it is difficult to explain why direction can change the specification of REGION
or POSITION. In short, interaction between two directions is far more intelligible than
interaction between DIRECTION and REGION or POSITION. Evidence also comes
from the present authors’ previous studies (Shinohara, Kojima and Matsunaka 2004a,
b). In these previous studies we carried out a similar kind of experiment but compared
the viewer-in-motion condition and the static condition. We obtained the results that
the viewer-in-motion condition works adversely for the standard, unmarked use of mae
(front). When a ball and a block were placed as in Figure 15, the ball was dominantly
(about 91%) judged as being mae of the block, but the viewer-in-motion condition
made this judgment significantly lower. That is, significantly greater numbers of negative
responses were obtained for the sentence ‘The ball is in mae of the block’ under the
viewer-in-motion condition.

FRONT (mae of the block)

Figure object

Figure 15. Arrangement of objects in Shinohara et al. (2004a, b)

 FRAMES OF REFERENCE, EFFECTS OF MOTION, AND LEXICAL MEANINGS OF JAPANESE FRONT/BACK TERMS 311

If mae denotes the region that is nearer to the viewer from the ground object, i.e.
between the viewer and the ground object, then it is quite difficult to explain why this
region is less likely to be called mae of the block when the motion of the viewer comes
into perceptual context. By contrast, if we assume that the meaning of mae defines the
orientation (frontal direction in this case) of the block, it becomes understandable that
some other motion that orients the object in the opposite direction can have an adverse
effect. This explains the increase of negative responses to the ball being called mae of
the block under the viewer-in-motion condition. The same phenomenon (increase of
negative responses to the unmarked, normal uses) is seen for ushiro and saki as well
(Shinohara, Kojima and Matsunaka 2004a, b).
To sum up, we suggest the following three points: (1) what frames of reference do is
to determine the (frontal) orientation of the ground object; (2) the spatial terms under
examination ‘prompt’ (Evans 2004: 54) certain spatial frames of reference rather than
denote specific regions or positions; (3) the effect of motion included in perceptual
context is not a part of the lexical meanings of these terms, but it is a kind of contextual
elaboration.
The second and third points especially concern Tyler and Evans’s (2003) and Evans’s
(2004) argument about lexical meanings. As we have shown, it is not reasonable to
attribute contextual effects to lexical meanings. We cannot specify all the different
motion conditions that can affect the uses of spatial lexemes. Nor can we describe
all the specific effects of motion as parts of the lexical properties because, as we have
demonstrated, such effects can only be described as a tendency in certain perceptual
contexts. The effect of motion is not truth-conditional: it only provides different pos-
sibilities of spatial construal with each expression. Each lexeme can be elaborated and
has various possibilities of interpretation when contexts permit. Thus, we support the
position taken by Tyler and Evans.
What strikes us is that the visual-perceptual context like the one we examined in
this study does not constitute the linguistic context that Tyler and Evans (2003) treat
as the material of lexical elaboration. Still, as we have seen, such context can affect the
interpretation of sentences so strongly that the totally opposite spatial direction can be
referred to by the same spatial lexeme. Such an effect might reside not at the linguistic
level but in a deeper cognitive level, and it may be that spatial lexemes such as the ones
we examined in this study have radial structures that include prototypes as the core
senses and gradually diffusing peripheral members, of which the latter may be more
susceptible to such cognitive-level influence of perceptual contexts. Tyler and Evans’s
analysis seems to concern the core senses, and we expect that our findings can add to
their theory.

5 Conclusion

In this paper we have examined the meanings of three Japanese FRONT/BACK terms:
mae, ushiro, and saki. After reviewing previous studies in Section 2, we have shown in
Section 3 that these lexemes have different specifications of frames of reference. Each
312 LANGUAGE, COGNITION AND SPACE

of the lexemes has, as part of its lexical properties, at least some information about the
unmarked frame of reference to refer to. Thus we support Levinson’s (2003) position
that there exists a certain degree of lexical specification of frames of reference, rather
than Svorou’s (1994) and Carlson-Radvansky and Irwin’s (1993) position that rejects
lexically specified frames of reference.
In Section 4, we have demonstrated that these spatial lexemes, when used for ground
objects that have no intrinsic directions or prominent axes, do not designate specific,
concrete spatial regions or positions. If the perceptual context includes the motion of
the viewer or of the objects, the direction of that motion can be added onto the ground
object, and thus, interpretation of the orientation of the ground object can vary. We
claim that these effects and the consequent interpretations about spatial relations are not
included in the lexical meanings of these words, but are a kind of contextual elaboration
(in a broad sense). This position is consonant with what Tyler and Evans (2003) and
Evans (2004) have suggested, i.e. the claim that lexical meanings should not include
contextual elaboration but rather be as narrow as can be.
In conclusion, we suggest that the three Japanese spatial lexemes, mae, ushiro,
and saki, can be made semantically narrow by eliminating the conceptual properties
of REGION and POSITION, but that they must have at least a certain specification of
unmarked frame of reference. Effects of motion can be observed and these lexemes
can have various uses in actual perceptual contexts, but this may be the consequence
of contextual elaboration in a broad sense. Thus, we have combined Levinson’s (2003)
position concerning linguistic specification of frames of reference and Tyler and
Evans’s (2003) position that lexical meanings and contextual elaboration should be
distinguished.

Notes
* This study is supported in part by a Grant-in-Aid for Scientific Research from the
Ministry of Education, Culture, Sports, Science and Technology of Japan, Grant No.
16500159. We would like to thank the participants at the 9th International Cogni-
tive Linguistics Conference for helpful comments and encouragement, as well as an
anonymous reviewer for very helpful comments. All remaining shortcomings are
ours.
** The mechanical parts of our experiment, i.e. computer graphics and the automated
data-output, were programmed by Takatsugu Kojima. We express deep thanks to him.

References
Carlson-Radvansky, L. A. and Irwin, D. E. (1993) Frames of reference in vision
and language: Where is above? Cognition 46: 223–244.
Clark, H. (1973) Space, time, semantics, and the child. In T. E. Moore (ed.)
Cognitive Development and the Acquisition of Language 27–63. New York:
Academic Press.
 FRAMES OF REFERENCE, EFFECTS OF MOTION, AND LEXICAL MEANINGS OF JAPANESE FRONT/BACK TERMS 313

Evans, V. (2004) The Structure of Time. Amsterdam: John Benjamins.

Harris, L. and Strommen, E. (1972) Understanding ‘FRONT’, ‘BACK’, and
‘Beside’: Experiments on the meaning of spatial concepts. In M. H. Siegel and
P. Zeigler (eds) Psychological Research: The Inside Story 198–212. New York:
Harper and Row.
Hill, C. (1975) Variation in the use of ‘front’ and ‘back’ by bilingual speakers.
Proceedings of the First Annual Meeting of the Berkeley Linguistics Society
196–206.
Hill, C. (1978) Linguistic representation of spatial and temporal orientation.
Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society
524–539.
Hill, C. (1982) Up/down, front/back, left/right: A contrastive study of Hausa and
English. In J. Weissenborn and W. Klein (eds) Here and There: Cross-linguistic
Studies on Deixis and Demonstration 11–42. Amsterdam: Benjamins.
Imai, M. and Ishizaki, S. (1999) Mae, ushiro, hidari, migi-no imi (The meanings
of front/back, right and left). Keio SFC Review 4: 81–88.
Imai, M., Nakanishi, T., Miyashita, H., Kidachi, Y. and Ishizaki, S. (1999) The
meanings of FRONT/BACK/LEFT/RIGHT. Cognitive Studies 6(2): 207–225.
Inoue, K. (1998) Moshi migi-ya hidari-ga nakattara (If there were no rights or
lefts). Tokyo: Taishukan Shoten.
Inoue, K. (2002) Zettai to Soutai no Hazama de – Kuukan Shijiwaku niyoru
Communication (Between absoluteness and relativity – communication
through spatial reference frame). In T. Ohori (ed.) Ninchi Gengogaku II
– Categorization (Cognitive linguistics II – categorization) 11–35. Tokyo:
University of Tokyo Press.
Inoue, K. (2005) Kuukan Ninchi to communication (Spatial cognition and
communication). In S. Ide and M. Hiraga (eds) Ibunka to Communication
(Different culture and communication) 118–128. Tokyo: Hitsuji Shobo.
Kataoka, K. (2003) Kanban kookoku-ni miru kuukan shijiwaku-no hen’i-
nitsuite (Variation of spatial frames of reference in commercial signboards).
Paper presented at The Eleventh Conference of Japanese Association of
Sociolinguistic Sciences.
Lakoff, G. (1987) Woman, Fire and Dangerous Things: What Categories Reveal
about the Mind. Chicago: Chicago University Press.
Langacker, R. (1987) Foundations of Cognitive Grammar Vol. 1. Stanford, CA:
Stanford University Press.
Levelt, W. J. M. (1996) Perspective taking and ellipsis in spatial descriptions. In
P. Bloom, M. Peterson, L. Nadel and M. Garrett (eds) Language and Space
77–108. Cambridge, MA: MIT Press.
Levinson, S. C. (1996) Frames of reference and Molyneaux’s questions:
Crosslinguistic evidence. In P. Bloom, M. Peterson, L. Nadel and M. Garrett
(eds) Language and Space 109–169. Cambridge, MA: MIT Press.
314 LANGUAGE, COGNITION AND SPACE

Levinson, S. C. (2003) Space in Language and Cognition. Cambridge: Cambridge

University Press.
Matsunaka, Y. and Shinohara, K. (2004) Spatial cognition and language of
space: A perspective from Japanese. In A. Soares da Silva, A. Torres and
M. Gonçalves (eds) Linguagem, Cultura e Cognição: Estudos de Linguística
Cognitiva 59–74. Coimbra: Almedina.
Matsunaka, Y. and Shinohara, K. (2005) Cognition and language of space:
Japanese words of FRONT. Paper presented at New Directions in Cognitive
Linguistics: First UK Cognitive Linguistics Conference.
Moore, K. E. (2000) Spatial experience and temporal metaphors in Wolof: Point
of view, conceptual mapping, and linguistic practice. PhD dissertation,
Department of Linguistics, University of California, Berkeley.
Odate, J., Shinohara, K. and Matsunaka, Y. (2003) Nihongo-ni-okeru mae-to
ushiro-no ninchi-to hyougen (Cognition and expression of front and back
in Japanese). Paper presented at The Eleventh Conference of Japanese
Association of Sociolinguistic Sciences.
Shinohara, K., Matsunaka, Y. and Odate, J. (2003) Cognition and expression of
FRONT and BACK in Japanese. Paper presented at The Eighth International
Cognitive Linguistics Conference.
Shinohara, K., Kojima, T. and Matsunaka, Y. (2004a) Lexical variation in frames
of reference: An empirical study of Japanese space terms. Paper presented at
the International Conference on Language, Culture and Mind.
Shinohara, K., Kojima, T. and Matsunaka, Y. (2004b) Nihongo-no kuukan goi
‘mae, ushiro, saki and temae’ to sono sanshou-waku-ni-kansuru jikken-teki
kenkyu (Japanese spatial terms ‘mae (front), ushiro (back), saki (front/ahead),
and temae (front-of-hand)’ and experimental study of their frames of refer-
ence). Paper presented at The Twenty-first Conference of Japanese Cognitive
Science Society.
Shinohara, K. and Matsunaka, Y. (2004) Spatial cognition and linguistic expres-
sion: Empirical research on frames of reference in Japanese. Annual Review of
Cognitive Linguistics 3: 261–283.
Shinohara, K., Odate, J. and Matsunaka, Y. (2003) ‘Mae’ to ‘ushiro’ no imi-to
sanshouwaku: taiji-teki houryaku-to seiretsu-teki houryku-no arawarekata
(Semantics and frame of reference of front and back: How Japanese spatial
terms adopt ego-opposed strategy and ego-aligned strategy). Paper presented
at The Twentieth Conference of Japanese Cognitive Science Society.
Svorou, S. (1994) The Grammar of Space. Amsterdam: John Benjamins.
Talmy, L. (1978) Figure and ground in complex sentences. In J. H. Greenberg
(ed.) Universals of Human Language 4: Syntax 625–649. Stanford: Stanford
University Press.
Talmy, L. (1983) How language structures space. In H. L. Pick and L. Acredolo
(eds) Spatial Orientation 225–82. New York: Plenum Press.
Talmy, L. (2000) Toward a Cognitive Semantics Vol. 1. Cambridge, MA: MIT Press.
 FRAMES OF REFERENCE, EFFECTS OF MOTION, AND LEXICAL MEANINGS OF JAPANESE FRONT/BACK TERMS 315

Tyler, A. and Evans, V. (2003) The Semantics of English Prepositions: Spatial Scenes,
Embodied Meaning, and Cognition. Cambridge: Cambridge University Press.
Vandeloise, C. (1991) Spatial Prepositions: A Case Study from French. Chicago and
London: The University of Chicago Press.
Yoshida, A. (2003) Nihonjin-no kuukan-ninchi-to mae-no imi (Spatial cogni-
tion of Japanese and the meaning of ‘mae’). Senior thesis, Department of
Computer, Information and Communication Sciences, Tokyo University of
Agriculture and Technology.
 Part VI
Space in sign-language and gesture

317
13 How spoken language and signed language
structure space differently
Leonard Talmy

1 Introduction1

This paper combines and relates new findings on spatial structuring in two areas of
investigation, spoken language and signed language. Linguistic research to date has
determined many of the factors that structure the spatial schemas found across spoken
languages (e.g. Gruber 1965, Fillmore 1968, Leech 1969, Clark 1973, Bennett 1975,
Herskovits 1982, Jackendoff 1983, Zubin and Svorou 1984, as well as myself, Talmy
1983, 2000a, 2000b). It is now feasible to integrate these factors and to determine the
comprehensive system they constitute for spatial structuring in spoken language. This
system is characterized by several features. With respect to constituency, there is a
relatively closed universally available inventory of fundamental spatial elements that
in combination form whole schemas. There is a relatively closed set of categories that
these elements appear in. And there is a relatively closed small number of particular
elements in each category, hence, of spatial distinctions that each category can ever mark.
With respect to synthesis, selected elements of the inventory are combined in specific
arrangements to make up the whole schemas represented by closed-class spatial forms.
Each such whole schema that a closed-class form represents is thus a ‘prepackaged’
bundling together of certain elements in a particular arrangement. Each language has in
its lexicon a relatively closed set of such pre-packaged schemas (larger than that of spatial
closed-class forms, due to polysemy) that a speaker must select among in depicting a
spatial scene. Finally, with respect to the whole schemas themselves, these schemas
can undergo a certain set of processes that extend or deform them. Such processes are
perhaps part of the overall system so that a language’s relatively closed set of spatial
schemas can fit more spatial scenes.
An examination of signed language 2 shows that its structural representation of
space systematically differs from that in spoken language in the direction of what appear
to be the structural characteristics of scene parsing in visual perception. Such differ-
ences include the following: Signed language can mark finer spatial distinctions with
its inventory of more structural elements, more categories, and more elements per
category. It represents many more of these distinctions in any particular expression. It
also represents these distinctions independently in the expression, not bundled together
into pre-packaged schemas. And its spatial representations are largely iconic with visible
spatial characteristics.
When formal linguistic investigation of signed language began several decades
ago, it was important to establish in the context of that time that signed language was

319
320 LANGUAGE, COGNITION AND SPACE

in fact a full genuine language, and the way to do this, it seemed, was to show that it fit
the prevailing model of language, the Chomskyan-Fodorian language module. Since
then, however, evidence has been steadily accruing that signed language does diverge
in various respects from spoken language. The modern response to such observa-
tions – far from once again calling into question whether signed language is a genuine
language – should be to rethink what the general nature of language is. Our findings
suggest that instead of some discrete whole-language module, spoken language and
signed language are both based on some more limited core linguistic system that then
connects with different further subsystems for the full functioning of the two different
language modalities.

2 Fundamental space-structuring elements and categories in

spoken language

An initial main finding emerges from analysis of the spatial schemas expressed by
closed-class (grammatical) forms across spoken languages. There is a relatively closed
and universally available inventory of fundamental conceptual elements that recombine
in various patterns to constitute those spatial schemas. These elements fall within a
relatively closed set of categories, with a relatively closed small number of elements
per category.

2.1 The target of analysis

As background to this finding, spoken languages universally exhibit two different

subsystems of meaning-bearing forms. One is the ‘open-class’ or ‘lexical’ subsystem,
comprised of elements that are great in number and readily augmented – typically, the
roots of nouns, verbs, and adjectives. The other is the ‘closed-class’ or ‘grammatical’
subsystem, consisting of forms that are relatively few in number and difficult to aug-
ment – including such bound forms as inflections and such free forms as prepositions
and conjunctions. As argued in Talmy (2000a, ch. 1), these subsystems basically perform
two different functions: open-class forms largely contribute conceptual content, while
closed-class forms determine conceptual structure. Accordingly, our discussion focuses
on the spatial schemas represented by closed-class forms so as to examine the concepts
used by language for structuring purposes.
Across spoken languages, only a portion of the closed-class subsystem regularly
represents spatial schemas. We can identify the types of closed-class forms in this
portion and group them according to their kind of schema. The types of closed-class
forms with schemas for paths or sites include the following: (1) forms in construction
with a nominal, such as prepositions like English across (as in across the field) or noun
affixes like the Finnish illative suffix -:n ‘into’, as well as prepositional complexes such as
English in front of or Japanese constructions with a ‘locative noun’ like ue ‘top surface’
(as in teeburu no ue ni ‘table GEN top at’ = ‘on the table’); (2) forms in construction
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 321

with a verb, such as verb satellites like English out, back and apart (as in They ran out
/ back / apart); (3) deictic determiners and adverbs such as English this and here; (4)
indefinites, interrogatives, relatives, etc., such as English everywhere/whither / wherever);
(5) qualifiers such as English way and right (as in It’s way / right up there); and (6)
adverbials like English home (as in She isn’t home).
Types of closed-class forms with schemas for the spatial structure of objects include
the following: (1) forms modifying nominals such as markers for plexity or state of
boundedness, like English -s for multiplexing (as in birds) or -ery for debounding (as
in shrubbery); (2) numeral classifiers like Korean chang ‘planar object’; and (3) forms
in construction with the verb, such as some Atsugewi Cause prefixes, like cu- ‘as the
result of a linear object moving axially into the Figure’.
Finally, sets of closed-class forms that represent a particular component of a spatial
event of motion/location include the following: (1) the Atsugewi verb-prefix set that
represents different Figures; (2) the Atsugewi verb-suffix set that represents different
Grounds (together with Paths); (3) the Atsugewi verb-prefix set that represents different
Causes; and (4) the Nez Perce verb-prefix set that represents different Manners (see
Talmy 2000b, ch. 1 and 2).

2.2 Determining the elements and categories

A particular methodology is used to determine fundamental spatial elements in lan-

guage. One starts with any closed-class spatial morpheme in any language, considering
the full schema that it expresses and a spatial scene that it can apply to. One then
determines any factor one can change in the scene so that the morpheme no longer
applies to it. Each such factor must therefore correspond to an essential element in
the morpheme’s schema. To illustrate, consider the English preposition across and the
scene it refers to in The board lay across the road. Let us here grant the first two elements
in the across schema (demonstrated elsewhere): (1) a Figure object (here, the board) is
spatially related to a Ground object (here, the road); and (2) the Ground is ribbonal – a
plane with two roughly parallel line edges that are as long as or longer than the distance
between them. The remaining elements can then be readily demonstrated by the
methodology. Thus, a third element is that the Figure is linear, generally bounded at
both ends. If the board were instead replaced by a planar object, say, some wall siding,
one could no longer use the original across preposition but would have to switch to
the schematic domain of another preposition, that of over, as in The wall siding lay
over the road. A fourth element is that the axes of the Figure and of the Ground are
roughly perpendicular. If the board were instead aligned with the road, one could no
longer use the original across preposition but would again have to switch to another
preposition, along, as in The board lay along the road. Additionally, a fifth element
of the across schema is that the Figure is parallel to the plane of the Ground. In the
referent scene, if the board were tilted away from parallel, one would have to switch
to some other locution such as The board stuck into / out of the road. A sixth element
is that the Figure is adjacent to the plane of the Ground. If the board were lowered
322 LANGUAGE, COGNITION AND SPACE

or raised away from adjacency, even while retaining the remaining spatial relations,
one would need to switch to locutions like The board lay (buried) in the road. / The
board was (suspended) above the road. A seventh element is that the Figure’s length
is at least as great as the Ground’s width. If the board were replaced by something
shorter, for example, a baguette, while leaving the remaining spatial relations intact,
one would have to switch from across to on, as in The baguette lay on the road. An
eighth element is that the Figure touches both edges of the Ground. If the board in
the example retained all its preceding spatial properties but were shifted axially, one
would have to switch to some locution like One end of the board lay over one edge of
the road. Finally, a ninth element is that the axis of the Figure is horizontal (the plane
of the Ground is typically, but not necessarily, horizontal). Thus, if one changes the
original scene to that of a spear hanging on a wall, one can use across if the spear is
horizontal, but not if it is vertical, as in The spear hung across the wall. / The spear hung
up and down on the wall. Thus, from this single example, the methodology shows that
at least the following elements figure in closed-class spatial schemas: a Figure and a
Ground, a point, a line, a plane, a boundary (a point as boundary to a line, a line as
boundary to a plane), parallelness, perpendicularity, horizontality, adjacency (contact),
and relative magnitude.
In the procedure of systematically testing candidate factors for their relevance, the
elements just listed have proved to be essential to the selected schema and hence, to
be in the inventory of fundamental spatial elements. But it is equally necessary to note
candidates that do not prove out, so as to know which potential spatial elements do
not serve a structuring function in language. In the case of across, for example, one can
probe whether the Figure, like the board in the referent scene, must be planar – rather
than simply linear – and coplanar with the plane of the Ground. It can be seen, though,
that this is not an essential element to the across schema, since this factor can be altered
in the scene by standing the board on edge without any need to alter the preposition,
as in The board lay flat / stood on edge across the road. Thus, coplanarity is not shown
by across to be a fundamental spatial element. However, it does prove to be so in other
schemas, and so in the end must be included in the inventory. This is seen for one of the
schemas represented by English over, as in The tapestry hung over the wall. Here, both
the Figure and Ground must be planes and coplanar with each other. If the tapestry here
were changed to something linear, say, a string of beads, it is no longer appropriate to
use over but only something like against, as in The string of beads hung *over / against the
wall. Now, another candidate element – that the Figure must be rigid, like the board in
the scene – can be tested and again found to be inessential to the across schema, since
a flexible linear object can be substituted for the board without any need to change the
preposition, as seen in The board/The cable lay across the road. Here, however, checking
this candidate factor across numerous spatial schemas in many languages might well
never yield a case in which it does figure as an essential element and so would be kept
off the inventory.
This methodology affords a kind of existence proof: it can demonstrate that some
element does occur in the universally available inventory of structural spatial elements
since it can be seen to occur in at least one closed-class spatial schema in at least one
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 323

language. The procedure is repeated numerous times across many languages to build
up a sizable inventory of elements essential to spatial schemas.
The next step is to discern whether the uncovered elements comprise particular
structural categories and, if so, to determine what these categories are. It can be observed
that for certain sets of elements, the elements in a set are mutually incompatible – only
one of them can apply at a time at some point in a schema. Such sets are here taken
to be basic spatial categories. Along with their members, such categories are also part
of language’s fundamental conceptual structuring system for space. A representative
sample of these categories is presented next.
It will be seen that these categories generally have a relatively small membership.
This finding depends in part on the following methodological principles. An element
proposed for the inventory should be as coarse-grained as possible – that is, no more
specific than is warranted by cross-schema analysis. Correlatively, in establishing a
category, care must be taken that it include only the most generic elements that have
actually been determined – that is, that its membership have no finer granularity than
is warranted by the element-abstraction procedure. For example, the principle of
mutual incompatibility yields a spatial category of ‘relative orientation’ between two
lines or planes, a category with perhaps only two member elements (both already seen
in the across schema): approximately parallel and approximately perpendicular. Some
evidence additionally suggests an intermediary ‘oblique’ element as a third member
of the category. Thus, some English speakers may distinguish a more perpendicular
sense from a more oblique sense, respectively, for the two verb satellites out and off, as
in A secondary pipe branches out / off from the main sewer line. In any case, though, the
category would have no more than these two or three members. Although finer degrees
of relative orientation can be distinguished by other cognitive systems, say, in visual
perception and in motor control, the conceptual structuring subsystem of language
does not include anything finer than the two- or three-way distinction. The procedures
of schema analysis and cross-schema comparison, together with the methodological
principles of maximum granularity for elements and for category membership, can lead
to a determination of the number of structurally distinguished elements ever used in
language for a spatial category.

2.3 Sample categories and their member elements

The fundamental categories of spatial structure in the closed-class subsystem of spoken

language fall into three classes according to the aspect of a spatial scene they pertain
to: the segmentation of the scene into individual components, the properties of an
individual component, and the relations of one such component to another. In a fourth
class are categories of nongeometric elements frequently found in association with
spatial schemas. A sampling of categories and their member elements from each of
these four classes is presented next. The examples provided here are primarily drawn
from English but can be readily multiplied across a diverse range of languages (see
Talmy 2000a, ch. 3).
324 LANGUAGE, COGNITION AND SPACE

2.3.1 Categories pertaining to scene segmentation

The class designated as scene segmentation may include only one category, that of
‘major components of a scene’, and this category may contain only three member
elements: the Figure, the Ground, and a Secondary Reference Object. Figure and
Ground were already seen for the across schema. Schema comparison shows the
need to recognize a third scene component, the Secondary Reference Object – in
fact, two forms of it: encompassive of or external to the Figure and Ground. The
English preposition near, as in The lamp is near the TV specifies the location of the
Figure (the lamp) only with respect to the Ground (the TV). But localizing the Figure
with the preposition above, as in The lamp is above the TV, requires knowledge not
only of where the Ground object is, but also of the encompassive earth-based spatial
grid, in particular, of its vertical orientation. Thus, above requires recognizing three
components within a spatial scene, a Figure, a Ground, and a Secondary Reference
Object of the encompassive type. Comparably, the schema of past in John is past the
border only relates John as Figure to the border as Ground. One could say this sentence
on viewing the event through binoculars from either side of the border. But John is
beyond the border can be said only by someone on the side of the border opposite
John, hence the beyond schema establishes a perspective point at that location as a
Secondary Reference Object – in this case, of the external type.

2.3.2 Categories pertaining to an individual scene component

A number of categories pertain to the characteristics of an individual spatial scene

component. This is usually one of the three major components resulting from scene
segmentation – the Figure, Ground, or Secondary Reference Object – but it could be
others, such as the path line formed by a moving Figure. One such category is that of
‘dimension’ with four member elements: zero dimensions for a point, one for a line, two
for a plane, and three for a volume. Some English prepositions require a Ground object
schematizable for only one of the four dimensional possibilities. Thus, the schema of the
preposition near as in near the dot requires only that the Ground object be schematizable
as a point. Along, as in along the trail, requires that the Ground object be linear. Over
as in a tapestry over a wall requires a planar Ground. And throughout, as in cherries
throughout the jello, requires a volumetric Ground.
A second category is that of ‘number’ with perhaps four members: one, two, several,
and many. Some English prepositions require a Ground comprising objects in one or
another of these numbers. Thus, near requires a Ground consisting of just one object,
between of two objects, among of several objects, and amidst of numerous objects, as in
The basketball lay near the boulder / between the boulders / among the boulders / amidst
the cornstalks. The category of number appears to lack any further members – that is,
closed-class spatial schemas in languages around the world seem never to incorporate
any other number specifications – such as ‘three’ or ‘even-numbered’ or ‘too many’.
A third category is that of ‘motive state’, with two members: motion and stationari-
ness. Several English prepositions mark this distinction for the Figure. Thus, in one of
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 325

its senses, at requires a stationary Figure, as in I stayed / *went at the library, while into
requires a moving Figure, as in I went / *stayed into the library. Other prepositions mark
this same distinction for the Ground object (in conjunction with a moving Figure).
Thus, up to requires a stationary Ground (here, the deer), as in The lion ran up to the
deer, while after requires a moving Ground as in The lion ran after the deer. Apparently
no spatial schemas mark such additional distinctions as motion at a fast vs. slow rate,
or being located at rest vs. remaining located fixedly.
A fourth category is that of ‘state of boundedness’ with two members: bounded and
unbounded. The English preposition along requires that the path of a moving Figure
be unbounded, as shown by its compatibility with a temporal phrase in for but not in,
as in I walked along the pier for 10 minutes / *in 20 minutes. But the spatial locution the
length of requires a bounded path, as in I walked the length of the pier in 20 Minutes /
*for 10 minutes.3 While some spatial schemas have the bounded element at one end of a
line and the unbounded element at the other end, apparently no spatial schema marks
any distinctions other than the two cited states of boundedness. For example, there is no
cline of gradually increasing boundedness, nor a gradient transition, although just such
a ‘clinal boundary’ appears elsewhere in our cognition, as in geographic perception or
conception, e.g., in the gradient demarcation between full forest and full meadowland
(Mark and Smith, 2004).
Continuing the sampling of this class, a fifth category is that of ‘directedness’ with
two members: basic and reversed. A schema can require one or the other of these ele-
ments for an encompassive Ground object, as seen for the English prepositions in The
axon grew along / against the chemical gradient, or for the Atsugewi verb satellites for
(moving) ‘downstream’ and ‘upstream’. Or it can require one of the member elements
for an encompassive Secondary Reference Object (here, the line), as in Mary is ahead
of / behind John in line.
A sixth category is ‘type of geometry’ with two members: rectilinear and radial. This
category can apply to an encompassive Secondary Reference Object to yield reference
frames of the two geometric types. Thus, in a subtle effect, the English verb satellite away,
as in The boat drifted further and further away / out from the island, tends to suggest a
rectilinear reference frame in which one might picture the boat moving rightward along
a corridor or sea lane with the island on the left (as if along the x-axis of a Cartesian
grid). But out tends to suggest a radial reference frame in which the boat is seen moving
from a center point along a radius through a continuum of concentric circles. In the
type-of-geometry category, the radial-geometry member can involve motion about a
center, along a radius, or along a periphery. The first of these is the basis for a further
category, that of ‘orientation of spin axis’, with two members: vertical and horizontal.
The English verb satellites around and over specify motion of the Figure about a vertical
or horizontal spin axis, respectively, as in The pole spun around / toppled over and in I
turned the pail around / over.
A seventh category is ‘phase of matter’, with three main members: solid, liquid, and
empty space, and perhaps a fourth member, fire. Thus, among the dozen or so Atsugewi
verb satellites that subdivide the semantic range of English into plus a Ground object,
the suffix -ik’s specifies motion horizontally into solid matter (as chopping an ax into
326 LANGUAGE, COGNITION AND SPACE

a tree trunk), -ic’t specifies motion into liquid, -ipsnu specifies motion into the empty
space of a volumetric enclosure, and -caw specifies motion into a fire. The phase of
matter category even figures in some English prepositions, albeit covertly. Thus, in can
apply to a Ground object of any phase of matter, whereas inside can apply only to one
with empty space, as seen in The rock is in/inside the box; in / *inside the ground; in /
*inside the puddle of water; in / *inside the fire.
A final category in this sampled series is that of ‘state of consolidation’ with appar-
ently two members: compact (precisional) and diffuse (approximative). The English
locative prepositions at and around distinguish these two concepts, respectively, for
the area surrounding a Ground object, as in The other hiker will be waiting for you at /
around the landmark. The two deictic adverbs in The hiker will be waiting for you there/
thereabouts mark the same distinction (unless there is better considered neutral to the
distinction). And in Malagasy (Imai, 2003), two locative adverbs for ‘here’ mark this
distinction, with eto for ‘here within this bounded region’, typically indicated with a
pointing finger, and ety for ‘here spread over this unbounded region’, typically indicated
with a sweep of the hand. In addition to this sampling, some ten or so further categories
pertaining to properties of an individual schema component, each category with a small
number of fixed contrasts, can be readily identified.

2.3.3 Categories pertaining to the relation of one scene component to another

Another class of categories pertains to the relations that one scene component can bear
to another. One such category was described earlier, that of ‘relative orientation’, with two
or three members: parallel, perpendicular, and perhaps oblique. A second such category
is that of ‘degree of remove’, of one scene component from another. This category appears
to have four or five members, two with contact between the components – coincidence
and adjacency – and two or three without contact – proximal, perhaps medial, and distal
remove. Some pairwise contrasts in English reveal one or another of these member
elements for a Figure relating to a Ground. Thus, the locution in the front of, as in The
carousel is in the front of the fairground, expresses coincidence, since the carousel as
Figure is represented as being located in a part of the fairground as Ground. But in front
of (without a the) as in The carousel is in front of the fairground, indicates proximality,
since the carousel is now located outside the fairground and near it but not touching it.
The distinction between proximal and distal can be teased out by noting that in front of
can only represent a proximal but not a distal degree of remove, as seen in the fact that
one can say The carousel is 20 feet in front of the fairground, but not, *The carousel is 20
miles in front of the fairground, whereas above allows both proximal and distal degrees
of remove, as seen in The hawk is 1 foot / 1 mile above the table. The distinction between
adjacency and proximality is shown by the prepositions on and over, as in The fly is on
/ over the table. Need for a fifth category member of ‘medial degree of remove’ might
come from languages with a ‘here / there / yonder’ kind of distinction in their deictic
adverbs or demonstratives.
A third category in this series is that of ‘degree of dispersion’ with two members:
sparse and dense. To begin with, English can represent a set of multiple Figures, say,
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 327

0-dimensional peas, as adjacent to or coincident with a 1-, 2-, or 3-dimensional Ground,

say, with a knife, a tabletop, or aspic, in a way neutral to the presence or absence of
dispersion, as in There are peas on the knife; on the table; in the aspic. But in representing
dispersion as present, English can (or must) indicate its degree. Thus, a sparse degree of
dispersion is indicated by the addition of the locution here and there, optionally together
with certain preposition shifts, as in There are peas here and thereon / along the knife;
on / over the table; in the aspic. And for a dense degree of dispersion, English has the
three specialized forms all along, all over and throughout, as seen in There are peas all
along the knife; all over the table; throughout the aspic.
A fourth category is that of ‘path contour’ with perhaps some four members:
straight, arced, circular, and meandering. Some English prepositions require one or
another of these contour elements for the path of a Figure moving relative to a Ground.
Thus, across indicates a straight path, as seen in I drove across the plateau / *hill, while
over – in its usage referring to a single path line – indicates an arced contour, as in I
drove over the hill / *plateau. In one of its senses, around indicates a roughly circular
path, as in I walked around the maypole, and about indicates a meandering contour, as
in I walked about the town. Some ten or so additional categories for relating one scene
component to another, again each with its own small number of member contrasts,
can be readily identified.

2.3.4 Nongeometric categories

All the preceding elements and their categories have broadly involved geometric
characteristics of spatial scenes or the objects within them – that is, they have been
genuinely spatial. But a number of nongeometric elements are recurrently found in
association with otherwise geometric schemas. One category of such elements is that
of ‘force dynamics’ (see Talmy 2000a, ch. 7) with two members: present and absent.
Thus, geometrically, the English prepositions on and against both represent a Figure
in adjacent contact with a Ground, but in addition, on indicates that the Figure is
supported against the pull of gravity through that contact, while against indicates
that it is not, as seen in The poster is on / *against the wall and The floating helium
balloon is against / *on the wall. Cutting the conceptualization of force somewhat
differently (Bowerman 1996), the Dutch preposition op indicates a Figure supported
comfortably in a natural rest state through its contact with a Ground, whereas aan
indicates that the Figure is being actively maintained against gravity through contact
with the Ground, so that flesh is said to be ‘op’ the bones of a live person but ‘aan’ the
bones of a dead person.
A second nongeometric category is that of ‘accompanying cognitive/affective
state’, though its extent of membership is not clear. One recurrent member, however,
is the attitude toward something that it is unknown, mysterious, or risky. Perhaps in
combination with elements of inaccessibility or nonvisibility, this category member is
associated with the Figure’s location in the otherwise spatial indications of the English
preposition beyond, whereas it is absent from the parallel locution on the other side of,
as in He is beyond / on the other side of the border (both these locutions – unlike past
328 LANGUAGE, COGNITION AND SPACE

seen above –are otherwise equivalent in establishing a viewpoint location as an external

Secondary Reference Object).
A third nongeometric category – in the class that relates one scene component to
another – is that of ‘relative priority’, with two members: coequal and main/ancillary.
The English verb satellites together and along both indicate joint participation, as seen
in I jog together / along with him. But together indicates that the Figure and the Ground
are coequal partners in the activity, whereas along indicates that the Figure entity is
ancillary to the Ground entity, who would be assumed to engage in the activity even if
alone (see Talmy 2000b, ch. 3).

2.4 Properties of the inventory

By our methodology, the universally available inventory of structural spatial elements

includes all elements that appear in at least one closed-class spatial schema in at least
one language. These elements may indeed be equivalent in their sheer availability for
use in schemas. But beyond that, they appear to differ in their frequency of occurrence
across schemas and languages, ranging from very common to very rare. Accordingly,
the inventory of elements – and perhaps also that of categories – may have the property
of being hierarchical, with entries running from the most to the least frequent. Such a
hierarchy suggests asking whether the elements in the inventory, the categories in the
inventory, and the elements in each category form fully closed memberships. That is,
does the hierarchy end at a sharp lower boundary or trail off indefinitely? With many
schemas and languages already examined, our sampling method may have yielded all
the commoner elements and categories, but as the process slows down in the discovery
of the rarer forms, will it asymptotically approach some complete constituency and
distinctional limit in the inventory, or will it be able to go on uncovering sporadic novel
forms as they develop in the course of language change?
The latter seems likelier. Exotic elements with perhaps unique occurrence in one or
a few schemas in just one language can be noted, including in English. Thus, in referring
to location at the interior of a wholly or partly enclosed vehicle, the prepositions in and
on distinguish whether the vehicle lacks or possesses a walkway. Thus, one is in a car but
on a bus, in a helicopter but on a plane, in a grain car but on a train, and in a rowboat
but on a ship. Further, Fillmore has observed that this on also requires that the vehicle
be currently in use as transport: The children were playing in / *on the abandoned bus
in the junkyard. Thus, schema analysis in English reveals the element ‘(partly) enclosed
vehicle with a walkway currently in use as transport’. This is surely one of the rarer ele-
ments in schemas around the world, and its existence, along with that of various others
that can be found, suggests that indefinitely many more of them can sporadically arise.
In addition to being only relatively closed at its hierarchically lower end, the inven-
tory may include some categories whose membership seems not to settle down to a small
fixed set. One such category may be that of ‘intrinsic parts’. Frequently encountered are
the five member elements ‘front’, ‘side’, ‘back’, ‘top’, and ‘bottom, as found in the English
prepositions in The cat lay before / beside / behind / atop / beneath the TV. But languages
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 329

like Mixtec seem to distinguish a rather different set of intrinsic parts in their spatial
schemas (Brugman and Macaulay,1986), while Makah distinguishes many more and
finer parts, such as with its verb suffixes for ‘at the ankle’ and ‘at the groin’ (Matthew
Davidson, personal communication).
Apart from any such fuzzy lower boundary or noncoalescing categories, though,
there does appear to exist a graduated inventory of basic spatial elements and categories
that is universally available and, in particular, is relatively closed. Bowerman (e.g. 1989)
has raised the main challenge to this notion. She notes, for example, that at the same time
that children acquiring English learn its in/on distinction, children acquiring Korean
learn its distinction between kkita ‘put [Figure] in a snug fit with [Ground]’ and nehta
‘put [Figure] in a loose fit with [Ground]’ she argues that since the elements ‘snug fit’
and ‘loose fit’ are presumably rare among spatial schemas across languages, they do
not come from any preset inventory, one that might plausibly be innate, but rather
are learned from the open-ended semantics of the adult language. My reply is that the
spatial schemas of genuinely closed-class forms in Korean may well still be built from
the proposed inventory elements, and that the forms she cites are actually open-class
verbs. Open-class semantics – whether for space or other domains – seems to involve
a different cognitive subsystem, drawing from finer discriminations within a broader
perceptual / conceptual sphere. The Korean verbs are perhaps learned at the same age
as English space-related open-class verbs like squeeze. Thus, English-acquiring children
probably understand that squeeze involves centripetal pressure from encircling or bi-/
multi-laterally placed Antagonists (typically the arm(s) or hand(s)) against an Agonist
that resists the pressure but yields down to some smaller compass where it blocks further
pressure, and hence that one can squeeze a teddy bear, a tube of toothpaste, or a rubber
ball, but not a piece of string or sheet of paper, juice or sugar or the air, a tabletop or the
corner of a building. Thus, Bowerman’s challenge may be directed at the wrong target,
leaving the proposed roughly preset inventory of basic spatial building blocks intact.

2.5 Basic elements assembled into whole schemas

The procedure so far has been analytic, starting with the whole spatial schemas expressed
by closed-class forms and abstracting from them an inventory of fundamental spatial
elements. But the investigation must also include a synthetic procedure: examining the
ways in which individual spatial elements are assembled to constitute whole schemas.
Something of such an assembly was implicit in the initial discussion of the across schema.
But an explicit example here can better illustrate this part of the investigation.
Consider the schema represented by the English preposition past as in The ball sailed
past my head at exactly 3 PM. This schema is built out of the following fundamental
spatial elements (from the indicated categories) in the indicated arrangements and
relationships: There are two main scene components (members of the ‘major scene
components’ category), a Figure and a Ground (here, the ball and my head, respec-
tively). The Figure is schematizable as a 0-dimensional point (a member element of the
‘dimension’ category). This Figure point is moving (a member element of the ‘motive
330 LANGUAGE, COGNITION AND SPACE

state’ category). Hence it forms a one-dimensional line (a member of the ‘dimension’

category’). This line constitutes the Figure’s ‘path’. The Ground is also schematizable as
a 0-dimensional point (a member of the ‘dimension’ category). There is a point P at
a proximal remove (a member of the ‘degree of remove’ category) from the Ground
point, forming a 1-dimensional line with it (a member of the ‘dimension’ category).
This line is parallel (a member of the ‘relative orientation’ category) to the horizontal
plane (a member of the ‘intrinsic parts’ category) of the earth-based grid (a member of
the ‘major scene components’ category). The Figure’s path is perpendicular (a member
of the ‘relative orientation’ category) to this line. The Figure’s path is also parallel to the
horizontal plane of the earth-based grid. If the Ground object has a front, side, and back
(members of the ‘intrinsic parts’ category), then point P is proximal to the side part. A
non-boundary point (a member of the ‘state of boundedness’ category) of the Figure’s
path becomes coincident (a member of the ‘degree of remove’ category) with point P
at a certain point of time.
Note that here the Figure’s path must be specified as passing through a point proxi-
mal to the Ground because if it instead passed through the Ground point, one would
switch from the preposition past to into, as in The ball sailed into my head, and if it
instead past through some distal point, one might rather say something like The ball
sailed along some ways away from my head. And the Figure’s path must be specified both
as horizontal and as located at the side portion of the Ground because, for example here,
if the ball were either falling vertically or traveling horizontally at my front, one would
no longer say that it sailed ‘past’ my head.
The least understood aspect of the present investigation is what well-formedness
conditions, if any, may govern the legality of such combinations. As yet, no obvious
principles based, say, on geometric simplicity, symmetry, consistency, or the like are
seen to control the patterns in which basic elements assemble into whole schemas.
On the one hand, some seemingly byzantine combinations –like the schemas seen
above for across and past – occur with some regularity across languages. On the other
hand, much simpler combinations seem never to occur as closed-class schemas. For
example, one could imagine assembling elements into the following schema: down
into a surround that is radially proximal to a center point. One could even invent a
preposition apit to represent this schema. This could then be used, say, in I poured
water apit my house to refer to my pouring water down into a nearby hole dug in
the field around my house. But such schemas are not found. Similarly, a number of
schematic distinctions in, for example, the domain of rotation are regularly marked
by signed languages, as seen below, and could readily be represented with the inven-
tory elements available to spoken languages, yet they largely do not occur. It could
be argued that the spoken language schemas are simply the spatial structures most
often encountered in everyday activity. But that would not explain why the additional
sign-language schemas – presumably also reflective of everyday experience – do not
show up in spoken languages. Besides, the different sets of spatial schemas found in
different spoken languages are diverse enough from each other that arguing on the
basis of the determinative force of everyday experience is problematic. Something
else is at work but it is not yet clear what that is.
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 331

2.6 Properties and processes applying to whole spatial schemas

It was just seen that selected elements of the inventory are combined in specific arrange-
ments to make up the whole schemas represented by closed-class spatial forms. Each such
whole schema is thus a ‘pre-packaged’ bundling together of certain elements in a particular
arrangement. Each language has in its lexicon a relatively closed set of such pre-packaged
schemas – a set larger than that of its spatial closed-class forms, because of polysemy. A
speaker of the language must select among these schemas in depicting a spatial scene. We
now observe that such schemas, though composite, have a certain unitary status in their
own right, and that certain quite general properties and processes can apply to them. In
particular, certain properties and processes allow a schema represented by a closed class
form to generalize to a whole family of schemas. In the case of a generalizing property,
all the schemas of a family are of equal priority. On the other hand, a generalizing process
acts on a schema that is somehow basic, and either extends or deforms it to yield nonbasic
schemas. (see Talmy 2000a ch. 1 and 3, 2000b ch. 5). Such properties and processes are
perhaps part of the overall spoken-language system so that any language’s relatively closed
set of spatial closed-class forms and the schemas that they basically represent can be used
to match more spatial structures in a wider range of scenes.
Looking first at generalizing properties of spatial schemas, one such property is
that they exhibit a topological or topology-like neutrality to certain factors of Euclidean
geometry. Thus, they are magnitude neutral, as seen in such facts as that the across
schema can apply to a situation of any size, as in The ant crawled across my palm / The
bus drove across the country. Further, they are largely shape-neutral, as seen by such
facts as that, while the through schema requires that the Figure form a path with linear
extent, it lets that line take any contour, as in I zig-zagged / circled through the woods.
And they are bulk-neutral, as seen by such facts as that the along schema requires a
linear Ground without constraint on the Ground’s radial extension, as in The caterpillar
crawled up along the filament /tree trunk. Thus, while holding to their specific constraints,
schemas can vary freely in other respects and so cover a range of spatial configurations.
Among the generalizing processes that extend schemas, one is that of ‘extendability
from the prototype’, which can actually serve as an alternative interpretation for some
forms of neutrality, otherwise just treated under generalizing properties. Thus, in the
case of shape, as for the through schema above, this schema could alternatively be
conceived as prototypically involving a straight path line for the Figure, one that can
then be bent to any contour. And, in the case of bulk, as for the along schema above,
this schema could be thought prototypically to involve a purely 1-dimensional line that
then can be radially inflated.
Another such process is ‘extendability in ungoverned dimensions’. By this process,
a scene component of dimensionality N in the basic form of a schema can generally be
raised in dimensionality to form a line, plane, or volume aligned in a way not conflict-
ing with the schema’s other requirements. To illustrate, it was seen earlier under the
‘type of geometry’ category that the English verb satellite out has a schema involving
a point Figure moving along a radius away from a center point through a continuum
of concentric circles, as in The boat sailed further and further out from the island. This
332 LANGUAGE, COGNITION AND SPACE

schema with the Figure idealizable as a point is the basic form. But the same satellite can
be used when this Figure point is extended to form a 1-dimensional line along a radius,
as in The caravan of boats sailed further and further out from the island. And the out can
again be used if the Figure point were instead extended as a 1-dimensional line forming
a concentric circle, as in A circular ripple spread out from where the pebble fell into the
water. In turn, such a concentric circle could be extended to fill in the interior plane, as
in The oil spread out over the water from where it spilled. Alternatively, the concentric
circle could have been extended in the vertical dimension to form a cylinder, as in A
ring of fire spread out as an advancing wall of flames. Or again, the circle could have
been extended to form a spherical shell, as in The balloon I blew into slowly puffed out.
And such a shell can be extended to fill in the interior volume, as in The leavened dough
slowly puffed out. Thus, the same form out serves for this series of geometric extensions
without any need to switch to some different form.
One more schema-extending process is ‘extendability across motive states’. A
schema basic for one motive state and Figure geometry can in general be systematically
extended to another motive state and Figure geometry. For example, a closed-class
form whose most basic schema pertains to a point Figure moving to form a path can
generally serve as well to represent the related schema with a stationary linear Figure in
the same location as the path. Thus, probably the most basic across schema is actually
for a moving point Figure, as in The gopher ran across the road. By the present process,
this schema can extend to the static linear Figure schema first seen in The board lay
across the road. All the spatial properties uncovered for that static schema hold as
well for the present basic dynamic schema, which in fact is the schema in which these
properties originally arise.
Among the generalizing processes that deform a schema, one is that of ‘stretching’,
which allows a slight relaxing of one of the normal constraints. Thus, in the across
schema, where the Ground plane is either a ribbon with a long and short axis or a
square with equal axes, a static linear Figure or the path of a moving point Figure must
be aligned with the short Ground axis or with one of its equal axes. Accordingly, one
can say I swam across the canal and I swam across the square pool when moving from
one side to the other, but one cannot say *I swam across the canal when moving from
one end of the canal to the other. But, by moderately stretching one axis length relative
to the other, one might just about be able to say I swam across the pool when moving
from one end to the other of a slightly oblong pool.
Another schema deforming process is that of ‘feature cancellation’, in which a particu-
lar complex of elements in the basic schema is omitted. Thus, the preposition across can
be used in The shopping cart rolled across the boulevard and was hit by an oncoming car,
even though one feature of the schema – ‘terminal point coincides with the distal edge
of the Ground ribbon’ – is canceled from the Figure’s path. Further, both this feature and
the feature ‘beginning point coincides with the proximal edge of the Ground ribbon’ are
canceled in The tumbleweed rolled across the prairie for an hour. Thus, the spoken language
system includes a number of generalizing properties and processes that allow the otherwise
relatively closed set of abstracted or basic schemas represented in the lexicon of any single
language to be applicable to a much wider range of spatial configurations.
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 333

3 Spatial structuring in signed language

All the preceding findings on the linguistic structuring of space have been based on
the patterns found in spoken languages. The inquiry into the fundamental concept
structuring system of language leads naturally to investigating its character in another
major body of linguistic realization, signed language. The value in extending the inquiry
in this way would be to discover whether the spatial structuring system is the same or
is different in certain respects across the two language modalities, with either discovery
having major consequences for cognitive theory.
In this research extension, a problematic issue is exactly what to compare between
spoken and signed language. The two language systems appear to subdivide into some-
what different sets of subsystems. Thus, heuristically, the generalized spoken language
system can be thought to consist of an open-class or lexical subsystem (generally
representing conceptual content); a closed-class or grammatical subsystem (generally
representing conceptual structure); a gradient subsystem of ‘vocal dynamics’ (includ-
ing loudness, pitch, timbre, rate, distinctness, unit separation); and an accompanying
somatic subsystem (including facial expression, gesture, and ‘body language’). On the
other hand, by one provisional proposal, the generalized sign language system might
instead divide up into the following: a subsystem of lexical forms (including noun, verb,
and adjective signs); an ‘inflectional’ subsystem (including modulations of lexical signs
for person, aspect); a subsystem of size-and-shape specifiers (or SASS’s; a subsystem of
so-called ‘classifier expressions’; a gestural subsystem (along a gradient of incorporation
into the preceding subsystems); a subsystem of face, head, and torso representations;
a gradient subsystem of ‘bodily dynamics’ (including amplitude, rate, distinctness,
unit separation); and an associated or overlaid somatic subsystem (including further
facial expression and ‘body language’). In particular here, the subsystem of classifier
expressions – which is apparently present in all signed languages – is a formally distinct
subsystem dedicated solely to the schematic structural representation of objects moving
or located with respect to each other in space (see Liddell 2003, Emmorey 2002). Each
classifier expression, perhaps generally corresponding to a clause in spoken language,
represents a so conceived event of motion or location. 4
The research program of comparing the representation of spatial structure across
the two language modalities ultimately requires considering the two whole systems
and all their subsystems. But the initial comparison – the one adopted here – should be
between those portions of each system most directly involved with the representation
of spatial structure. In spoken language, this is that part of the closed-class subsystem
that represents spatial structure and, in signed language, it is the subsystem of classifier
constructions. Spelled out, the shared properties that make this initial comparison apt
include the following. First, of course, both subsystems represent objects relating to each
other in space. Second, in terms of the functional distinction between ‘structure’ and
‘content’ described earlier, each of the subsystems is squarely on the structural side. In
fact, analogous structure-content contrasts occur. Thus, the English closed-class form
into represents the concept of a path that begins outside and ends inside an enclosure
in terms of schematic structure, in contrast with the open-class verb enter that repre-
334 LANGUAGE, COGNITION AND SPACE

sents the same concept in terms of substantive content (see Talmy 2000a, ch. 1 for this
structure-content distinction). Comparably, any of the formations within a classifier
expression for such an outside-to-inside path represents it in terms of its schematic
structure, in contrast with the unrelated lexical verb sign that can be glossed as ‘enter’.
Third, in each subsystem, a schematic structural form within an expression in general
can be semantically elaborated by a content form that joins or replaces it within the same
expression. Thus, in the English sentence I drove it (– the motorcycle–) in (to the shed)
the parenthesized forms optionally elaborate on the otherwise schematically represented
Figure and Ground. Comparably, in the ASL sentence ‘(SHED) (MOTORCYCLE)
vehicle-move-into-enclosure’, the optionally signed forms within parentheses elaborate
on the otherwise schematic Figure and Ground representations within the hyphenated
classifier expression.
To illustrate the classifier system, a spatial event that English could express as The
car drove past the tree could be expressed in ASL as follows: The signer’s dominant
hand, used to represent the Figure object, here has a ‘3 handshape’ (index and middle
fingers extended forward, thumb up) to represent a land vehicle. The nondominant
hand, used to represent the Ground object, here involves an upright ‘5 handshape’
(forearm held upright with the five fingers extended upward and spread apart) to
represent a tree. The dominant hand is moved horizontally across the signer’s torso
and past the nondominant forearm. Further though, this basic form could be modified
or augmented to represent additional particulars of the referent spatial event. Thus,
the dominant hand can show additional characteristics of the path. For example, the
hand could move along a curved path to indicate that the road being followed was
curved, it could slant upward to represent an uphill course, or both could be shown
together. The dominant hand can additionally show the manner of the motion. For
example, as it moves along, it could oscillate up and down to indicate a bumpy ride,
or move quickly to indicate a swift pace, or both could be shown together, as well as
with the preceding two path properties. And the dominant hand can show additional
relationships of the Figure to the Ground. For example, it could pass nearer or farther
from the nondominant hand to indicate the car’s distance from the tree when passing
it, it could make the approach toward the nondominant hand longer (or shorter) than
the trailing portion of the path to represent the comparable relationship between the
car’s path and the tree, or it could show both of these together or, indeed, with all the
preceding additional characteristics.
The essential finding of how signed language differs from spoken language is that it
more closely parallels what appear to be the structural characteristics of scene parsing
in visual perception. This difference can be observed in two venues, the universally
available spatial inventory and the spatial expression.
These two venues are discussed next in turn.
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 335

3.1 In the inventory

The inventory of forms for representing spatial structure available to the classifier
subsystem of signed language has a greater total number of fundamental elements, a
greater number of categories, and generally a greater number of elements per category
than the spoken language closed-class inventory. While many of the categories and their
members seem to correspond across the two inventories, the signed language inven-
tory has an additional number of categories and member elements not present in the
spoken language inventory. Comparing the membership of the corresponding categories
in terms of discrete elements, the number of basic elements per category in signed
language actually exhibits a range: from being the same as that for spoken language to
being very much greater. Further, though, while the membership of some categories
in signed language may well consist of discrete elements, that of others appears to be
gradient. Here, any procedure of tallying some fixed number of discrete elements in a
category must give way to determining the approximate fineness of distinctions that can
be practicably made for that category. So while some corresponding categories across
the two language modalities may otherwise be quite comparable, their memberships
can be of different types, discrete vs. analog. Altogether, then, given its greater number
of categories, generally larger membership per category, and a frequently gradient type
of membership, the inventory of forms for building a schematic spatial representation
available to the classifier subsystem of signed language is more extensive and finer than
for the closed-class subsystem of spoken language. This greater extensiveness and finer
granularity of spatial distinctions seems more comparable to that of spatial parsing in
visual perception.
The following are some spatial categories in common across the two language
modalities, but with increasing disparity in size of membership. First, some categories
appear to be quite comparable across the two modalities. Thus, both the closed-class sub-
system of spoken language and the classifier subsystem of signed language structurally
segment a scene into the same three components, a Figure, a Ground, and a secondary
Reference Object. Both subsystems represent the category of dimensionality with the
same four members – a point, a line, a plane, and a volume. And both mark the same
two degrees of boundedness: bounded and unbounded.
For certain categories, signed language has just a slightly greater membership than
does spoken language. Thus, for motive state, signed language structurally represents
not only moving and being located, but also remaining fixedly located – a concept that
spoken languages typically represent in verbs but not in their spatial preposition-like
forms.
For some other spatial categories, signed language has a moderately greater mem-
bership than spoken language. In some of these categories, the membership is probably
gradient, but without the capacity to represent many fine distinctions clearly. Thus,
signed language can apparently mark moderately more degrees of remove than spoken
language’s four or five members in this category. It can also apparently distinguish mod-
erately more path lengths than the two – short and long – that spoken language marks
336 LANGUAGE, COGNITION AND SPACE

structurally (as in English The bug flew right / way up there). And while spoken language
can mark at most three distinctions of relative orientation – parallel, perpendicular, and
oblique – signed language can distinguish a moderately greater number, for example, in
the elevation of a path’s angle above the horizontal, or in the angle of the Figure’s axes
to that of the Ground (e.g. in the placement of a rod against a wall).
Finally, there are some categories for which signed language has an indefinitely
greater membership than spoken language. Thus, while spoken language structurally
distinguishes some four path contours as seen in section 2.3.3, signed language can
represent perhaps indefinitely many more, including zigzags, spirals, and ricochets.
And for the category ‘locus within referent space’, spoken language can structurally
distinguish perhaps at most three loci relative to the speaker’s location – ‘here’, ‘there’,
and ‘yonder’ – whereas sign language can distinguish indefinitely many more within
sign space.
Apart from membership differences across common categories, signed language
represents some categories not found in spoken language. One such category is the
relative lengths of a Figure’s path before and after encounter with the Ground. Or again,
signed language can represent not only the category of ‘degree of dispersion’ (which
spoken language was seen to represent in section 2.3.3), but also the category ‘pattern
of distribution’. Thus, in representing multiple Figure objects dispersed over a planar
surface, it could in addition structurally indicate that these Figure objects are linear (as
with dry spaghetti over a table) and are arrayed in parallel alignment, crisscrossing, or
in a jumble.
This difference in the number of structurally marked spatial category and element
distinctions between spoken and signed language can be highlighted with a closer
analysis of a single spatial domain, that of rotational motion. As seen earlier, the closed-
class subsystem in spoken language basically represents only one category within this
domain, that of ‘orientation of spin axis’, and within this category distinguishes only two
member elements, vertical and horizontal. These two member elements are expressed,
for example, by the English verb satellites around and over as in The pole spun around
/ toppled over. ASL, by contrast, distinguishes more degrees of spin axis orientation
and, in addition, marks several further categories within the domain of rotation. Thus,
it represents the category of ‘amount of rotation’ and within this category can readily
distinguish, say, whether the arc of a Figure’s path is less than, exactly, more than, or
many times one full circuit. These are differences that English might offer for inference
only from the time signature, as in I ran around the house for 20 seconds / in 1 minute
/ for 2 minutes / for hours, while using the same single spatial form around for all these
cases. Further, while English would continue using just around and over, ASL further
represents the category of ‘relation of the spin axis to an object’s geometry’ and marks
many distinctions within this category. Thus, it can structurally mark the spin axis as
being located at the center of the turning object – as well as whether this object is planar
like a CD disk, linear like a propeller, or an aligned cylinder like a pencil spinning on its
point. It distinguishes this from the spin axis located at the boundary of the object – as
well as whether the object is linear like the ‘hammer’ swung around in a hammer toss, a
transverse plane like a swinging gate, or a parallel plane like a swung cape. And it further
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 337

distinguishes these from the spin axis located at a point external to the object – as well
as whether the object is point-like like the earth around the sun, or linear like a spinning
hoop. Finally, ASL can structurally represent the category of ‘uniformity of rotation’ with
its two member elements, uniform and nonuniform, where English could mark this
distinction only with an open-class form, like the verbs in The hanging rope spun / twisted
around, while once again continuing with the same single structural closed-class form
around. Thus, while spoken language structurally marks only a minimal distinction of
spin axis orientation throughout all these geometrically distinct forms of rotation, signed
language marks more categories as well as finer distinctions within them, and a number
of these appear to be distinguished as well by visual parsing of rotational movement.
To expand on the issue of gradience, numerous spatial categories in the classifier
subsystem of signed language – for example, many of the 30 spatial categories listed in
section 3.2.3.1 are gradient in character. Spoken language has a bit of this, as where the
vowel length of a waaay in English can be varied continuously. But the preponderant
norm is the use of discrete spatial elements, typically incorporated into distinct mor-
phemes. For example, insofar as they represent degree of remove, the separate forms
in the series on / next to / near / away from represent increasing distance in what can
be considered quantal jumps. That is, the closed-class subsystem of spoken language is
a type of cognitive system whose basic organizing principle is that of the recombina-
tion of discrete elements (i.e., the basic conceptual elements whose combinations, in
turn, comprise the meanings of discrete morphemic forms). By contrast, the classifier
subsystem of signed language is the kind of cognitive system whose basic organizing
principle largely involves gradience, much as would seem to be the case as well for the
visual and motor systems. In fact, within a classifier expression, the gradience of motor
control and of visual perception are placed in sync with each other (for the signer and
the addressee, respectively), and conjointly put in the service of the linguistic system.
While this section provides evidence that the classifier subsystem in signed language
diverges from the schematizing of spoken language in the direction of visual parsing,
one must further observe that the classifier subsystem is also not ‘simply’ a gestural
system wholly iconic with visual perception. Rather, it incorporates much of the discrete,
categorial, symbolic, and metaphoric character that is otherwise familiar from the
organization of spoken language. Thus, as already seen above, spatial representation in
the classifier subsystem does fall into categories, and some of these categories contain
only a few discrete members – in fact, several of these are much the same as in spoken
language. Second, the hand-shapes functioning as classifiers for the Figure, manipula-
tor, or instrument within classifier expressions are themselves discrete (nongradient)
members of a relatively closed set. Third, many of the hand movements in classifier
expressions represent particular concepts or meta-concepts and do not mimic actual
visible movements of the represented objects. Here is a small sample of this property.
After one lowers one’s two extended fingers to represent a knife dipping into peanut
butter – or all one’s extended fingers in a curve to represent a scoop dipping into coffee
beans – one curls back the fingertips while moving back up to represent the instrument’s
‘holding’ the Figure, even though the instrument in question physically does nothing
of the sort. Or again, the free fall of a Figure is represented not only by a downward
338 LANGUAGE, COGNITION AND SPACE

motion of the dominant hand in its classifier handshape, but also by an accompanying
rotation of the hand – whether or not the Figure in fact rotated in just that way during
its fall. As another example, a Figure is shown as simply located at a spot in space by
the dominant hand in its classifier handshape being placed relaxedly at a spot in signing
space, and as remaining fixedly at its spot by the hand’s being placed tensely and with a
slight final jiggle, even though these two conceptualizations of the temporal character of
a Figure’s location are visually indistinguishable. Or, further, a (so-conceivedly) random
spatial distribution of a mass or multiplex Figure along a line, over a plane, or through
a volume is represented by the Figure hand being placed with a loose nonconcerted
motion, typically three times, at uneven spacings within the relevant n-dimensional
area, even though that particular spacing of three exemplars may not correspond to the
actual visible distribution. And finally, a classifier hand’s type of movement can indicate
whether this movement represents the actual path of the Figure, or is to be discounted.
Thus, the two flat hands held with palms toward the signer, fingertips joined, can be
moved steadily away to represent a wall’s being slid progressively outward (as to expand
a room), or instead can be moved in a quick up-and-down arc to a point further away to
represent a wall relocated to a further spot, whatever its path from the starting location.
That is, the latter quick arc movement represents a meta-concept: that the path followed
by the hands does not represent the Figure’s actual path and is to be disregarded from
calculations of iconicity. All in all, then, the classifier subsystem presents itself as a
genuine linguistic system, but one having more extensive homology with the visual
structuring system than spoken language has.

3.2 In the expression

The second venue, that of any single spatial expression, exhibits further respects in which
signed language differs from spoken language in the apparent direction of visual scene
parsing. Several of these are outlined next.

3.2.1 Iconic representation in the expression

Spatial representation in signed classifier expressions is iconic with scene parsing in

visual perception in at least the following four respects.

3.2.1.1 Iconic clustering of elements and categories

The structural elements of a scene of motion are clustered together in the classifier
subsystem’s representation of them in signed language more as they seem to be
clustered in perception. When one views a motion event, such as a car driving bumpily
along a curve past a tree, it is perceptually the same single object, the car, that exhibits
all of the following characteristics: it has certain object properties as a Figure, it
moves, it has a manner of motion, it describes a path of a particular contour, and it
relates to other surrounding objects (the Ground) in its path of motion. The Ground
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 339

object or objects are perceived as separate. Correspondingly, the classifier subsystem

maintains exactly this pattern of clustering. It is the same single hand, the dominant
hand, that exhibits the Figure characteristics, motion, manner, path contour, and
relations to a Ground object. The other hand, the nondominant, separately represents
the Ground object.
All spoken languages diverge to a greater or lesser extent from this visual fidelity.
Thus, consider one English counterpart of the event, the sentence The car bumped along
past the tree. Here, the subject nominal, the car, separately represents the Figure object
by itself. The verb complex clusters together the representations of the verb and the
satellite: The verb bumped represents both the fact of motion and the manner of motion
together, while its sister constituent, the satellite along represents the presence of a path
of translational motion. The prepositional phrase clusters together the preposition
past, representing the path conformation, and its sister constituent, the nominal the
tree, representing the Ground object. It in fact remains a mystery at this point in the
investigation why all spoken languages using a preposition-like constituent to indicate
path always conjoin it with the Ground nominal and basically never with the Figure
nominal5, even though the Figure is what executes the path, and is so represented in
the classifier construction of signed language.

3.2.1.2 Iconic representation of object vs. action

The classifier subsystem of signed language appears to be iconic with visual parsing not
only in its clustering of spatial elements and categories, as just seen, but largely also in
its representation of them. For example, it marks one basic category opposition, that
between an entity and its activity, by using an object like the hand to represent an object,
and motion of the hand to represent motion of the object. More specifically, the hand
or other body part represents a structural entity (such as the Figure) – with the body
part’s configuration representing the identity or other properties of the entity – while
movements or positionings of the body part represent properties of the entity’s motion,
location, or orientation. For example, the hand could be shaped flat to represent a
planar object (e.g. a sheet of paper), or rounded to represent a cup-shaped object. And,
as seen, any such hand-shape as Figure could be moved along a variety of trajectories
that represent particular path contours.
But an alternative to this arrangement could be imagined. The handshape could
represent the path of a Figure – e.g., a fist to represent a stationary location, the out-
stretched fingers held flat together to represent a straight line path, the fingers in a
curved plane for a curved path, and the fingers alternately forward and backward for a
zigzag path. Meanwhile, the hand movement could represent the Figure’s shape – e.g.,
the hand moving in a circle to represent a round Figure and in a straight line for a
linear Figure. However, no such mapping of referents to their representations is found.6
Rather, the mapping in signed language is visually iconic: it assigns the representation
of a material object in a scene to a material object in a classifier complex, for example,
the hand, and the representation of the movements of that object in the scene to the
movements of the hand.
340 LANGUAGE, COGNITION AND SPACE

No such iconic correspondence is found in spoken language. Thus, while material

objects are prototypically expressed by nouns in English, they are instead prototypically
represented by verb roots in Atsugewi (see Talmy 2000b, ch. 1). And while path configu-
rations are prototypically represented in Spanish by verbs, this is done by prepositions
and satellites in English.

3.2.1.3 Iconic representation of further particular categories

Finer forms of iconicity are also found within each branch of the broad entity-activity
opposition. In fact, most of the spatial categories listed in section 3.2.3.1 that a classifier
expression can represent are largely iconic with visual parsing. Thus, an entity’s form
is often represented by the form of the hand(s), its size by the compass of the hand(s),
and its number by the number of digits or hands extended. And, among many other
categories in the list, an entity’s motive state, path contour, path length, manner of
motion, and rate of motion are separately represented by corresponding behaviors of
the hand(s).
Spoken language, again, has only a bit of comparable iconicity. As examples, path
length can be iconically represented in English by the vowel length of way, as in The
bird flew waay / waaaay / waaaaaay up there. Path length can also be semi-iconically
represented by the number of iterations, as in The bird flew up/ up up / up up up and
away. Perhaps the number of an entity can be represented in some spoken language by a
closed-class reduplication. But the great majority of spoken closed-class representations
show no such iconicity.

3.2.1.4 Iconic representation of the temporal progression of a trajectory

The classifier subsystem is also iconic with visual parsing in its representation of tempo-
ral progression, specifically, that of a Figure’s path trajectory. For example, when an ASL
classifier expression represents ‘The car drove past the tree’, the ‘past’ path is shown by
the Figure hand progressing from the nearer side of the Ground arm to a point beside
it and then on to its further side, much like the path progression one would see on
viewing an actual car passing a tree. By contrast, nothing in any single closed-class path
morpheme in a spoken language corresponds to such a progression. Thus, the past in
The car drove past the tree is structurally a single indivisible linguistic unit, a morpheme,
whose form represents no motion ahead in space. Iconicity of this sort can appear in
spoken language only where a complex path is treated as a sequence of subparts, each
with its own morphemic representation, as in I reached my hand down around behind
the clothes hamper to get the vacuum cleaner.

3.2.2 A narrow time-space aperture in the expression

Another way that the classifier expression in signed language may be more like visual
perception is that it appears to be largely limited to representing a narrow time-space
aperture. The tentative principle is that a classifier complex readily represents what would
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 341

appear within a narrow scope of space and time if one were to zoom in with one’s scope
of perception around a Figure object, but little outside that narrowed scope. Hence,
a classifier expression readily represents the Figure object as to its shape or type, any
manipulator or instrument immediately adjacent to the Figure, the Figure’s current state
of Motion (motion or located-ness), the contour or direction of a moving Figure’s path,
and any Manner exhibited by the Figure as it moves. However, a classifier expression
can little represent related factors occurring outside the current time, such as a prior
cause or a follow-up consequence. And it can little represent even concurrent factors
if they lie outside the immediate spatial ambit of the Figure, factors like the ongoing
causal activity of an intentional Agent or other external instrumentality.
By contrast, spoken languages can largely represent such nonlocal spatiotempo-
ral factors within a single clause. In particular, such representation occurs readily in
satellite-framed languages such as English (see Talmy 2000b, ch. 1 and 3). In representing
a Motion event, this type of language regularly employs the satellite constituent (e.g.
the verb particle in English) to represent the Path, and the main verb to represent a
‘co-event’. The co-event is ancillary to the main Motion event and relates to it as its
precursor, enabler, cause, manner, concomitant, consequence, or the like.
Satellite-framed languages can certainly use this format to represent within-aperture
situations that can also be represented by a classifier complex. Thus, English can say
within a single clause – and ASL can sign within a single classifier expression – a motion
event in which the Figure is moved by an adjacent manipulator, as in I pinched some
moss up off the rock and I pulled the pitcher along the counter, or in which the Figure is
moved by an adjacent instrument, as in I scooped jelly beans up into the bag. The same
holds for a situation in which a moving Figure exhibits a concurrent Manner, as in The
cork bobbed past the seaweed.
But English can go on to use this same one-clause format to include the representa-
tion of co-events outside the aperture, either temporally or spatially. Thus, temporally,
English can include the representation of a prior causal event, as in I kicked the football
over the goalpost (first I kicked the ball, then it moved over the goalpost). And it can
represent a subsequent event, as in They locked the prisoner into his cell (first they put
him in, then they locked it). But ASL cannot represent such temporally extended event
complexes within a single classifier expression. Thus, it can represent the former sentence
with a succession of two classifier expressions: first, flicking the middle finger of the
dominant hand across the other hand’s upturned palm to represent the component
event of kicking an object, and next moving the extended index finger of the dominant
hand axially along a line through the space formed by the up-pointing index and little
fingers of the nondominant hand, representing the component event of the ball’s pass-
ing over the goalpost. But it cannot represent the whole event complex within a single
expression – say, by flicking one’s middle finger against the other hand whose extended
index finger then moves off axially along a line.
Further, English can use the same single-clause format to represent events with
spatial scope beyond a narrow aperture, for example, an Agent’s concurrent causal
activity outside any direct manipulation of the Figure, as in I walked / ran / drove/flew
the memo to the home office. Again, ASL cannot represent the whole event complex of,
342 LANGUAGE, COGNITION AND SPACE

say, I ran the memo to the home office within a single classifier expression. Thus, it could
not, say, adopt the classifier for holding a thin flat object (thumb pressed against flat
fingers) with the dominant hand and placing this atop the nondominant hand while
moving forward with it as it shows alternating strokes of two downward pointed fingers
to indicate running (or concurrently with any other indication of running). Instead a
sequence of two expressions would likely be used, for example, first one for taking a
memo, then one for a person speeding along.7
Although the unacceptable examples above have been devised, they nevertheless
show that it is physically feasible for a signed language to represent factors related to
the Figure’s Motion outside its immediate space-time ambit. Accordingly, the fact that
signed languages, unlike spoken languages, do avoid such representations may follow
from deeper structural causes, such as a greater fidelity to the characteristics of visual
perception.
However apt, though, such an account leaves some facts still needing explanation.
Thus, on the one hand, it makes sense that the aperture of a classifier expression is
limited temporally to the present moment – this accords with our usual understand-
ing of visual perception. But it is not clear why the aperture is also limited spatially.
Visual perception is limited spatially to a narrow scope only when attention is being
focused, but is otherwise able to process a wide scoped array. Why then should classifier
expressions avoid such wide spatial scope as well? Further, sign languages can include
representation of the Ground object within a single classifier expression (typically with
the nondominant hand), even where that object is not adjacent to the Figure.

3.2.3 More independent distinctions representable in the expression

This third property of classifier expressions has two related aspects – the large number of
different elements and categories that can be represented together, and their independent
variability – and these are treated in succession next.

3.2.3.1 Many more elements / categories representable within a single expression

Although the spatiotemporal aperture that can be represented within a single classifier
expression may be small compared to that in a spoken-language clause, the number of
distinct factors within that aperture that can be represented is enormously greater. In
fact, perhaps the most striking difference between the signed and the spoken representa-
tion of space in the expression is that the classifier system in signed language permits the
representation of a vastly greater number of distinct spatial categories simultaneously
and independently. A spoken language like English can separately represent only up
to four or five different spatial categories with closed-class forms in a single clause. As
illustrated in the sentence The bat flew way back up into its niche in the cavern, the verb
is followed in turn by: a slot for indication of path length (with three members: ‘zero’
for ‘neutral’, way for ‘relatively long’, right for ‘relatively short’); a slot for state of return
(with two members: ‘zero’ for ‘neutral’, back for ‘return’); a slot for displacement within
the earth-frame (with four members: ‘zero’ for ‘neutral’, up for ‘positive vertical displace-
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 343

ment’, down for ‘negative vertical displacement’, over for ‘horizontal displacement’); a
slot for geometric conformation (with many members, including in, across, past); and
perhaps a slot for motive state and vector (with two members: ‘zero’ for ‘neutral between
location AT and motion TO’ as seen in in / on, and -to for ‘motion TO’ as seen in into /
onto). Even a polysynthetic language like Atsugewi has closed-class slots within a single
clause for only up to six spatial categories: path conformation combined with Ground
type, path length, vector, deixis, state of return, and cause or manner. In contrast, by
one tentative count, ASL has provision for the separate indication of thirty different
spatial categories. These categories do exhibit certain cooccurrence restrictions, they
differ in obligatoriness or optionality, and it is unlikely – perhaps impossible – for all
thirty of them to be represented at once. Nevertheless, a sizable number of them can be
represented in a single classifier expression and varied independently there. The table
below lists the spatial categories that I have provisionally identified as available for
concurrent independent representation. The guiding principle for positing a category
has been that its elements are mutually exclusive: different elements in the same category
cannot be represented together in the same classifier expression. If certain elements
can be concurrently represented, they belong to different categories. Following this
principle has, on the one hand, involved joining together what some sign language
analyses have treated as separate factors. For example, the first category below covers
equally the representation of Figure, instrument, or manipulator (handling classifier),
since these three kinds of elements apparently cannot be separately represented in a
single expression – one or another of them must be selected. On the other hand, the
principle requires making distinctions within some categories that spoken languages
treat as uniform. Thus, the single ‘manner’ category of English must be subdivided
into a category of ‘divertive manner’ (e.g. moving along with an up-down bump) and
a category of ‘dynamic manner’ (e.g. moving along rapidly) because these two factors
can be represented concurrently and varied independently.
A. Entity properties
1. identity (form or semantic category) of Figure / instrument / manipulator
2. identity (form or semantic category) of Ground
3. magnitude of some major entity dimension
4. magnitude of a transverse dimension
5. number of entities
B. Orientation properties
1. an entity’s rotatedness about its left-right axis (‘pitch’)
2. an entity’s rotatedness about its front-back axis (‘roll’)
3.a. an entity’s rotatedness about its top-bottom axis (‘yaw’)
3.b. an entity’s rotatedness relative to its path of forward motion
C. Locus properties
1. Locus within sign space
344 LANGUAGE, COGNITION AND SPACE

D. Motion properties
1. motive state (moving / resting / fixed)
2. internal motion (e.g. expansion/contraction, form change, wriggle, swirling)
3. confined motion (e.g. straight oscillation, rotary oscillation, rotation, local
wander)
4. translational motion
E. Path properties
1. state of continuity (unbroken / saltatory)
2. contour of path
3. state of boundedness (bounded / unbounded)
4. length of path
5. vertical height
6. horizontal distance from signer
7. left-right positioning
8. up-down angle (‘elevation’)
9. left-right angle (‘direction’)
10. transitions between motion and stationariness (e.g. normal, decelerated,
abrupt as from impact)
F. Manner properties
1. divertive manner
2. dynamic manner
G. Relations of Figure or Path to Ground
1. path’s conformation relative to Ground
2. relative lengths of path before and after encounter with Ground
3. Figure’s path relative to the Path of a moving Ground
4. Figure’s proximity to Ground
5. Figure’s orientation relative to Ground

It seems probable that something more on the order of this number of spatial categories
are concurrently analyzed out by visual processing on viewing a scene than the much
smaller number present in even the most extreme spoken language patterns.

3.2.3.2 Elements / categories independently variable in the expression – not in pre-

packaged schemas

The signed-spoken language difference just presented was mainly considered for the
sheer number of distinct spatial categories that can be represented together in a single
classifier expression. Now, though, we stress the corollary: their independent variability.
That is, apart from certain constraints involving cooccurrence and obligatoriness in a
classifier expression, a signer can generally select a category for inclusion independently
of other categories, and select a member element within each category independently
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 345

of other selections. For example, a classifier expression can separately include and
independently vary a path’s contour, length, vertical angle, horizontal angle, speed,
accompanying manner, and relation to Ground object.
By contrast, it was seen earlier that spoken languages largely bundle together a
choice of spatial member elements within a selection of spatial categories for representa-
tion within the single complex schema that is associated with a closed-class morpheme.
The lexicon of each spoken language will have available a certain number of such ‘pre-
packaged’ spatial schemas, and the speaker must generally choose from among those
to represent a spatial scene, even where the fit is not exact. The system of generalizing
properties and processes seen in section 2.6 that apply to the set of basic schemas in the
lexicon (including their plastic extension and deformation) may exist to compensate
for the pre-packaging and closed stock of the schemas in any spoken language. Thus,
what are largely semantic components within a single morpheme in spoken language
correspond to what can be considered separate individually controllable morphemes
in the signed classifier expression.
The apparent general lack in classifier expressions of pre-packaging, of a fixed set
of discrete basic schemas, or of a system for generalizing, extending, or deforming such
basic schemas may well accord with comparable characteristics of visual parsing. That
is, the visual processing of a viewed scene may tend toward the independent assessment
of spatial factors without much pre-packeting of associated factors or of their plastic
alteration. If shown to be the case, then signed language will once again prove to be
closer to perceptual spatial structuring than spoken language is.

4 Cognitive implications of spoken / signed language

differences

The preceding comparison of the space-structuring subsystems of spoken and of signed

language has shown a number of respects in which these are similar and in which they
are different. It can be theorized that their common characteristics are the product of a
single neural system, what can be assumed to be the core language system, while each
set of distinct characteristics results from the activity of some further distinct neural
system. These ideas are outlined next.

4.1 Where signed and spoken language are alike

We can first summarize and partly extend the properties above found to hold both in
the closed-class subsystem of spoken language and in the classifier subsystem of signed
language. Both subsystems can represent multifarious and subtly distinct spatial situa-
tions – that is, situations of objects moving or located with respect to each other in space.
Both represent such spatial situations schematically and structurally. Both have basic
elements that in combination make up the structural schematizations. Both group their
basic elements within certain categories that themselves represent particular categories
346 LANGUAGE, COGNITION AND SPACE

of spatial structure. Both have certain conditions on the combination of basic elements
and categories into a full structural schematization. Both have conditions on the cooc-
currence and sequencing of such schematizations within a larger spatial expression.
Both permit semantic amplification of certain elements or parts of a schematization
by open-class or lexical forms outside the schema. And in both subsystems, a spatial
situation can often be conceptualized in more than one way, so that it is amenable to
alternative schematizations.

4.2 Where spoken and signed language differ

Beside the preceding commonalities, though, the two language modalities have been
seen to differ in a number of respects. First, they appear to divide up into somewhat
different sets of subsystems without clear one-to-one matchups. Accordingly, the spatial
portion of the spoken language closed-class subsystem and the classifier subsystem of
signed language may not be exactly corresponding counterparts, but only those parts
of the two language modalities closest to each other in the representation of schematic
spatial structure. Second, within this initial comparison, the classifier subsystem seems
closer to the structural characteristics of visual parsing than the closed-class subsystem
in all of the following ways: It has more basic elements, categories, and elements per
category in its schematic representation of spatial structure. Its category membership
exhibits much more gradient representation, in addition to discrete representation. Its
elements and categories exhibit more iconicity with the visual in the pattern in which
they are clustered in an expression, in their observance of an object/action distinction,
in their physical realization, and in their progression through time. It can represent only
a narrow temporal aperture in an expression (and only a narrow spatial aperture as well,
though this difference from spoken language might not reflect visual fidelity). It can
represent many more distinct elements and categories together in a single expression.
It can more readily select categories and category elements independently of each
other for representation in an expression. And it avoids pre-packaged category-element
combinations as well as generalizations of their range and processes for their extension
or deformation.

4.3 A new neural model

In its strong reading, the Fodor-Chomsky model relevant here is of a complete inviolate
language module in the brain, one that performs all and only the functions of language
without influence from outside itself – a specifically linguistic ‘organ’. But the evidence
assembled here challenges such a model. What has here been found is that two different
linguistic systems, the spoken and the signed, both of them undeniably forms of human
language, share extensive similarities but – crucially – also exhibit substantial differences
in structure and organization. A new neural model can be proposed that is sensitive to
this finding. We can posit a ‘core’ language system in the brain, more limited in scope
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 347

than the Fodor-Chomsky module, that is responsible for the properties and performs
the functions found to be in common across both the spoken and the signed modalities.
In representing at least spatial structure, this core system would then further connect
with two different outside brain systems responsible, respectively, for the properties and
functions specific to each of the two language modalities. It would thus be the interaction
of the core linguistic system with one of the outside systems that would underlie the
full functioning of each of the two language modalities.
The particular properties and functions that the core language system would pro-
vide would include all the spoken-signed language properties in section 4.1 specific
to spatial representation, though presumably in a more generic form. Thus, the core
language system might have provision for: using individual unit concepts as the basis for
representing broader conceptual content; grouping individual concepts into categories;
associating individual concepts with overt physical representations, whether vocal or
manual; combining individual concepts -and their physical representations – under
certain constraints to represent a conceptual complex; and establishing a subset of
individual concepts as the basic schematic concepts that, in combinations, represent
conceptual structure.
When in use for signed language, this core language system might then further
connect with particular parts of the neural system for visual perception. I have previously
called attention to the already great overlap of structural properties between spoken
language and visual perception (see Talmy 2000a, ch. 2), which might speak to some
neural connection already in place between the core language system and the visual
system. Accordingly, the proposal here is that in the case of signed language, still further
connections are brought into play, ones that might underlie the finer granularity, iconic-
ity, gradience, and aperture limitations we have seen in signed spatial representations.
When in use for spoken language, the core language system might further connect
with a putative neural system responsible for some of the characteristics present in
spoken spatial representations but absent from signed ones. These could include the
packaging of spatial elements into a stable closed set of patterned combinations, and a
system for generalizing, extending, and deforming the packets. It is not clear why such
a further system might otherwise exist but, very speculatively, one might look to see if
any comparable operations hold, say, for the maintenance and modification of motor
patterns.
The present proposal of a more limited core language system connecting with
outlying subsystems for full language function seems more consonant with contempo-
rary neuroscientific findings that relatively smaller neural assemblies link up in larger
combinations in the subservience of any particular cognitive function. In turn, the
proposed core language system might itself be found to consist of an association and
interaction of still smaller units of neural organization, many of which might in turn
participate in subserving more than just language functions.
348 LANGUAGE, COGNITION AND SPACE

Notes
1 Talmy (2003) has been reprinted as the present paper with the permission of Lawrence
Erlbaum. The references have been updated. Since the initial publication, all of section
2 on spoken language has been greatly expanded and refi ned in Talmy (2006). And the
implications of spoken-signed differences for the evolution of language are explored in
Talmy (2007), which also appears on the author’s website:
http://linguistics.buffalo.edu/people/faculty/talmy/talmyweb/index.html
2 I here approach signed language from the perspective of spoken language because it is
not at this point an area of my expertise. For their help with my questions on signed lan-
guage, my thanks to Paul Dudis, Karen Emmorey, Samuel Hawk, Nini Hoiting, Marlon
Kuntze, Scott Liddell, Stephen McCullough, Dan Slobin, Ted Suppala, Alyssa Wolf, and
others – who are not responsible for my errors and oversights.
3 As it happens, most motion prepositions in English have a polysemous range that covers
both the unbounded and the bounded sense. Thus, through as in I walked through the
tunnel for 10 minutes refers to traversing an unbounded portion of the tunnel’s length,
whereas in I walked through the tunnel in 20 minutes, it refers to traversing the entire
bounded length.
4 The ‘classifier’ label for this subsystem – originally chosen because its constructions
largely include a classifier-like handshape – can be misleading, since it names the whole
expression complex for just one of its components. An apter term might be the ‘Motion-
event subsystem’.
5 As the only apparent exception, a ‘demoted Figure’ (see Talmy 2000b, ch. 1) can acquire
either of two ‘demotion particles’ – e.g., English with and of – that mark whether the
Figure’s path had a ‘TO’ or a ‘FROM’ vector, as seen in The fuel tank slowly filled with gas
/ drained of its gas.
6 The size and shape specifiers (SASS’s) in signed languages do permit movement of the
hands to trace out an object’s contours, but the hands cannot at the same time adopt a
shape representing the object’s path.
7 The behavior here of ASL cannot be explained away on the grounds that it is simply
structured like averb-framed language, since such spoken languages typically can repre-
sent concurrent Manner outside a narrow aperture, in effect saying something like: ‘I
walking / running / driving / flying carried the memo to the home office’.

References
Bennett, David C. (1975) Spatial and temporal uses of English prepositions: An essay in
stratificational semantics. London: Longman.
Bowerman, Melissa. (1989) Learning a semantic system: What role do cognitive
predispositions play? In Mabel L. Rice and Richard L. Schiefelbusch (eds) The
Teachability of Language. Baltimore PH: Brookes Pub. Co.
Bowerman, Melissa. (1996) The origins of children’s spatial semantic categories:
Cognitive vs. linguistic determinants. In J.J. Gumperz and S.C. Levinson (eds)
Rethinking Linguistic Relativity 145–176. Cambridge, UK: Cambridge University
Press.
 HOW SPOKEN LANGUAGE AND SIGNED LANGUAGE STRUCTURE SPACE DIFFERENTLY 349

Brugmann, Claudia and Macaulay, Monica. (1986) Interacting Semantic Systems:

Mixtec expressions of location. In Proceedings of the Thirteenth Annual Meeting
of the Berkeley Linguistics Society 315–328. Berkeley: Berkeley Linguistics Society.
Clark, Herb. (1973) Space, time, semantics, and the child. In Timothy E. Moore (ed.)
Cognitive Development and the Acquisition of Language. New York: Academic
Press.
Emmorey, Karen. (2002) Language, Cognition and the Brain: Insights from Sign
Language Research. Mahwah NJ: Lawrence Erlbaum.
Fillmore, Charles. (1968) The case for case. In Emmon Bach and Robert T. Harms
(eds) Universals in Linguistic Theory. New York: Holt, Rinehart and Winston.
Gruber, Jeffrey S. (1965) Studies in lexical relations. PhD dissertation, MIT.
Reprinted as part of Lexical structures in syntax and semantics, 1976. Amsterdam:
North-Holland.
Herskovits, Annette. (1982) Space and the prepositions in English: Regularities and
irregularities in a complex domain. PhD dissertation, Stanford University.
Imai, Shingo. (2003) Spatial Deixis: How Demonstratives Divide Space. Doctoral
dissertation. University at Buffalo, State University of New York.
Jackendoff, Ray. (1983) Semantics and Cognition. Cambridge, MA: MIT Press.
Leech, Geoffrey. (1969) Towards a semantic description of English. New York:
Longman Press.
Liddell, Scott. (2003) Sources of meaning in ASL classifier predicates. In Karen
Emmorey (ed.) Perspectives on Classifier Constructions in Sign Language 199–
220. Mahwah, NJ: Erlbaum.
Mark, David M. and Smith, Barry. (2004) A science of topography: From qualitative
ontology to digital representations. In Michael P. Bishop and John F. Shroder
(eds) Geographic Information Science and Mountain Geomorphology 75–100.
Chichester, England: Springer-Praxis.
Talmy, Leonard. (1983) How language structures space. In Herbert L. Pick, Jr. and
Linda P. Acredolo (eds) Spatial orientation: Theory, research, and application.
New York: Plenum Press.
Talmy, Leonard. (2000a) Toward a Cognitive Semantics, volume I: Concept structuring
systems. Cambridge, MA: MIT Press.
Talmy, Leonard. (2000b) Toward a Cognitive Semantics, volume II: Typology and proc-
ess in concept structuring. Cambridge, MA: MIT Press.
Talmy, Leonard. (2003) The representation of spatial structure in spoken and signed
language. In Karen Emmorey (ed.) Perspectives on Classifier Constructions in Sign
Language 169–195. Mahwah, NJ: Lawrence Erlbaum.
Talmy, Leonard. (2006) The fundamental system of spatial schemas in language.
In Beate Hampe (ed.) From perception to meaning: Image schemas in Cognitive
Linguistics 199–234. Berlin: Mouton de Gruyter.
Talmy, Leonard. (2007) Recombinance in the Evolution of Language. In Jonathon
E. Cihlar, David Kaiser, Irene Kimbara and Amy Franklin (eds) Proceedings of
the 39th Annual Meeting of the Chicago Linguistic Society: The Panels. Chicago:
Chicago Linguistic Society.
Zubin, David and Soteria Svorou. (1984) Orientation and gestalt: Conceptual organ-
izing principles in the lexicalization of space. With S. Choi. In David Testen,
Veena Mishra and Joseph Drogo (eds) Lexical semantics. Chicago: Chicago
Linguistic Society.
14 Geometric and image-schematic patterns in
gesture space
Irene Mittelberg

1 Introduction

The human body exists, moves, interacts, and communicates in space and time.
Inseparable from the human body, manual gestures, too, unfold and vanish in space
and time. They derive their meaning in part from the coinciding speech and in part
from particular combinations of hand shapes, hand motions, and their location in
gesture space. Over the last few decades, research on co-speech gesture and signed
languages has shown that these dynamic visuo-motor modalities do not only exploit
various dimensions of physical space as their articulatory medium, but that they can also
provide a window into how physical, conceptual, social, and discourse spaces interact
(e.g., Emmorey and Reilly 1995; Liddell 2003; Kendon 2004; McNeill 1992, 2000, 2005;
Müller 1998; Núñez and Sweetser 2006; Parrill and Sweetser 2004; Sweetser 2007; Taub
2001; Wilcox 2000; Wilcox and Morford 2007).
Within cognitive linguistics, gesture data from typologically different languages
have proven to be a valuable source of multimodal evidence for conceptual meta-
phor and particularly for spatial metaphor. A considerable body of research done
on metaphorical gestures, e.g., representations of abstract ideas and structures, has
demonstrated their capacity to reveal source domain information not necessarily
captured by concurrent verbal expression (Bouvet 2001; Cienki 1998a, 1998b; Cienki
and Müller to appear; McNeill 1992, 2005; Mittelberg to appear; Müller 1998, 2004b;
Núñez 2004; Sweetser 1998, 2007; inter alia). Moreover, a recent experimental study
(Cienki 2005) suggests that basic image and force schemas manifest themselves in
gesture. Due to their specific materiality and logic, gestures are particularly apt at
depicting spatial and dynamic properties of conceptual structure and processes, thus
supporting the theory of the embodied mind (Gibbs 1994, 2003, 2006; Lakoff and
Johnson 1980, 1999).
Indeed, basic physical activities that involve hand motions and/or bodily movement
through space – such as walking, grasping, touching, pointing, placing, and exchanging
physical objects – exhibit metaphorical correspondences in the domains of thought and
speech: we understand something if we can ‘grasp’ it, we ‘walk’ people through texts,
‘point out’ certain aspects, ‘push an issue’, or try to ‘get ideas across’ to our interlocutors
(cf. Sweetser 1992). Exploring how such habitual actions play out in gesture, the aim
of this paper is to offer insights into the ways in which scholars employ gestures to

351
352 LANGUAGE, COGNITION AND SPACE

illustrate their discourse about abstract knowledge domains. On the basis of academic
discourse videotaped in linguistics courses, I will show how gestural depictions may
bring intangible subject matters into physical existence that can be shared by professors
and their students. The main point of interest here is the spatialization of abstract
information pertaining to grammatical concepts and theories. I will demonstrate that
the prominent hand shapes and motion patterns that were found to recur across subject
matters and speakers form a set of patterns which are reminiscent of simple geometric
figures (e.g., squares, triangles, cubes, circles), as well as image and motor schemas
proposed in the cognitive linguistics literature (e.g., object, path, balance, support,
container, rotation; cf. Hampe 2005; Johnson 1987; Lakoff 1987; Mandler 1996, 2004;
Talmy 1988). The term geometric here refers to basic shapes evoked by constellations
of arms and hands and by forms resulting from imaginary lines drawn in the air. It
will be suggested that a kind of ‘common-sense geometry’ (Deane 2005:245) may be,
among other kinds of conceptual structures and motor routines, one of the factors
that motivate what have turned out to be fairly systematic representations of linguistic
form, grammatical categories, and syntactic relations. In view of the important role
such embodied schemas have been found to assume in language acquisition (Mandler
1996, 2004), language per se (Talmy 1988), and also in the visual arts (Johnson 1987;
Mittelberg 2002, 2006, in prep.), it might not be all that surprising to also see some
of them reflected in gesture. The aim here is to show that discerning them in this
dynamic bodily modality is useful in diagnosing less monitored aspects of cognition
during communication.
While the work presented here is part of a larger study investigating how such
patterns play into the iconic, metaphorical, and metonymic meaning construction in
multimodal discourse (Mittelberg 2006, 2007, 2008), the discussion below will focus
almost exclusively on the material side of the semiotic processes that seem to ground
abstract thought in the speakers’ bodies and the surrounding space.1 This paper is thus
about how abstract information is spatially represented through gesture – and not
about the gestural depiction of spatial concepts or scenes per se (see Sweetser 2007 for
an overview).
Before moving into the heart of the study, let us look at an example from the data
in order to get a first impression of how gestures may ascribe meaning to chunks and
regions of space. In the sequence from which the image below is taken (Figure 1), the
speaker talks about the difference between main verbs and auxiliaries. During his
explanation leading up to this particular gesture, he points to instances of both verb
types contained in sentences projected onto the screen behind him. He then goes on
to say that auxiliaries such as ‘have’, ‘will’, ‘being’ and ‘been’, ‘must all belong to some
subcategory’. Upon mentioning ‘some subcategory’, he produces the gesture shown
below, consisting of two hands that seem to be loosely holding an imaginary object.
The extended arms and almost flat hands jointly evoke two diagonally descending lines.
The meaning of the term ‘subcategory’ is effectively represented by a gesture that
is produced in a comparatively low region of gesture space, low not only in relation to
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 353

the speaker’s body, but also in relation to preceding and subsequent gestures. In fact,
the hand configuration appears well below the region where this speaker and also the
other subjects of the this study produce the majority of gestures referring to grammatical
categories and sentence structure. It is thus an unusual, marked usage of space (Waugh
1982), which receives some of its semantic properties in relation to the unmarked region
of gesture space (in front of the speaker’s torso) which indirectly functions as a point
of reference.

Figure 1. Gesture representing ‘subcategory’ placed comparatively low in gesture space

If one were to accompany the same term (‘subcategory’) with the same object gesture
but located, say, in front of one’s chest, the effect of the gestural illustration would
not be as insightful. And, if one were to produce the same gesture on the mention
of a word referring to a concrete item, it would express that concrete entity and not,
as in this case, an abstract category. Here, the abstract category is metaphorically
represented in terms of an imaginary physical object (or container) that fills the
space between the two hands. It can be seen as reflecting the metaphorical concept
IDEAS ARE OBJECTS or CATEGORIES ARE CONTAINERS (Lakoff and Johnson
1980). At the same time, a second spatial metaphor is evoked: the ‘subcategory’ is
literally placed underneath the superordinated category it relates to. In the course
of the paper, we will explore various ways in which space becomes meaningful in
gestural representations of grammar.
The structure of the chapter is as follows: section 2 describes the data and meth-
odology of this study. Section 3 presents the results of the form analysis, providing an
overview of the prominent hand configurations and motion patterns. In section 4, the
findings are discussed in light of A) image and motor schemas proposed in the cognitive
linguistics literature and B) issues of object representation and spatial relations more
generally. The chapter concludes with a summary of the main characteristics of the
gestures discussed and suggestions for further research.
354 LANGUAGE, COGNITION AND SPACE

2 Data and methodology: discourse genre, transcription, and

coding parameters

2.1 Corpus

The corpus designed for this research comprises twenty-four hours of naturalistic aca-
demic discourse and co-speech gestures produced by four linguists (all native speakers
of American English; three females and one male). The subjects were videotaped while
lecturing in introductory linguistics courses at two American universities. The focus of
attention is on the communicative behavior of the professor lecturing; student behavior
and teacher–student interaction are not considered here. Topics covered include general
aspects of morphology, syntax, and phonology as well as different linguistic theories:
generative grammar, emergent grammar, and relational grammar. Correspondingly, a
major part of the discourse revolves around the introduction of new concepts and techni-
cal terms. In this highly specialized type of multimodal discourse, the objects referred
to are for the most part abstract entities and structures: linguistic units (morphemes,
words, phrases, etc.), grammatical categories (verb classes, cases, semantic roles, etc.),
syntactic structures (clauses, sentences, etc.), as well as operations (the active-passive
transformation, subordination, reiteration, etc.). In search for multimodal representa-
tions of these entities, the corpus was assessed from a thematic point of view, selecting
and capturing episodes in which gestures portraying grammatical phenomena occurred.
Such ‘referential gestures’ may depict, according to Müller’s functionalist typology of
gestures (1998:110–113), objects, attributes of objects and people, actions, behaviors,
etc. Müller further distinguishes referential gestures of concrete entities from gestures
depicting abstract entities. As most of the gestures discussed here refer to abstract
phenomena, they can be said to be essentially metaphorical in nature. In each semiotic
act different iconic and indexical (i.e., metonymic) modes were found to interact to
different degrees, but we will not be able to go into these issues of interpretation here
(see McNeill 1992, 2005 on gesture categorization and Mittelberg 2008 and Mittelberg
and Waugh 2009 for more details on the interaction of metaphor and metonymy in
meta-linguistic gestures).
Not only the subject matter talked about, but also cultural practices and pedagogical
routines influence the kinds of gestures that accompany meta-linguistic discourse.
Given that in Western cultures language is represented as horizontally oriented strings
of written words, habits of writing and reading from left to right and filling text spaces
from top to bottom can be expected to motivate, among other factors, the graphic
representation of language and grammar in gesture. Common practices in grammar and
linguistics courses also need to be taken into account, such as diagramming sentence
structure and dissecting sentences into functional parts (see Jakobson 1966 for an
account of why grammatical patterns lend themselves so well for graphic representation).
These factors as well as the use of mediational tools such as blackboards, whiteboards,
overhead projectors, and laptops influence the kinds of gestural signs produced in this
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 355

specific context as well as their exact execution in relation to the technical equipment
and the spatial environment of the classroom.
Working with multimodal usage data involves a series of steps which will be only
briefly sketched here. First, the speech of each segment was transcribed adapting the
discourse transcription convention provided by Du Bois and colleagues (Du Bois et al.
1993). Then, the gestures were coded according to their kinetic features (see section
below) and, in relation to the concurrent speech, the exact speech–gesture synchrony
was documented in annotated transcripts. 2 To this end, the course of each gestural
movement (which may include onset, preparation, peak, hold, and return to rest) gets
translated into typographic representations, superimposed on the transcribed speech.
Each gesture was traced from the moment the articulators (here hands and arms) begin
to depart from a rest position until the moment when they return to rest or relaxation.
Such a full movement excursion (Kendon 2004:111) is called a gesture-unit (G-unit):
‘The G-unit is defined as the period of time between successive rests of the limbs; a
G-unit begins the moment the limb begins to move and ends when it has reached a
rest position again’ (McNeill 1992:83). Only gestures articulated with hands and arms
were taken into account, leaving aside facial expressions, gaze, self-grooming, and
movements of the head and torso (for more details on methods and sample transcripts
see Mittelberg 2007).

2.2 Physical gesture features: hand shape, palm orientation, and movement

In gesture research, the most widely used coding parameters are hand presence and
hand dominance, hand shape, palm orientation, movement (trajectory and type), and the
location in gesture space where a gesture is performed (cf. McNeill 1992, 2005; Kendon
2004; Müller 1998, 2004; Webb 1996). These kinetic features were also used to describe
the referential gestures in the present corpus, thereby determining those qualities of
a gesture gestalt that contribute most significantly to its meaning and function. For
example, in certain cases, the movement proved to be more salient with respect to the
meaning of a gesture than the particular shape of the hand performing the movement
(e.g., in certain pointing gestures it did not matter whether the hand pointing was a
relaxed flat hand or whether the index finger was extended); in other cases, the hand
shape is more salient than the contextual movements (e.g., in the case of hands forming
a closed fist); and in yet other cases, both dimensions are significant (e.g., a push with
an open palm facing the addressee, thus building a barrier and evoking the idea of
‘stop’ or ‘rejection’). As we saw in the subcategory example above, the location in which
the gesture is produced may also significantly contribute to its meaning and function.
In order to categorize the hand shapes, a data-driven typology of manual signs
was developed.3 The data were searched for hand shapes and arm configurations that
recurred across speakers and contexts, and a label was assigned to each prominent form.
For example, one of the most frequently used hand shapes is a flat open hand with the
palm turned upwards, thus building a sort of surface. Here it seemed worthwhile to
356 LANGUAGE, COGNITION AND SPACE

build on conventions introduced by Müller (2004) in her study of forms and functions
of the palm-up open hand gesture (hereafter referred to as ‘puoh’). Each variant of the
open hand gesture that occurred in the data was given an abbreviation such as ‘puoh’,
indicating the orientation of the palm, plus a short name evoking the degree of openness
of the hand (‘tray’, ‘cup’, ‘lid’, etc.) as well as an indication of which hand performed the
gesture. For instance, ‘puoh-tray-lh’ stands for a flat palm-up open hand, produced with
the left hand, evoking the shape of a tray. Or, ‘pcoh-box-bh’ stands for another frequent
gesture consisting of two hands held apart, with both palms being held vertically and
facing each other and thus pointing to the center of gesture space (i.e., ‘pcoh’ stands for
palm-center open hand and the ‘center’ denotes the direction that the palm is facing).
A variant of this gesture was discussed above in the subordination example (Figure 1).
Gestures typically involve some sort of movement through space and are as such
a comparatively fluid medium: they usually vanish as quickly as they emerge, often
melting into one other. Describing such manual actions entails the range and trajectory
of the performed motion (for example, along horizontal, vertical, or diagonal axes)
as well as the manner of the movement (straight line, wave, rotation of the wrist,
etc.). When a gesture appeared unusually forceful, the energy level with which the
movement was carried out was taken into account. Instances in which a movement
is discontinued or a configuration is being held (e.g., the so-called gesture hold, cf.
McNeill 1992) were also recorded. In keeping with the notational conventions used
for hand shapes, the prominent movement patterns were given labels that inform
about their trajectory and manner. For example, ‘vert-trace-rh’ signifies a line that
is traced vertically with the right hand, and ‘wrist-rota-lh’ refers to a wrist rotation
performed with the left hand.

2.3 Location in gesture space

Manual gestures take shape in physical space. The range, organization, and preferred use
of a person’s gesture space is conditioned by factors such as age (children vs. adults), cul-
tural background, and personal style, among others (cf. Calbris 1990; Goldin-Meadow
2003; Kendon 2004; McNeill 1992; Müller 1998). Not surprisingly, the space param-
eter has entered gesture research in various ways, shedding light on spatial cognition,
culturally-determined conceptualizations of space, etc. (cf. Haviland 2000; Levinson
1997, 2003; Núñez and Sweetser 2006; Sweetser 2007). Gesture space is relative to, and
constituted by, the position and posture of the speaker-gesturer who, in each commu-
nicative instance, sets up the coordinates of gesture space around her, according to the
dimensions and movements of her body, her gestural articulators (here arms and hands),
her physical environment, and, if applicable, also according to the interpersonal, social
space spanning between herself and her interlocutor(s). The location of a gesture can be
described from various angles: relative to the gesturer’s body, relative to previously or
subsequently produced gestures, or relative to the addressee’s gesture space. In gesture,
space is exploited to indicate and describe the location of objects, people, places, events,
and ideas, as well as the spatial relationships among entities and persons, a task that
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 357

is generally more difficult to master with purely linguistic means (cf. Emmorey 1996;
Emmorey and Reilley 1995 regarding the use of space in signed languages).
In terms of the perspective from which a scene or an object may be described in a
given speech event, the speaker-gesturer can represent alternate viewpoints: observer
viewpoint, character/participant viewpoint, as well as the addressee’s viewpoint (cf.
McNeill 1992:118–25; Sweetser 2007). It is probably a matter of teaching experience
and pedagogical awareness whether a teacher assumes her or his own point of view or
the audience’s perspective. In any event, these considerations determine how the use of
gesture space is organized. When freely gesturing (and not pointing at information on
the blackboard or screen), the professors videotaped for the present study were most
of the time facing their student audience, and both observer viewpoint and addressee’s
viewpoint could be made out in their gestural descriptions of grammatical categories
and structures. For example, the subjects alternatively illustrated the word order in a
sentence by drawing an imaginary line starting either on the left side and ending on
the right side of their body, or in the opposite direction, from the students’ left to the
students’ right side. Some cognitive and perceptive flexibility thus needs to be assumed
at both ends of the speech and gesture event (for a discussion of frames of reference in
ASL see Emmorey 1996; Liddell 2003; Wilcox and Morford 2007).
To document the locations where gestures occur and the trajectory they trace,
gesture researchers have developed systems to compartmentalize gesture space into
sectors. For example, McNeill established a shallow disk consisting of concentric squares
superimposed on a drawing of a seated person, thus reflecting the semi-experimental
set-up in which speakers were asked to retell animated cartoons (McNeill 1992:86–89,
2005:274). Since the conditions under which the present data were collected were
not controlled in any way, and since teachers tend to walk around in the classroom
and constantly change their position and the angle with which they turn towards the
audience, blackboards, overhead projectors, laptops, etc., there were no stable space
coordinates. Instead of investigating the relative density of occurrence of certain gesture
types in particular sectors of gesture space (e.g., in relation to different body parts), or
correlating gesture location and discourse function, which are possible ways to exploit
the space factor in gestural communication (cf. McNeill 1992: 88ff.), one of the main
interests here was to determine the ways in which the speakers’ use of gesture space
could reveal aspects of their spatial representations of abstract phenomena. This is, as
will be shown below, where different geometric and image-schematic representations
of linguistic form and structure come into play.

3 Study: prominent hand configurations and motion patterns in

meta-grammatical gestures

The aim of this section is to provide an overview of the prominent gestural forms that
were found to illustrate verbal explanations of linguistic form, grammatical relations
and syntactic functions. The point of departure here was the physical forms of gestures
exhibited in the data (i.e., hand figurations, manual actions, or imaginary lines drawn
358 LANGUAGE, COGNITION AND SPACE

in the air). Only then did the analysis turn to the abstract ideas and structures which
gestural signs stand for in a given moment, taking into account the concurrent speech.
Since the scope of the paper does not allow for a detailed account of the cross-modal
distribution of semantic features and pragmatic functions, the discussion below will be
mainly restricted to the material properties of the gestures (see Mittelberg 2006, 2008
and Mittelberg and Waugh 2009 for detailed content analyses).4

3.1 Prominent hand and arm configurations

The gestalt of a given gesture relies on the semiotic collaboration of several parameters,
of which the hand shape is only one. Yet, the hand shape and/or arm configuration can
be said to be salient in a gesture if it is the most notable feature in the process of its
articulation. While most of the gestures to be discussed below involve some kind of
movement, it is the hand shapes and arm configurations that, especially when being
held for a moment, tend to stand out perceptually. As in the example discussed above
(Figure 1), the movement leading up to the object-holding gesture is not as perceptually
and semantically salient as the bimanual configuration produced on the mention of the
term ‘subcategory’. Factoring in the speech content it becomes evident that both the
specific hand and arm configuration plus its location contribute key qualities to the
bi-modally achieved message.
Across the four subjects, the data show recurrent representations of linguistic units as
readily manoeuvrable objects. There are several different ways of holding and manipulating
such imaginary items, some of which allude to the geometry and/or size of the object,
while in other cases no or very little information about the size or form of the object can be
inferred. One way to refer to an abstract item is to seemingly hold something placed on a
palm-up open hand (puoh). The degree to which the hand is flat, relaxed, or cupped varies
from case to case. The potential functions of this basic hand shape have been matched with
the actions of holding, presenting, or offering an imaginary object for inspection, and these
functions have been observed in diverse contexts (Müller 2004). Variants of the palm-up
open hand gesture, also called ‘palm presentation’ gestures (Kendon 2004) or ‘conduit
gesture’ (McNeill 1992, 2005), were frequently observed in the teaching contexts under
investigation here, especially when professors talk about abstract categories or linguistic
examples not visibly present in the immediate environment (an alternative would be to
point to words written on the blackboard).
The following list comprises the different open-hand variants found in the data,
some of which will be illustrated and discussed in more detail below. As indicated in
the methods section above, each type was assigned an abbreviation referring to the
openness and orientation of the palm (such as ‘puoh’) plus a ‘name’ and an indication
of which hand was used (some of the palm-up open hand abbreviations follow Müller
2004). Finally, an abbreviation signals which hand was used. While, theoretically, the
hand shapes listed below could be produced simultaneously by each hand, they were
for the most part observed to be executed with only one hand at a time.
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 359

Single open and closed hands

rh: right hand; lh: left hand

A. puoh-tray-lh/rh hand as flat surface, supporting imaginary objects

B. puoh-cup-lh/rh hand with curled fingers, forming a receptacle
C. pfoh-stop-lh/rh ‘f ’ stands for ‘front’, palm facing audience
D. pdoh-lid-lh/rh ‘d’ stands for ‘down’, flat hand
E. pdoh-claw-lh/rh open hand facing down, fingers curled
F. pcoh-blade-lh/rh ‘c’ stands for palm facing center of gesture space
G. fist-lh/rh closed fist

The last gesture type listed above is in fact the opposite of an open hand: it is a closed
hand forming a fist. Other hand shapes involving specific finger configurations include
‘measure’ (thumb and index finger are stretched apart, tips pointing upwards, similar to
the way one might take measure in inches), ‘pinch’ (the tips of index finger and thumb
are pressed against one another), and ‘scrunch’ (fingers are held closely together, facing
audience, tips pointing towards the floor).

Specific finger configurations

H. t-i-measure-lh/rh ‘t’ for thumb, ‘i’ for index

I. pinch-lh/rh fingertips of index and thumb pressed together
J. scrunch-lh/rh similar to pinch, but different orientation and
finger configuration, back of hand facing audience,
tips pointing towards floor

Another category of gestural shapes engages not only hands but also parts of a speaker’s
arm(s). Most of the observed pointing gestures fall into this category, as they are usually
produced with both an extended arm and hand, exhibiting either an extended index
finger [‘ind-index’] or the entire, mostly relaxed hand [‘hand-index’]. Together, hand and
arm build a vector, or a path, leading to the targeted referent (e.g., an object, a person,
information written on the blackboard, or to certain locations in gesture space right in
front of the speaker). In addition, there were arm configurations depicting chunks of a
syntactic tree diagram by mirroring the triangle-like shape of such diagonally downward
branching structures [‘diag-arm’].

Pointing gestures and other kinds of arm configurations

K. ind-index-lh/rh [pointing with generic extended index finger]

L. hand-index-lh/rh [pointing with full, relaxed hand]
M. diag-arm-lh/rh [arm held diagonally, forming a triangle-like shape
if both arms are involved]
360 LANGUAGE, COGNITION AND SPACE

Other gestures observed in the data are always performed with two hands, evoking an
internal structure, or what has been called ‘syntax’ (cf. Kendon (2004:275ff.) on Open
Hand Supine gestures with lateral movement). Examples are the gesture mentioned
above in which the imaginary object is held between two hands, or a gesture conveying
the idea of a balance by seemingly weighing two things, with two palm-up open hands
moving alternately up and down, one on each side of the body.

Open hand variants performed with both hands (bh)

N. puoh-tray-lateral [balance]
O. puoh-cup-lateral [balance]
P. puoh-sym-offshoot [hands thrown laterally up into the air, from center
outward]
Q. pcoh-box-bh [refers to an elongated object held between both
hands]

The data were searched for instantiations of each of these identified shapes (and
movement patterns, to be discussed below) across topics and speakers. For most
of these forms, several instances were identified and assessed with regard to the
concurrent speech content and the overall meaning of the multimodally achieved
representation. Below, a selected set of these hand shapes will be illustrated and
discussed in more detail.

3.1.1 Single open and closed hands: surfaces and containers for abstract entities

Comparatively small linguistic units, such as morphemes, words, and categories were
represented as objects seemingly resting on a variant of the palm-up open hand gesture
or inside a closed fist. The gestures shown in Figures 2 and 3 are instances of palm-up
open-hand gestures with a flat palm or cupped hand evoking a kind of surface or a
receptacle where items can be placed (i.e., imagined) and presented to the audience.
From just looking at the hand shape it might not be clear whether the action the hand
is performing represents an act of offering, receiving, showing, or requesting an item.
In conjunction with the speech content, however, it turns out that the gesture in Figure
2, for instance, represents the action of receiving. It denotes a technical term, namely
the semantic role ‘recipient’, by showing an open hand ready to receive an object. A
similarly shaped gesture fulfills a different function in Figure 3, where the speaker is
explaining the fact that an idea can materialize in discourse in the form of a noun or a
verb. On the mention of ‘a noun’ she creates a sort of tray on which the emerged form
is being presented to the audience.
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 361

Figure 2. puoh-cup stands for recipient Figure 3. puoh-tray stands for a noun

By seemingly handling small imaginary objects, linguistic units are thus reified and
made graspable for the mind. Flat open hands provide surfaces, planes, or, put more
generally, support structures, exposed to the eye of the addressee, on which the item
referred to in the speech modality can be imagined. Alternatively, the absence of an
entity or the expectation to receive something can be signaled. Similar functions can
be performed by cupped hands (with clearly curled fingers), building a sort of open
container (see Figure 12 below). While the focus here is on the formal properties of
open hand gestures, it needs to be kept in mind that the meaning of a gesture results
from both its form and the function it plays in a given speech event (see Müller 2004
for a detailed account of forms and uses of the palm-up open hand gesture and also
Kendon 2004). Abstracting from these pragmatic considerations, the central point
here is that these open hand gestures seem to embody the image schemas SUPPORT
(Mandler 1996) and CONTAINMENT (Johnson 1987; Lakoff and Johnson 1980)
respectively.
As the next examples suggest, imaginary small objects can also be held in tightly
closed hands. In Figure 4, the speaker refers to grammatical ‘knowledge’ while forming
a fist (left hand) and to the idea that ‘knowledge becomes automatized’ with usage
when forming a second fist (right hand). While talking about the fact that the word
‘teacher’ consists of two parts (the morphemes ‘teach-’ and ‘-er’), the speaker in Figure
5 encloses each component in a fist: the right hand holds the lexical morpheme ‘teach-’
and the left hand the grammatical morpheme ‘-er’. The spatial difference between the two
hands evokes the conceptual difference between the two functionally distinct elements
forming one word, thus instantiating the metaphorical concept PHYSICAL DISTANCE
IS CONCEPTUAL DISTANCE (Sweeter 1998). At the same time, the two hands jointly
allude to the internal structure of the word ‘teach/er’. In both cases, the fists are first
formed successively and then held simultaneously, as shown in the figures below (see
Mittelberg 2008 on diagrammatic iconicity holding between the two hands).
362 LANGUAGE, COGNITION AND SPACE

Figure 4. Fist(s) for knowledge (grasp/mastery) Figure 5. Fists containing morphemes (teach-er)

While here, too, the image schema CONTAINMENT manifests itself in these gestural
representations, the fist seems to have, compared to the open hand variants, a different
semantic import. It evokes the idea of having, literally and metaphorically, captured a
concept, of having a firm grasp of it: one knows how to handle a certain phenomenon.
Inside the closed hand, there is no space for maneuvering. At the same time, the object
enclosed in the hand container is invisible and not much information about it is acces-
sible, which stands in contrast to exposing an idea on an open hand for inspection
and commentary, or alluding to the fact that one does not have an answer and is thus
‘empty-handed’ (cf. Müller 2004).

3.1.2 Different amounts of space between the articulators

We will now look at some hand shapes where the configuration of individual fingers
and the existence or nonexistence of space between the articulators play a signifi-
cant role. The two examples below represent cases of what was called a pinch in the
list provided above. A pinch involves the index finger and thumb pressed together.
For example, the gesture shown in Figure 6 expresses the idea of a precise list of
categories in the theory of relational grammar, by drawing, with the index finger
and thumb pressed together (indicating the idea of ‘precise’) a vertically descend-
ing line (depicting the idea of a ‘list’). In a similar fashion, the gesture in Figure 7
features no space between the fingertips. However, unlike the gesture in Figure 6, it
bears a stronger resemblance to what is generally known as the ring gesture due to
the slightly more rounded fingers; this gesture occurs across cultures and contexts
with different coded meanings, ranging from tangibility, to precision and perfection
(Kendon 2004; McNeill 2005; Müller 1998). Here, in the context of a syntax lecture,
it has a different designation: it co-occurs with the mention of the technical term
‘node’ which is a juncture point at the top of a branching structure in tree diagrams
used in the framework of generative grammar.
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 363

Figure 6. Pinch indicating precise list of categories Figure 7. Pinch/ring indicating ‘node’ (tree)

An alternative way to refer to small items is to seemingly hold them between the tips
of thumb and index finger, as if one were taking measure. In other words, there is some
space between the two fingers, which might suggest a virtual object filling the space.
For example, in Figure 8, the small space between index finger and thumb indicates
the compact nature of the pronoun ‘it’, alluding to the placement and function of such
minimal forms in phrasal verb constructions. The gesture in Figure 9 stands for a verb
form (‘fell’) at the end of a sentence (‘Diana fell.’).

Figure 8. Measure representing pronoun ‘it’ Figure 9. Measure representing verb ‘fell’

A gesture heavily used to represent parts of speech, words, phrases, and sentences
depicts a comparatively bigger imaginary object as being held by two, relatively relaxed,
open hands with palms facing each other. The examples below show two of the more
expansive versions in which speakers hold the hands relatively far apart to represent
364 LANGUAGE, COGNITION AND SPACE

a sentence (Figure 10) or a constituent (Figure 11). These gestures can also be said to
reflect the image schema CONTAINMENT or, if one focuses on the fact that phrases
and sentences have a beginning and an end, by the SOURCE-PATH-GOAL schema
(Johnson 1987; Lakoff and Johnson 1980, 1999).

Figure 10. pcoh-box representing a sentence Figure 11. pcoh-box representing a constituent

In view of the representations discussed so far, space seems to carry meaning in specific
ways. Although there is no direct correspondence between the amount of space extend-
ing between the articulators and the physical characteristics of the elements referred to,
there is a tendency for smaller individual linguistic units to be represented as being held
in one hand (Figures 2, 3, 5, 8, 9) and for comparably more complex constructs such
as entire phrases or sentences to be represented by objects held (or space extending)
between both hands of the speaker (Figures 10 and 11). In the latter cases, the geometry
of the objects held between two hands is specified to a higher degree than the shape
of objects seemingly sitting on open hands and remains rather undefined. In both
scenarios, however, the mind needs to fill in information according to the cues provided
by the hand constellations as well as the concurrent speech content (see Mittelberg and
Waugh 2009). It should be noted that it is difficult at times to decide whether one can
assume objects or whether it is rather about delineating the space extending between
fingers or hands.

3.1.3 Pointing gestures and specific arm configurations

While parts of the speakers’ arms were involved in different fashions in many of the
gestures discussed above, we now turn to configurations in which arms are instrumental
in the gestural sign formation. As we will see, arms may be recruited to build signposts in
pointing gestures or to directly stand for elements of the object they depict (cf. Müller’s
(1998) modes of gestural representation).
The spatial orientation and angle of pointing gestures depend each time on the
location of the object towards which they are directed. Through the act of pointing
at something in the proximity of the speaker (e.g., on the mention of a demonstra-
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 365

tive pronoun) the object is established via a vector consisting of a path evoked by the
extended arm and hand and its virtual extension leading to the targeted object. Such
deictic gestures highlight spatial relationships between the speaker and objects, loca-
tions, or people, whether they are present in the environment, imagined, or previously
introduced in the unfolding discourse (cf. Fricke 2002, 2007; Furuyama 2001; Kita 2003;
McNeill 1992, 2005; McNeill et al. 1993; Sweetser 2007; Williams 2004).
To illustrate and anchor their explanations, the speakers frequently point to infor-
mation presented on blackboards, whiteboards, or overhead screens. An example of this
is given below (Figure 12). Talking about the difference between main verbs and auxiliary
verbs, the speaker points with his right hand to words projected onto the screen behind
him (on the mention of ‘there is’), thus creating a vector between the position of his
body (i.e., the deictic center or origo of the speech act, according to Bühler 1934) and the
referent of the concurrent deictic expression. It can also be taken as an instantiation of
the SOURCE-PATH-GOAL schema with the path leading the interpreting mind to the
object referred to. Completing his sentence (started with ‘there is’), the speaker forms
with the left hand a cupped palm-up open hand gesture (on the mention of ‘the main
verb’,). A concrete example of a ‘main verb’ is being pointed at on the screen (‘taught’),
while the abstract category as such is to be imagined as being inside the cupped hand
directed towards the student audience.

Figure 12. Index (‘there is’) plus cup ‘the main verb’ Figure 13. Semantic roles ‘bounce around’

Another way of assigning meaning to space is to virtually place things in gesture space
or to simply point to locations in space, for instance when enumerating a list of things.
In the example above (Figure 13), the speaker talks about the different ‘semantic roles
that bounce around in linguistics’, and represents each type of semantic role with a
different gesture produced in a different place. The gesture shown here is made on the
mention of the term ‘agent’ (we already looked at the gesture for ‘recipient’, cf. Figure 2).
Metaphorically speaking, this gesture can be interpreted to reflect the metaphor IDEAS
ARE LOCATIONS; it can also be seen as an instance of metonymy of place (PLACE FOR
OBJECT). By dispersing categories in space, the physical distance between the assigned
locations represents the conceptual distance between the different semantic roles and
366 LANGUAGE, COGNITION AND SPACE

their respective functions (agent, patient, goal, recipient, experiencer) thus evoking
the metaphorical mapping CONCEPTUAL DISTANCE IS PHYSICAL DISTANCE
(Sweetser 1998).
As for gestural constellations involving both arms, let us look at Figure 14. Here
the speaker illustrates a part of a syntactic tree diagram by forming a triangle-like
shape, achieved with the fingertips touching at the center top and both forearms
held diagonally with elbows pointing outwards. Mirroring a part of the diagram
on the blackboard behind the speaker, the evoked pyramid directly imitates a tree
chunk. Put differently, the arms of the speaker embody conceptual structure. Such
depictions provide more substance than lines quickly traced into the air and lend,
as such, otherwise relatively fleeting representations a higher degree of stability in
space and time.

Figure 14. Phrase structure as a lateral diagonally Figure 15. index pointing to
branching tree chunk ground: subordination

Illustrating the idea of subordination, the speaker in Figure 15 combines a pointing

gesture with a representational gesture that can also be interpreted as standing for a
tree branch descending to the right lower side of her body. She indicates that there are
certain cases in which embedded sentences go ‘all the way down’, at which point she
directs her fully extended right arm towards the floor and points with her index finger
straight to the ground. As it was the case in the subcategory example discussed above
(Figure 1), the descending arm evokes a spatialization of the idea of subordination by
reaching into comparably low regions of gesture space. Alternatively, the speaker drew
the same kind of geometric configurations in the air, tracing either only a single diagonal
line or two diagonal lines downward, one to each side of her body (see also Figure 14).
This kind of dynamic representation serves as a bridge into the section below where
motion patterns will be discussed.
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 367

3.2 Motion patterns

In gesture, form can be created not only by hand and arm constellations, but also by
fleeting hand movements that draw simple lines or contours of objects in the air, thus
leaving imaginary traces in gesture space. Identifying significant motion patterns recur-
ring in the data entailed determining for each dynamic gestural gestalt those qualities
that contribute most significantly to its meaning. Again, this can ultimately only be
done in correlation with the concurrent speech content and particularly with those
speech segments that coincide with the peak, or ‘stroke’ phase, of a gesture (McNeill
1992). In most cases, hand shape and movement do interact in one way or another;
yet, the discussion below concentrates on the different trajectories and/or manners of
those hand motions that appear constitutive of the gestural signs (which in turn stand
for the abstract ideas and structures they convey).
One can generally distinguish between several types of gestural movements. For
example, the movement of a hand can result in the evocation of a form (such as the
size and shape of a guitar). It may also be influenced by the object that is involved in
the action imitated by the hand movement (such as the unlocking of a door with an
imaginary key), or it can simply imitate a manual action (such as waving at somebody)
or the manner and/or speed of a movement executed by a person or an object (for
research on motion events and the description of movement and manner in gesture
see McNeill 1992, 2000; Müller 1998; Slobin 2003). The hand movements observed in
the present data were also found to exhibit several intrinsic logics: first, movements
carried out by hands tracing straight lines or curved lines imitating the shape of a wave,
circle, or arch (these movement types bring to bear the different planes in the gesture
space such as horizontal, vertical, and front-back); second, there are pointing gestures
whose direction and range depend on the location of the object or person pointed at
(cf. section 3.1.3); third, object-oriented actions such as placing something; and fourth,
basic motor actions with no object involved, such as two hands rotating around each
other. These distinctions concur with previously made observations that a large number
of gestural shapes and movements originate in concrete object manipulation and are
abstracted from and structured by routinized interactions between the human body and
the physical and social world. Accounting for the noted variety, Müller’s (1998) system
of modes of gestural representation include manual actions such as such as drawing,
molding, enacting or embodying (see also Calbris 2003; LeBaron and Streeck 2000;
Streeck 2002). Although the schematic representations and drawings provided below
only render frozen visualizations of dynamic gestural gestalts, and while this sort of
qualitative approach needs to be complemented by quantitative investigations across
subject matters and speakers, the identified patterns offer a window into some of the
ways in which hand movements unfolding in a teacher’s gesture space may reveal aspects
of the underlying conceptualizations of abstract concepts and structures. The following
typology of gestural motion patterns was established:
368 LANGUAGE, COGNITION AND SPACE

Linear movements (horizontal/vertical/diagonal)

A. hori-trace-lh/rh horizontal line
B. vert-trace-lh/rh vertical line
C. diag-trace-rh diagonal line

D. diag-trace-ll diagonal line

E. diag-trace-lat lateral diagonal line

F. scale-lh/rh hand trace vertically organized steps/levels

G. hori-join-lat horizontal line drawn with both hands going
inward or lateral inward movement or a more
forceful push
H. hori-part-lat horizontal line drawn with both hands,

(lateral outward movement)
I. push-lh/rh/bh push away from body along a straight line, not
curved (exploiting depth along sagittal axis)
J. pull-lh/rh/bh pull toward body along a straight line, not
curved (exploiting depth along sagittal axis)

While the movements listed above exhibit linear trajectories along the major axes,
non-linear representations along the horizontal and the vertical axes also occurred;
additional non-linear configurations include both half and full circles:

non-linear traces
K. hori-wave-lh/rh wavy line traced in the air,
along a horizontal axis
L. diag-wave-lh/rh/ wavy line traced in the air,
along a diagonal axis

curves and circles

M. curve-up-lf/rh hand(s) move(s) along upper
half of circle
N. curve- dn-lf/rh hand(s) move(s) along lower
half of circle
O. circle-lh/rh/bh hand(s) complete(s) one full
cycle, rotation
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 369

Other motor actions of hands, not involving simple traces of the manipulation of imagi-
nary objects, include the following two types of rotations:

P. rotation-lateral both hands (and arms) draw circles repeatedly rotating

around one another
Q. wrist-rota-lh/rh/bh wrist rotation, occurs with different orientations

Below, I will discuss several examples of hand movements that evoke dynamic images
of abstract entities and processes – even if it is just via an imaginary trace left in the
air (the dynamic nature of the movements can unfortunately not be fully appreciated
without viewing the video clips).

3.2.1 Linear movements (horizontal and vertical traces)

Sentences and other sequences of linguistic units were found to be represented by

movements tracing the horizontal alignment of words from the left to the right of the
speaker (‘hori-trace’), or, if the viewpoint of the audience was assumed, from right to
left. A slight variation of such schematic representations of sentence structure is shown
in Figure 16 below, where the gesture starts out with both hands joined at the center of
gesture space, right in front of the upper torso of the speaker. Subsequently, the hands
move laterally outward until both arms are fully extended, as if they were tracing, as
mentioned in the concurrent speech, ‘a string of words’ (‘hori-part’).

Figure 16. A sentence as a string of words Figure 17. The infix goes into the
middle of another morpheme

The gesture in Figure 16 depicts ‘a sentence’ as a ‘string of words’ drawn horizontally in

the air, with both hands starting in front of the speaker’s chest and being pulled outward
to each side of the body. A vertically descending line was already shown in the gesture
representing a list of categories (Figure 6); it also underlies the gesture whose beginning
370 LANGUAGE, COGNITION AND SPACE

point is illustrated in Figure 17 above. After setting the stage by mentioning that words
in English may have prefixes and suffixes, the speaker explains the position infixes take
in the structure of a complex word. In the moment captured above, the speaker is just
about to insert an infix into a word stem. The idea of insertion is depicted by a well-
defined vertical trajectory traced by the hand (executed on the mention of ‘morphemes
that go right into the middle of another morpheme’), until the hand seems to hit the
base form which he quickly sketches as a container by drawing its horizontal base line
and then alluding to its two outer sides with a bimanual palm-center open hand gesture.
In addition to horizontal lateral outward movements such as the string depicted
in Figure 16, the data also exhibit lateral inward movements that are executed with a
higher energy level. For example, as shown in Figure 18, the idea that, according to
the theory of emergent grammar, boundaries between grammar and language use are
‘blurred’ is illustrated by a gesture that starts out with two hands apart, palms facing each
other, but the palms then get suddenly pushed towards each other to convey the idea of
fusion. Similarly, the speaker in Figure 19 talks about the behavior of words that like to
‘go together’ and ‘travel together to the front of the sentence’, which is portrayed by two
fists being quickly and repeatedly brought together. In both cases, physical closeness
signals conceptual closeness and is achieved through physical forceful action. We can
thus observe an interaction between image and force schemas.

Figure 18. ‘hori-join’, blurring boundaries Figure 19. ‘hori-join’, words go concepts
together (travel)

3.2.2 Non-linear traces

Let us now look at some non-linear motion patterns. The first two images below show
instances of wave-like motions along a horizontal axis. In Figure 20, the speaker draws,
on the mention of ‘non-linearity’, a wave-like graph consisting of a first curve going
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 371

down and a second one going up. The speaker in Figure 21 makes an almost identical
motion to represent the concept of ‘intonation contour’, except that the motion goes in
the opposite direction.

Figure 20. Horizontal wave for ‘non-linearity’ Figure 21. Horizontal wave for ‘intonation
contour’

There are also instances of larger arch-like structures that are executed with both hands.
The following gestural demonstration, taken from a morphology lecture, provides an
example of the understanding that the two elements that jointly build a circumfix (by
surrounding the word stem) seem to be attached at a level above the word level. In her
attempt to illustrate the hidden organization of such complex morphological structure
(or, the strings attached), the speaker makes an arch-like gesture whose initial phase is
captured in the image below (Figure 22). After holding both hands above head level,
the speaker simultaneously draws them down to waist level, one hand to the left and
one to the right of her body. The idea that the ‘circumfix encompasses the front and
back of the word’ is subsequently represented by a bimanual palm-center open hand
gesture (not shown here; it resembles the gesture in Figure 11). Her two hands seem to
be holding the entire morphological structure by its front and back, where the indica-
tions ‘front’ and ‘back’ do not refer to spaces closer to or farther away from the speaker’s
body (which would refer to the sagittal axis that runs through her body from the space
behind her back to the space in front of her). Rather, the front of the word is located to
the left of the speaker and the back of the word to her right, in accordance with the
conceptualization of written words and sentences as extending from left to right in front
of the speaker/reader/writer (in Western cultures).
372 LANGUAGE, COGNITION AND SPACE

Figure 22. Circumfix as arch gesture

Another kind of arch-like gesture was found in the context of teaching the framework
of relational grammar. Whereas the gesture in Figure 22 is a spontaneous depiction
of morphological structure, other arch-like gestures have been observed that were
motivated by a standardized diagram used in the framework of relational grammar (the
diagrams are often compared to igloos or umbrellas). For example, a speaker explains
the concept of ‘multi-attachment’ (i.e., the idea that subject and reflexive pronoun refer
to the same person) as follows: first, the right (dominant) hand rises to head level and
comes down making a slight arch-like swing to the right. Then, the left hand rises and
makes a similar arch-like movement downward (this gesture is not reproduced here). In
the videotape one sees the corresponding diagram on the blackboard in the background
of the speaker; it shows exactly the kind of lines that the speaker draws in the air.
Correspondingly, this gesture visualizes syntactic relations in terms of spatial structure:
schematic arch-like lines cutting through several zones layered on top of each other.
It is important to keep in mind that some of the gestures discussed above are
informed by a particular theoretical view of grammatical concepts and relations (i.e.,
generative grammar or relational grammar). Without the relevant theoretical back-
ground it would probably be difficult to make sense of such gestural diagrams. They
are dynamic renditions of hypothesized conceptual relations translated into spatial
configurations; without any kind of visual support, their adequate description in solely
linguistic terms would probably be less economic and also less effective (see Mittelberg
2008 for a Peircean approach on image and diagrammatic iconicity in such metaphoric
gestures). What all these gestures have in common is that they are based on theories that
rely on a specific set of metaphors representing different understandings of language
and grammar.
To conclude this section, we can say that some of the geometric gestural representa-
tions (diagonals, triangles, and arches) are in fact not the spontaneous creations of the
speakers, but they instead are rooted in scientific conventions. The manual routine of
literally drawing diagrams on paper or blackboards is likely to influence how speakers
represent connections between words or grammatical constituents via hand move-
ments through space. Those shapes and motion patterns that are created ad-hoc seem
to be motivated, at least in part, by object-oriented actions (such as drawing, writing,
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 373

and manipulating objects), and specific motor actions (e.g., wrist rotation). In all the
cases, however, the anchor points for these representations are the human body and
its articulators’ range of possible movements as well as the dimensions constituted by
the physical classroom setting and teaching tools. Embodied practices are exploited
to fleetingly visualize conceptual images of abstract entities and structures in terms
of physical objects, bodily actions, and locations in space. These forms of mediation
between the conceptual and the embodied may offer insights, as will be detailed below,
into the central role played by image and motor schemas (and their metaphorical projec-
tion) which seem to motivate and structure, at least partly, gestural representations of
abstract knowledge domains and other types of intangible things such as values and
beliefs in a systematic way.

4 Discussion: dynamic manifestations of geometric and image-

schematic patterns

The gestures examined here are ephemeral and partial representations of objects and
actions that metaphorically refer to abstract entities and operations. As the spectrum of
emergent patterns discussed above suggests, some of the gestural forms and movements
indeed reflect geometric and image-schematic representations of grammatical concepts
and structures. Due to the fluid character of the gestural medium, the schematic images
are never fully visible at once; they may find expression in a virtual trace left by a hand
movement or by invoking the manual action of holding an object. It is left to the mental
eye of the addressees, or to their own bodily experience with such actions, to fill in
the missing pieces. In what follows, I will take these observations a step further and
address some of their implications in terms of image and motor schemas (section 4.1)
and regarding geometric representations of objects and spatial relations more generally
(section 4.2).

4.1 Gestural instantiations of image and motor schemas

As we have seen above, hand shapes and movements collaborate in building holistic
gestural gestalts. The study presented here has revealed some of the ways in which
the salient properties of such multidimensional figurations give minimal information
that may evoke full schemas of objects and actions. In what follows, I would like to
elaborate the idea that the prominent patterns identified in the data can be recruited
as tangible, non-verbal evidence for image schemas which are assumed to be part
of the ‘cognitive unconscious’ (Lakoff and Johnson 1999:9–15). Rereading Johnson’s
(1987:XIV) original definition of image schemas as ‘recurring, dynamic patterns of our
perceptual interactions and motor programs that give coherence and structure to our
experience’ with gesture in mind, reinforces the assumption that gesture is a crucial
source of manifestations of such embodied patterns and that in order to account for
374 LANGUAGE, COGNITION AND SPACE

their dynamic nature one needs to consider not only visual but also kinesthetic aspects
of image schemas (see also Cienki 1998 a/b, 2005 and Sweetser 1998, 2007).
Based on gestural representations of grammar, the present work offers support for
the ‘semiotic reality of image schemas’ (Danaher 1998:190). In particular, the following
correspondences between gestural patterns (cf. section 3) and basic image schemas are
suggested (cf. Johnson 1987; Lakoff and Johnson 1980, 1999; Mandler 1996, 2004):

support (‘puoh-tray’, ‘puoh-cup’)

containment (‘puoh-cup’, ‘fist’)
object (‘puoh-tray’, ‘puoh-cup’, ‘pcoh-box’,, ‘fist’)
source-path-goal (‘hori-trace’, ‘vert-trace’, ‘diag-trace;’
deictics such as ‘hand-index’, ‘ind-index’, ‘diag-arm’)
extension (‘hori-trace’, ‘vert-trace’, ‘diag-trace;’
deictics such as ‘hand-index’, ‘ind-index’, ‘diag-arm’)
balance (‘puoh-tray-bh’, ‘puoh-cup-bh’, ‘fist-bh’, ‘sym-offshoot’)
scale (‘scale’)
center-periphery (‘sym-offshoot’ ‘hori-join’, ‘hori-part’)
cycle (‘circle-bh’, ‘wrist-rotation’, ‘rotation lateral’)
iteration (‘wrist-rotation’, ‘rotation lateral’)
front-back (‘push’, ‘pull’)
force (‘push’, ‘pull’, ‘hori-join’, ‘sym-offshoot’)

The schemas part-whole, link, contact, and adjacency are not discussed here, yet
contiguity relations (i.e., metonymy) proved to be particularly relevant for represent-
ing relationships between individual elements jointly constituting an entire phrase or
sentence (cf. Mittelberg 2008 and Mittelberg and Waugh 2009. Furthermore, basic
geometric shapes (e.g., circles, semi-circles, triangles, rectangles, squares) were identified
as well as straight and curved lines traced along horizontal, vertical, and diagonal axes, as
well as the sagittal axis (front-back). Perhaps not too surprisingly, the list above contains
for the most part spatial and spatial relations image schemas which are assumed to struc-
ture systems of spatial relations cross-linguistically (Lakoff and Johnson 1999:35). In his
experimental study on image schema manifestations in co-speech gesture, Cienki (2005)
tested the potential of image schemas (i.e., PATH, CONTAINER, CYCLE, OBJECT, and
FORCE) as descriptors for several types of gestures accompanying discourse on matters
of honesty. Results suggest that ‘images schemas are readily available, indeed “on hand”
for recruitment as gestural forms’ (Cienki 2005:435); they may be represented in the
gesture modality as either static entities or dynamic processes. Cienki also found that
gestures can invoke different schemas than the accompanying linguistic track, thus
providing additional information to discourse participants.
The array of image schemas Cienki (2005) employed in his experimental study as
well as the above list of image-schematic patterns found in the present discourse data
contains some of the schemas that belong, according to Mandler (1996:373–8), to the
preverbal, spatially structured meaning system: SUPPORT, CONTAINMENT, PATH,
and CONTACT. Mandler maintains that these image schemas are spatial representa-
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 375

tions, or spatial abstractions, that result from perceptual analysis, which in the develop-
ment of infants comes before object manipulation. 5 Moreover, such spatial analyses
performed by the infant are supposed to be important in learning the relational aspects
of language, e.g., the meaning of verbs and locative prepositions such as ‘on’ and ‘in’
(cf. Bowerman 1996; E. Clark 1973; H. Clark 1973). Although no conclusive statements
can be made on the basis of the observations presented here, Mandler’s assertions are
relevant regarding the spatialization of grammatical relations in a modality that is
utilized for communication by infants prior to language (cf. Goldin-Meadow 2003 on
gesture and language development). Mandler also concedes that dynamics and internal
feelings would be more difficult to analyze. Here, too, gesture research promises to
further augment our understanding of the bodily logic of image and particularly of
force schemas (Talmy 1988).
Recent work on image schemas comprises a variety of understandings and defini-
tions (see contributions in Hampe 2005), but overall the notion of embodiment seems
to be taken more and more literally: there is a tendency towards the realization that
the human body’s intuitive expressions and culturally-shaped practices represent a rich
source of insight into how higher cognitive activities may be grounded in dynamic
patterns not only of bodily perception and movement, but also of social behavior.
Johnson (2005) strongly advocates the importance of putting flesh on image-schematic
skeletons and of trying to account for the felt qualities of meanings and situations (see
also Cienki 2005; Deane 2005; Gibbs 2005; Zlatev 2005). One of the central questions
still seems to be how multi-faceted meanings, especially in abstract reasoning, emerge
from embodied experience:

But let us not forget that the truly significant work done by image schemas is tied
to the fact that they are not merely skeletons or abstractions. They are recurring
patters of organism-environment interactions that exist in the felt qualities of our
experience, understanding, and thought. Image schemas are the sort of structures
that demarcate the basic contours of our experience as embodied creatures. […]
Their philosophical significance, in other words, lies in the way they bind together
body and mind, inner and outer, and thought and feeling. They are an essential part
of the embodied meaning and provide the basis for much of our abstract inference.
(Johnson 2005:31)

In view of Johnson’s (2005:31) exhortation to ‘analyze various additional strata of mean-

ing, such as the social and affective dimensions, to flesh out the full story of meaning and
thought’, it seems safe to say that bodily semiotics generally bear the potential to inform
us about qualities that are difficult to access via purely linguistic inquiry. Gesture data
remain a promising source to explore both structured and intuitive aspects of how we
make meaning and also of how we make sense of what others try to convey. In light of
these considerations, we can perhaps better appreciate the extent to which the present
gesture data bring out the dynamic and embodied aspects of image-schematic and
geometric representations of abstract objects and structures: gestures are not simply
visual, but visuo-motoric and a bodily medium; hence, they have the capacity to shed
376 LANGUAGE, COGNITION AND SPACE

additional light on the assumed multimodal character of concepts and image schemas
(cf. Evans and Green 2006). Contrary to static visual representations of words, sentences,
and diagrams captured on paper or blackboards, these gestures afford a ‘representation
of abstract processes as dynamic patterns’ (Kendon 1997:112) through a ‘dynamic visuo-
spatial imagery’ (McNeill et al. 2001:11). Linguistic form and structure seem to come
to life: branches branch out, words move or travel together to the front of a sentence,
and boundaries between concepts get blurred. Instrumental hand actions seemingly
manipulating items highlight the process character of operations such as prefixation,
suffixation, infixation, or the construction of a sentence. In addition, grammatical
operations such as ‘reiteration’ and ‘recursion’ were found to be represented by the
rotation of a single hand or by two hands revolving around each other, and a similar
motor schema was observed to signify the function of a morphological case or the idea
of active language use as opposed to the knowledge of grammar. In gesture research, the
‘bodily basis of meaning, imagination and reason’, the title of Johnson’s (1987) ground-
laying book, may be taken literally, thus trying to illuminate not only the relationship
between the gesturer’s body and the imaginary objects and forces it interacts with, but
also to explore how meanings are conveyed through minimal movements or forceful
hand actions.

4.2 Dynamic representations of objects in places: some preliminary

considerations on the ‘what’ and ‘where’ in gesture space

Being aware of the preliminary character of the following reflections, I would like to
draw together two central aspects that make co-speech gesture a promising source of
insights into the relationship between cognition, space, and language: its spontaneous,
unreflective character on the one hand and its tendency to reflect schematic imagery
and basic geometric forms on the other.
Due to the attention gesture draws to what I like to think of as the ‘ex-bodiment’
(Mittelberg 2006, 2008) of internalized imagery and experiences with the physical and
social world, and due to its propensity to directly portray spatial and sensory-motor
aspects of concepts and source domains of metaphorical mappings, gesture research
has yielded insights into our understanding of abstract knowledge domains (Calbris
2003; Cienki 1998, 2005; McNeill 1992; Müller 1998, 2004; Sweetser 1998, 2007; Núñez
2004; Taub 2001). Since gestures unfold in space, they are naturally apt at illuminating
spatial metaphor, not only regarding linguistic form and structure, but also regarding, for
instance, the spatial representation of moral concepts (Cienki 1998 a/b), mathematical
thought (McNeill 1992; Núñez 2004; Smith 2003), and concepts belonging to the domain
of speech communication (Sweetser 1998).
It is because of their unreflective character that gestural representations of abstract
phenomena can offer fresh insights into the metaphorical nature of the conceptual
system and, more generally, into less monitored aspects of cognition during com-
munication. Crucially, in the present data, metaphorical understandings of abstract
entities are frequently expressed in the gesture modality even if the accompanying
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 377

speech is non-metaphorical. The technical term ‘subcategory’ (shown in Figure 1) is

a good example of this kind of multimodal representation of abstract concepts: the
metaphorical understanding of a category in terms of a container or object is conveyed
only in the gesture modality, not in speech. Other examples would be technical terms
such as ‘noun’, ‘constituent’, ‘node’, ‘sentence’, and ‘morpheme’ or words or parts of
words such as ‘fell’, ‘teach-,‘ and ‘-er’. In contrast to carefully planned and executed
pictorial metaphors deployed in advertisements, cartoons, and paintings, spontaneous
metaphorical gestures may provide more intuitive renditions of mental imagery, created
locally and online (see Mittelberg and Waugh 2009 and Müller and Cienki 2009 on
multimodal metaphor).
Arguing in favor of a multimodal approach to spatial representations, Deane
(2005:245) discusses instances in which spatial prepositions evoke a ‘common-sense
geometry’; he asserts that ‘the same spatial relation may receive distinct representations
in multiple representational modalities’ (p. 247). In view of the configurations observed
in the present gesture data, it seems that the speakers do apply a sort of common-sense
geometry when ascribing basic shapes to linguistic entities (e.g., in the form of bounded
objects) and structures (e.g., in the form of lines and diagrams, the latter exploiting both
horizontal and vertical axes to spatially portray hierarchical relations).
A question that poses itself here concerns the degree to which the imaginary meta-
phorically construed objects are geometrically specified. Talmy (1983) suggested univer-
sal constraints as to how figure object and ground object are geometrically schematized
in locative expressions; he noted an asymmetry to the effect that the figure object tends
to be relatively shapeless and the ground object tends to be more precisely defined (cf.
Landau 1996:321ff.; Landau and Jackendoff 1993). Investigating how the visual-spatial
modality might condition descriptions of the relation between two objects, Emmorey
(1996:175–9) found the tendencies identified by Talmy to hold in ASL, where, in fact, ‘the
use of space to directly represent spatial relations stands in marked contrast to spoken
languages’ (p. 175). She also found that signers tend to express the ground first and then
the figure object, conceiving of the figure as a point with respect to a more complex
ground (p. 179). In the present data, this process was found in gestural descriptions
in which, for instance, a string of words (as in Figure 16) was first drawn in the air
and subsequently functioned as a sort of virtual reference structure in which the word
order of particular linguistic units was pointed out. The same is true in regard to tree
diagrams which, once they are sketched out in air, provide slots where elements such as
embedded clauses may be placed (cf. Mittelberg 2006). However, much more research
is needed to develop a better understanding of the mechanisms of what one could call,
with recourse to Landau and Jackendoff (1993), the ‘what’ and ‘where’ in gesture space.
Now, if we wanted to describe the relationship between objects and gestural articula-
tors in light of figure/ground relationships as well as the relative specification of objects
in terms of their geometry, we could, in a first approach, say the following: in cases where
an imaginary object (i.e., the figure) is sitting on a palm-up open hand (i.e., the ground),
it exhibits a less specific geometry than the hand itself (see Figures 2, 3, 12). In most of
these scenarios, details of size or shape are not provided for the figure object, except for
the fact that a single hand cannot hold a very large object. In gesture, space may carry
378 LANGUAGE, COGNITION AND SPACE

meaning in various ways, and, as we saw above, the different amounts of space between
hands or fingers may signify linguistic units of different degrees of complexity (e.g., a
morpheme in Figure 9 vs. a sentence in Figure 10). The object/box gestures (Figures 1,
10, 11) seem to be more strongly profiled in terms of their size and volume and might
thus qualify as geometrically idealized representations of objects, i.e. manifestations of
what Talmy (1983) referred to as the ‘flexible schematizing of objects’ (Landau 1996:319).
By contrast, different kinds of pointing gestures were found to simply assign a location,
but no shape, to grammatical categories (such as semantic roles; see Figure 13). Here,
we could conceive of the space in front of the speaker as the ground, this time rather
vaguely defined. One could argue that these objects do not receive much specification
because they signify imaginary abstract entities and that, since that which they stand
for is revealed in the concurrent speech, it might be sufficient to just point to their
existence and, if applicable, to their specific spatial arrangement. In fact, the gestures
here take care of the ‘where’ of the entities, which also entails their position with respect
to one another (e.g. the placement pronouns in phrasal word constructions, Figure 8,
or the insertion of an infix, Figure 17). This bimodal strategy is highly economic and
makes verbal paraphrases (i.e., prepositional phrases) unnecessary. While some of
these observations indicate the kind of asymmetry suggested by Talmy (1983), more
research is needed to correlate the geometry of objects and their relations in the gesture
modality with cognitive and discourse-pragmatic factors such as, for instance, attention,
perceptual saliency, information flow, pragmatic inferencing, and the exact cross-modal
encoding of spatial information.

5 Concluding remarks

Gesture assigns meaning to space. It employs hand shapes, movement, and space to
describe not only physical objects and their spatial relationships, but also spatial models
underlying abstract knowledge domains and other concepts that are difficult to represent
such as time, values or emotions. The gestures discussed in the present paper have,
as I hope to have shown, the capacity to unite phenomena that at first might appear
contrasting in one way or another, including the interrelation between form and motion,
spontaneity and systematicity, and the abstract and the concrete.
First, in the gesture modality form may become motion and motion may become
form (form is motion, cf. Lakoff and Turner 1989). Hands may dynamically represent
the form of an object by drawing its contours in the air (such as the wave-like movements
representing the notion ‘intonation contour’, see Figure 21); or the virtual trace left by
a manual motion may evoke a form (such as a virtual container in which items can be
subsequently placed, see Figure 10). A gestural sign may depict the formal essence of an
entity and/or its characteristic movement, both of which can be used independently of
the perception or presence of the object. In addition, gestures can portray the process
character of mental operations of which we often only see the final product, for example
an assembled word or sentence (e.g., infixation, see Figure 17).
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 379

Second, despite their spontaneous and unreflective dimensions, gestural representa-

tions have been shown to exhibit a considerable degree of systematicity regarding both
the form they take and the space they exploit. There is more and more converging
evidence that the factors motivating the structure of gestures of the abstract include
embodied image and motor schemas, conceptual metaphor and metonymy (Bouvet
2001; Cienki 1998, 2005; Cienki and Müller 2008; McNeill 1992, 2005; Mittelberg 2006;
Müller 1998, 2004b; Núñez 2004; Núñez and Sweetser 2006; Sweetser 1998, 2007; Taub
2001), as well as routine object-oriented actions and practices of social interaction
(Calbris 2003; Clark 2003; LeBaron and Streeck 2000; Kendon 2004; Müller 1998, 2004;
Streeck 2002; inter alia).
Third, metaphoric gestures mediate between the abstract and the concrete: while
being abstracted from physical objects and actions, they make abstract phenomena
tangible. By isolating the essential properties of the objects and actions they represent,
they provide insights into the abstractive capacities and embodied structures of the
human mind, and incarnate the principles of conceptual metaphor and abstract infer-
encing (Johnson 2005). In the meta-grammatical discourse analyzed here, linguistic
form and structure seem to propel manifestations of a set of image-schematic and
geometric patterns in the gesture modality. Embodied ‘common-sense geometry’
(Deane 2005:245) thus manifests itself in these gestures to a certain degree, and it
would be interesting to see whether such tendencies appear in gestures accompanying
discourses about other abstract subject matters (cf. Cienki 2005; Núñez 2004; Smith
2003; Sweetser 2007). Such work could further attest to the embodied nature of basic
image and motor schemas in general and spatial-relations concepts in particular
(Lakoff and Johnson 1999:34ff.; Hampe 2005; Talmy 1988). Another promising avenue
for further research would be to explore the pragmatics of the ‘flexible schematizing
of objects’ and the relative geometry of figure and ground objects in co-speech gesture
(Talmy 1983; Emmorey 1995; Landau 1996; Landau and Jackendoff 1993).
Theoretical, academic discourse might have the reputation of being dry, technical,
and objective; however, the multimodal classroom discourse examined here is strik-
ingly dynamic, immediate, and engaging. The professors’ gestures convey not only
visuo-spatial illustrations of grammatical concepts and theories, but also intuitive, felt
qualities of thought and meaning-making processes which no doubt deserve further
(cross-disciplinary) attention.

Acknowledgements
I am grateful to the editors and an anonymous reviewer as well as to Jana Bressem, Alan
Cienki, Jacques Coursil, Sotaro Kita, Silva Ladewig, Cornelia Müller, Michael Spivey,
Eve Sweetser, and Linda Waugh for stimulating discussions and insightful comments
on earlier versions of this chapter. I also thank Allegra Giovine, Joel Ossher, and Daniel
Sternberg for their valuable help with database design and data coding and Yoriko Dixon
for providing the artwork.
380 LANGUAGE, COGNITION AND SPACE

Notes
1 The approach to multimodal discourse developed in Mittelberg (2006) combines
Peircean semiotics (Peirce 1955), Jakobson’s theory of metaphor and metonymy
(Jakobson 1956), and contemporary cogntivist approaches to metaphor and metonymy
(see also Mittelberg and Waugh 2009).
2 Gesture researchers have suggested various schemes for how to graphically capture not
only the close temporal relationship between speech and co-speech gesture, but also
the kinetic features of gestures (cf. Calbris 1990; Duranti 1997:144–154; Kendon 2004;
McNeill 1992, 2005; Müller 1998:175–199, 284ff.; Parrill and Sweetser 2004; inter alia).
This study has particularly been inspired by the methods of transcription, coding, and
analysis developed by members of the McNeill Lab (McNeill 1992), Müller (1998, 2004a)
and Webb (1996).
3 Another possibility would have been to adopt the form inventory of a signed language
such as American Sign Language (c.f. McNeill 1992:86–88; Webb 1996).
4 I thank Allegra Giovine and Daniel Sternberg for their invaluable collaboration on this
part of the analysis.
5 Here a link can be made to abstraction in the visual arts. Georges Braque and Pablo
Picasso developed their Cubist transformations of people and everyday objects through
extracting their most essential characteristics (see Mittelberg 2006 and in prep.).

References
Boroditsky, L. (2001) Does language shape thought? Mandarin and English speakers’
conceptions of time. Cognitive Psychology 43: 1–22.
Bouvet, D. (2001) La dimension corporelle de la parole. Les marques posturo-mimo-
gestuelles de la parole, leurs aspects métonymiques et métaphoriques, et leur rôle au
cours d’un récit. Paris: Peeters.
Bowermann, M. (1996) Learning how to structure space for language: A cross-
linguistic perspective. In P. Bloom, M. A. Peterson, L. Nadel and M. F. Garrett
(eds) Language and Space 385–436. Cambridge, MA: MIT Press.
Bühler, K. (1934/1965) Sprachtheorie. Zur Darstellungsfunktion der Sprache. 2nd ed.
Stuttgart: Fischer.
Calbris, G. (1990) The semiotics of French gestures. Bloomington: Indiana University
Press.
Calbris, G. (2003) From cutting an object to a clear cut analysis: Gesture as the
representation of a preconceptual schema linking concrete actions to abstract
notions. Gesture 3(1): 19–46.
Cienki, A. (1998a) Metaphoric gestures and some of their relations to verbal meta-
phoric expressions. In J.-P. Koenig (ed.) Discourse and cognition: Bridging the gap
189–204. Stanford: CSLI Publications.
Cienki, A. (1998b) STRAIGHT: An image schema and its metaphorical extensions.
Cognitive Linguistics 9: 107–149.
Cienki, A. (2005) Image schemas and gesture. In B. Hampe (ed.) in cooperation
with J. Grady, From perception to meaning: Image schemas in cognitive linguistics
421–442. Berlin/New York: Mouton de Gruyter.
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 381

Cienki, A. and Müller, C. (eds) (2008) Metaphor and gesture. Amsterdam/

Philadelphia: John Benjamins.
Clark, E. V. (1973) Non-linguistic strategies and the acquisition of word meanings.
Cognition 2: 161–182.
Clark, H. H. (1973) Space, time, semantics, and the child. In T. E. Moore (ed.)
Cognitive development and the acquisition of language 27–63. New York:
Academic Press.
Clark, H. H. (1996) Using language. Cambridge: Cambridge University Press.
Clark, H. H. (2003) Pointing and placing. In S. Kita (ed.) Pointing: Where language,
culture, and cognition meet 243–268. Mahwah, NJ: Lawrence Erlbaum Associates.
Danaher, D. (1998) Peirce’s semiotic and cognitive metaphor theory. Semiotica
119(1/2): 171–207.
Deane, P. D. (2005) Multimodal spatial representations: On the semantic unity of
over. In B. Hampe (ed.) in cooperation with J. Grady, From perception to mean-
ing: Image schemas in cognitive linguistics 235–282. Berlin: Mouton de Gruyter.
DuBois, J., Schuetze-Coburn, S., Cumming, S. and Paolino, D. (1993) Outline of
discourse transcription. In J. A. Edwards and M. D. Lampert (eds) Talking data:
Transcription and coding in discourse research 45–87. Mahwah, NJ: Lawrence
Erlbaum Associates.
Duranti, A. (1997) Linguistic anthropology. Cambridge: Cambridge University Press.
Emmorey, K. and Reilly, J. (eds) (1995) Language, gesture, and space. Hillsdale, NJ:
Lawrence Erlbaum Associates.
Emmorey, K. (1996) The confluence of space and language in signed languages. In
P. Bloom, M. A. Peterson, L. Nadel and M. F. Garrett (eds) Language and space
171–209. Cambridge, MA: MIT Press.
Evans, V. and Green, M. (2006) Cognitive linguistics: An introduction. Edinburgh:
Edinburgh University Press.
Gibbs, R. W., Jr. (1994) The poetics of mind: Figurative thought, language, and under-
standing. Cambridge, UK: Cambridge University Press.
Gibbs, R. W., Jr. (2003) Embodied experience and linguistic meaning. Brain and
Language 84: 1–15.
Gibbs, R. W., Jr. (2005) The psychological status of image schemas. In B. Hampe
(ed.) in cooperation with J. Grady, From perception to meaning: Image schemas in
cognitive linguistics 113–135. Berlin/New York: Mouton de Gruyter.
Gibbs, R. W., Jr. (2006) Embodiment and cognitive science. New York: Cambridge
University Press.
Goldin-Meadow, S. (2003) Hearing gesture: How hands help us think. Cambridge,
MA/London: The Belknap Press of Harvard University Press.
Hampe, B. (ed.) (2005) From perception to meaning: Image schemas in cognitive
linguistics. Edited in cooperation with J. Grady. Berlin/New York: Mouton de
Gruyter.
Haviland, J. B. (1993) Anchoring, iconicity and orientation in Guugu Yimithirr
pointing gestures. Journal of Linguistic Anthropology 3: 3–45.
Haviland, J. B. (2000) Pointing, gesture spaces, and mental maps. In D. McNeill (ed.)
Language and gesture 13–46. Cambridge: Cambridge University Press.
382 LANGUAGE, COGNITION AND SPACE

Jakobson, R. (1956/1990) Two aspects of language and two types of aphasic distur-
bances. In L. R. Waugh and M. Monville-Burston (eds) Roman Jakobson, on
language 115–133. Cambridge/London: Harvard University Press.
Jakobson, R. (1961/1987) Poetry of grammar and grammar of poetry. In K.
Pomorska and S. Rudy (eds) Roman Jakobson, language in literature 121–144.
Cambridge/London: Belknap Press of Harvard University Press.
Jakobson, R. (1966/1990) Quests for the essence of language In L. R. Waugh and M.
Monville-Burston (eds) Roman Jakobson, on language 407–421. Cambridge/
London: Harvard University Press.
Johnson, M. (1987) The body in the mind: The bodily basis of meaning, imagination,
and reason. Chicago: University of Chicago Press.
Johnson, M. (2005) The philosophical significance of image schemas. In B. Hampe
(ed.) in cooperation with J. Grady, From perception to meaning: Image schemas in
cognitive linguistics 15–33. Berlin/New York: Mouton de Gruyter.
Kendon, A. (1997) Gesture. Annual Review of Anthropology 26: 109–128.
Kendon, A. (2000) Language and gesture: Unity or duality? In D. McNeill (ed.)
Language and gesture 47–63. Cambridge: Cambridge University Press.
Kendon, A. (2004) Gesture: Visible action as utterance. Cambridge: Cambridge
University Press.
Kita, S. (ed.) (2003) Pointing: Where language, culture, and cognition meet. Mahwah,
NJ: Lawrence Erlbaum Associates.
Lakoff, G. (1987) Women, fire, and dangerous things. What categories reveal about the
mind. Chicago: University of Chicago Press.
Lakoff, G. (1993) The contemporary theory of metaphor. In A. Ortony (ed.)
Metaphor and thought 202–251. 2nd ed. Cambridge: Cambridge University Press.
Lakoff, G. and Johnson, M. (1980) Metaphors we live by. Chicago: University of
Chicago Press.
Lakoff, G. and Johnson, M. (1999) Philosophy in the flesh: The embodied mind and its
challenge to Western thought. New York: Basic Books.
Landau, B. (1999) Multiple geometric representations of objects in languages and
language learners. In P. Bloom, M. A. Peterson, L. Nadel and M. F. Garrett (eds)
Language and space 316–363. Cambridge, MA: MIT Press.
Landau, B. and Jackendoff, R. (1993) ‘What’ and ‘Where’ in spatial language and
spatial cognition. Behavioral and Brain Sciences 16: 217–238.
LeBaron, C. and Streeck, J. (2000) Gestures, knowledge, and the world. In D. McNeill
(ed.) Language and gesture 118–138. Cambridge: Cambridge University Press.
Levinson, J. (1996) Frames of reference and Molyneux’s question: Crosslinguistic
evidence. In P. Bloom, M.A. Peterson, L. Nadel and M. F. Garrett (eds) Language
and space 109–169. Cambridge, MA: MIT Press.
Levinson, J. (1997) Language and cognition: The cognitive consequences of spatial
description in Guugu Yimithirr. Journal of Linguistic Anthropology 7 (1): 93–131.
Levinson, J. (2003) Space in language and cognition: Explorations in cognitive diversity.
Cambridge: Cambridge University Press.
Liddell, S. (2003) Grammar, gesture, and meaning in American Sign Language.
Cambridge: Cambridge University Press.
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 383

Mandler, J. (1996) Preverbal representation and language. In P. Bloom, M. A.

Peterson, L. Nadel and M. F. Garrett (eds) Language and space 365–384.
Cambridge, MA: MIT Press.
Mandler, J. (2004) The foundations of mind. Oxford/New York: Oxford University
Press.
McNeill, D. (1992) Hand and mind: What gestures reveal about thought. Chicago:
Chicago University Press.
McNeill, D. (ed.) (2000) Language and gesture. Cambridge: Cambridge University
Press.
McNeill, D. (2005) Gesture and thought. Chicago: Chicago University Press.
McNeill, D., Cassell, J. and Levy, E. T. (1993) Abstract deixis. Semiotica 95 (1/2):
5–19.
McNeill, D., Quek, F., McCullough, K.-E., Duncan, S.D., Furuyama, N., Bryll, R. and
Ansari, R. (2001) Catchments, prosody and discourse. Gesture 1(1): 9–33.
Mittelberg, I. (2002) The visual memory of grammar: Iconographical and metaphori-
cal insights. Metaphorik.de 2/2002: 69–89.
Mittelberg, I. (2006) Metaphor and metonymy in language and gesture: Discourse
evidence for multimodal models of grammar. Unpublished Doctoral
Dissertation, Cornell University.
Mittelberg, I. (2007) Methodology for multimodality: One way of working with
speech and gesture data. In M. Gonzalez-Marquez, I. Mittelberg, S. Coulson
and M. J. Spivey (eds) Methods in cognitive linguistics 225–248. Amsterdam/
Philadelphia: John Benjamins.
Mittelberg, I. (2008) Peircean semiotics meets conceptual metaphor: Iconic modes in
gestural representations of grammar. In A. Cienki and C. Müller (eds) Metaphor
and gesture 115–154. Amsterdam/New York: John Benjamins.
Mittelberg, I. (in prep.) Metonymy in gesture and Cubism: A comparative study of
abstraction, essence, and relativity in semiotic structure.
Mittelberg, I. and Waugh, L. R. (2009) Metonomy first, metaphor second: A cogni-
tive-semiotic approach to multimodal figures of thought in co-speech gesture. In
C. Forceville and E. Urios-Aparisi (eds) Multimodal metaphor 322–358. Berlin/
New York: Mouton de Gruyter.
Müller, C. (1998) Redebegleitende Gesten. Kulturgeschichte – Theorie –
Sprachvergleich. Berlin: Berlin Verlag A. Spitz.
Müller, C. (2004) Forms and uses of the palm up open hand: A case of a gesture
family? In C. Müller and R. Posner (eds) The semantics and pragmatics of every-
day gestures. The Berlin Conference 233–256. Berlin: Weidler Verlag.
Müller, C. (2008) Metaphors. Dead and alive, sleeping and waking. A cognitive
approach to metaphors in language use. Chicago: Chicago University Press.
Müller, C. and Cienki, A. (2009) Metaphor, gestures, and beyond: Forms of multimo-
dal metaphor in the use of spoken language. In C. Forceville and E. Urios-Aparisi
(eds) Multimodal metaphor 293–321. Berlin/New York: Mouton de Gruyter.
Núñez, R. E. (2004) Language, thought, and gesture: The embodied cognitive
foundations of mathematics. In F. Iida, R. Pfeifer, L. Steels and Y. Kuniyoshi (eds)
Embodied artificial intelligence 54–73. Berlin: Springer-Verlag.
384 LANGUAGE, COGNITION AND SPACE

Núñez, R. E. and Sweetser, E. E. (2006) Looking ahead to the past: Converging

evidence from Aymara language and gesture in the crosslinguistic comparison of
spatial construals of time. Cognitive Science.
Ortony, A. (1993) Metaphor and thought. 2nd edn. Cambridge: Cambridge University
Press.
Özyürek, A. (2000) The influence of addressee location on spatial language and
representational gestures of direction. In D. McNeill (ed.) Language and gesture
64–83. Cambridge, U.K.: Cambridge University Press.
Parrill, F. and Sweetser, E. E. (2004) What we mean by meaning: Conceptual integra-
tion in gesture analysis and transcription. Gesture 4(2): 197–219.
Peirce, C. S. (1955) Logic as semiotic: The theory of signs (1893–1920). In J. Bucher
(ed.) Philosophical writings of Peirce 98–119. New York: Dover.
Richardson, D. C., Spivey, M. J., McRae, K. and Barsalou, L. W. (2003) Spatial repre-
sentations activated during real-time comprehension of verbs. Cognitive Science
27: 767–780.
Slobin, D. (1996) From ‘thought and language’ to ‘thinking for speaking’. In J. J.
Gumperz and S. C. Levinson (eds) Rethinking linguistic relativity 97–114.
Cambridge: Cambridge University Press.
Slobin, D. (2000) Verbalized events: A dynamic approach to linguistic relativity and
determinism. In S. Niemeier and R. Dirven (eds) Evidence for linguistic relativity.
Amsterdam/New York: John Benjamins.
Smith, N. (2003) Gesture and beyond. Unpublished Undergraduate Honors Thesis,
Program in Cognitive Science, University of California at Berkeley.
Sweetser, E. E. (1990) From etymology to pragmatics: Metaphorical and cultural
aspects of semantic structure. Cambridge: Cambridge University Press.
Sweetser, E. E. (1992) English metaphors for language: Motivations, conventions, and
creativity. Poetics Today 13(4): 705–724.
Sweetser, E. E. (1998) Regular metaphoricity in gesture: Bodily-based models of
speech interaction. Actes du 16e Congrès International des Linguistes (CD-ROM),
Elsevier.
Sweetser, E. E. (2007) Looking at space to study mental spaces: Co-speech gesture
as a crucial data source in cognitive linguistics. In M. Gonzalez-Marquez, I.
Mittelberg, S. Coulson and M. Spivey (eds) Methods in Cognitive Linguistics
202–224. Amsterdam/New York: John Benjamins.
Streeck, J. (2002) A body and its gestures. Gesture 2(1): 19–44.
Talmy, L. (1983) How language structures space. In H. L. Pick and L. P. Acredolo
(eds) Spatial orientation: Theory, research, and application 225–282. New York:
Plenum Press.
Talmy, L. (1988) Force dynamics in language and cognition. Cognitive Science 12:
49–100.
Taub, S. (2001) Language from the body: Iconicity and metaphor in American Sign
Language. Cambridge: Cambridge University Press.
Waugh, L. R. (1982) Marked and unmarked – a choice between unequals in semiotic
structure. Semiotica 39(3/4): 299–318.
 GEOMETRIC AND IMAGE -SCHEMATIC PATTERNS IN GESTURE SPACE 385

Webb, R. (1996) Linguistic features of metaphoric gestures. Unpublished Doctoral

Dissertation, University of Rochester.
Wilcox, P. P. (2000) Metaphor in American Sign Language. Washington, DC:
Gallaudet University Press.
Wilcox, S. and Morford, J. (2007) Empirical methods in signed language research. In
M. Gonzalez-Marquez, I. Mittelberg, S. Coulson and M. J. Spivey (eds) Methods
in cognitive linguistics 171–200. Amsterdam/Philadelphia: John Benjamins.
Williams, W. (2004) Making meaning from a clock: Material artifacts and conceptual
blending in time-telling instruction. Unpublished PhD Dissertation, University
of California at San Diego.
Zlatev, J. (2005) What’s in a schema? Bodily mimesis and the grounding of language.
In B. Hampe (ed.) in cooperation with J. Grady, From perception to mean-
ing:  Image schemas in cognitive linguistics 313–342. Berlin/New York: Mouton de
Gruyter.
Part VII
Motion

387
15 Translocation, language and the
categorization of experience
Jordan Zlatev, Johan Blomberg and Caroline David

1 Introduction

The phenomenon of motion is prevalent in experience: the rising and falling of our
chests in breathing, the tapping of our feet against the floor, the flying of birds, the
ripples of water in the brook. Panta rei. But all instances of (perceived) motion are not
of the same kind. In the case of the rising chest, the tapping foot and the rippling water
we do not experience any change of location of the moving object. On the other hand,
in following by gaze the flight of birds, or perhaps a boat floating down the river, we
do experience such a change of location. At the same time, there is a difference in the
latter two cases: birds fly through perceived self-motion, while the boat is being moved
by the flow of the river, or possibly by people rowing it.
The goal of this chapter is twofold. The first is to provide an experientially-based
classification of perceived motion situations. We believe that the one we offer in Section 3
is more systematic than the various distinctions made in the current literature on ‘motion
events’ (e.g. Talmy 2000, Slobin 2003, Pourcel 2005, cf. Section 2). Notice also that by
emphasizing experience, rather than the objective fact of motion, we adopt a phenom-
enological perspective situating motion in the lifeworld of the human subject (Husserl
1999 [1907]), rather than in ‘objective reality’. This is consistent with the assumption,
often emphasized by cognitive linguists nowadays (e.g. Lakoff 1987), but with roots in
antiquity (cf. Itkonen 1991), that language refers to and classifies not reality in itself – but
reality as conceived by human beings. This brings us naturally to the second goal of the
chapter: to use the proposed taxonomy of motion situations in addressing the questions
of how different languages express motion, and if linguistic differences imply differences
in conceptualization. Such (neo-)Whorfian questions have been explored extensively
in the literature in recent years (see Pourcel 2005 and Section 4 below for a review), but
unless we can define the classes of motion experiences independently of language, we
are left without a compass in addressing the issues of linguistic relativity. Indeed, one
finds an acknowledgment of the need for a language-independent characterization of
experience in the writings of the father of the ‘principle of linguistic relativity’ himself,
Benjamin Lee Whorf:

To compare ways in which different languages differently ‘segment’ the same

situation of experience, it is desirable to analyze or ‘segment’ the experience first
in a way independent of any language or linguistic stock, a way which will be same
for all observers. (Whorf 1956: 162)

389
390 LANGUAGE, COGNITION AND SPACE

After reviewing some of the neo-Whorfian research on motion in Section 4, we ask in

Section 5 whether the different ways in which French, Swedish and Thai speakers express
motion situations imply conceptual and experiential differences in tasks involving the
categorization of translocation. Describing a series of experimental studies using the
Event Triads elicitation tool (Bohnemeyer, Eisenbeiss and Naranhimsan 2001), and an
extension of it (Blomberg 2006, 2007) we show that the answer to this question appears
to be not unambiguous. To anticipate, our empirical findings suggest that the categoriza-
tion of motion situations can be either more direct – and thus relatively unaffected by
language – or more mediated (Vygotsky 1978), and that language can play a considerable
role at least in the second case. As we discuss in Section 6, the change of emphasis from
linguistic relativity to linguistic mediation can help interpret not only our own results,
but also some of the contradictory findings reported in the recent literature.

2 Motion and ‘motion-event typology’

If an essential aspect of motion is the perception of physical instability (Durst-Andersen

1992: 53) then what exactly is a ‘motion event’, given that this has been the dominant
term in the relevant literature during the past decades? Talmy offers the following
answer: ‘A Motion event […] is a situation containing motion or the continuation of
stationary location.’ (Talmy 2000: 162, our emphasis). But whatever advantages this may
have in terms of capturing commonalities across static and dynamic locative predication,
it is much too general for our purposes by glossing over the major experiential division:
spatial change vs. stasis.
Talmy (1985, 2000) considers the ‘presence of motion’, or motion with a small
letter, along with the conceptual components figure, ground, path and manner/cause
to be building blocks of a ‘motion event’, and depending on the way they are mapped
to different constituents in the clause, formulates the basis for his well-known motion-
event typology, shown schematically in Figure 1, with example sentences from English
(a satellite-framed, or S-language) and French (a verb-framed, or V-language). This
typology has been claimed to be exhaustive, i.e. that every one of the world’s languages
can be categorized as being, predominantly, an S- or a V-language.

S-languages I swam across the river

(e.g. English)
motion manner path
co-event core-schema

V-languages
(e.g. French) J' ai traversé le fleuve (à la nage)
ͳ•‰ …”‘••Ȁ   ”‹˜‡” •™‹‹‰

Figure 1. Different mapping patterns between the conceptual components of motion events and
parts-of-speech in satellite-framed (S) languages and verb-framed (V) languages
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 391

However, it has become increasingly clear that this binary typology cannot do justice
to the complexity found in the world’s languages: either more ‘exotic’ ones such as
Tzeltal (cf. Brown 2004), or more familiar ones such as Russian (cf. Smith 2003), as
many of the contributions to the volume edited by Strömqvist and Verhoeven (2004),
e.g. Slobin (2004) testify. In some of our own work (Zlatev and David 2003, Zlatev and
Yangklang 2004), we have documented how Thai, and by extension other similar serial
verb languages, constitute a distinct ‘third’ type (which Slobin 2004 has generalized as
the ‘equipollently-framed type’, together with other languages which permit the easy
encoding of both Manner and Path in the same clause). For example, Thai resembles
V-languages in some respects (e.g. path expression by a main verb), S-languages in
other respects (e.g. manner expression by a main verb), while in yet other respects it
resembles neither (e.g. by having a separate ‘slot’ in the serial verb construction for
path+manner conflating verbs), cf. Zlatev and David (2003) for discussion.
But perhaps more troublesome for the Talmian typology than the empirical
problems are certain unresolved conceptual and definitional issues, such as the
following:

• What exactly is ‘path’? The extended trajectory traversed by the moving entity,
or some sort of schematic representation of this, e.g. as in the model of Regier
(1996), related to the beginning, middle and/or end of the motion trajectory?1
And how does this relate to the concept of direction of motion, expressed in
e.g. up?
• What exactly is ‘manner’ (of motion)? Does this include information pertaining
to the vehicle of motion (e.g. fly vs. ride), the speed (e.g. stroll vs. run), the body
parts (e.g. hop vs. climb), the medium (sink vs. fall) or all of these?
• Why is path regarded as the ‘core schema’, and is this so for all languages and for
all types of motion (for this and the following point, see the discussion below)?
• What is a ‘co-event’? Is it really an event and does it always pertain to informa-
tion related to the ‘manner’ or ‘cause’ of motion?
• What exactly is a ‘satellite’? Talmy (2000: 102) defines it as a constituent standing
in a ‘sister relation to the verb root’, but it is, for example, unclear if Swedish
verbal particles (e.g. gå in) can be thus grouped with Bulgarian verb-prefixes
(e.g. v-liza): while both examples correspond to English ‘go in’, and the ‘satellite’
carries the meaning INTERIOR, the Bulgarian stem does not exist as an inde-
pendent verb.

The basic, and yet unresolved, question however, remains ‘What is motion?’ and corre-
spondingly: ‘What is a motion event?’ Prior to a clear answer to these questions, it is not
certain that we are comparing equivalent semantic structures across languages. Talmy
is clearly aware that his initial definition of a ‘Motion event’ needs further specification,
since he repeatedly points out the difference between translational motion: ‘an object’s
basic location shifts from one point to another in space’ and self-contained motion, where
‘an object keeps its basic or ‘average’ location’ (Talmy 2000: 35) and emphasizes that
the typology concerns motion only of the first kind. However, it is not altogether clear
392 LANGUAGE, COGNITION AND SPACE

what this distinction amounts to and what is meant by ‘basic location’. As examples (1)
show, it is not possible to decide on the basis of the semantics of the verb alone what
type of motion is involved: in (1a) John’s motion is clearly ‘self-contained’ while in (1c)
John’s location has ‘shifted’ from outside to inside the room. But what about (1b): is the
motion involved considerable enough to be ‘translational’?

(1) (a) John ran on the treadmill.

(b) John ran in the park.
(c) John ran into the room.

In a recent monograph, Pourcel (2005) endeavors to clarify these issues through an ‘alter-
native model’, that is claimed to be based on conceptual analysis, rather than semantic
analysis, as is the case with Talmy, or discourse analysis as done by Slobin (e.g. 1996,
1997, 2003). The core of Pourcel’s proposal seems to be to distinguish between motion
events and motion activities, illustrating these with examples (2) and (3) – with identical
numbers in (Pourcel 2005: 153–154):

(2) The dog ran out of the barn across the field to the house.

(3) The dog is running around the house.

On this basis, it as argued that:

[t]here is therefore a distinction between motion that is source-and-goal-oriented,

as in (2), and motion that is not, as in (3). Conceptually, it is relevant to distinguish
between motion event and motion activity as the conceptual emphasis of an event
consists of the PATH of motion…; whereas the conceptual emphasis of an activity
consists of the MANNER of motion, which specifies a motion in progress, e.g. (3).
In other words, the core schema of activity is no longer PATH, but MANNER.
(Pourcel 2005: 154)

In general, this proposal is quite reasonable. But if indeed the ‘core schema’ in activity
representations is Manner rather than Path, this goes clearly against Talmy’s terminology,
where Path is always the core schema, irrespective of language and construction type,
which brings us back to one of the conceptual/definitional problems listed earlier. Still
more troublesome is that Pourcel (2005) does not provide any clear conceptual criterion
for what distinguishes ‘events’ from ‘activities’ that would explain the corresponding
focus on Path vs. Manner. The qualification ‘specifies a motion in progress’ for activities
can hardly be correct since it is based on the progressive aspect marking of (3), while
(1a) and (arguably) (1b) are representations of ‘activities’, even though they are not
presented as being ‘in progress’.
Furthermore, the concept of ‘motion event’ is extended by Pourcel (2005) to involve
not only ‘telic paths’, such as those on (2), but ‘atelic’ or ‘locative’ paths, ‘e.g. DOWN,
ALONG, AROUND’ (Pourcel 2005: 154), illustrated in the English example (4) and the
French examples (5) and (6):2
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 393

(4) The dog ran up the street.

(5) Marc monte les escaliers sur la pointe des pieds.

Marc goes up the stairs on tiptoes.

(6) Marc longe les bords de la rivièr.

Marc goes along the river bank.

What are the grounds for grouping these as examples of ‘events’ along with (2) rather
than as activities along with (3), which, note, even includes the so-called AROUND
path? We believe that the reasons are twofold. First, a language-independent conceptual
analysis is not provided, but rather one which is influenced by the ‘grammatical features
of motion event encoding in French’ such as that ‘PATH information is obligatory …
in the main verb’ (Pourcel 2005: 180), along with a priori classification of verbs such
as monter and longer as PATH verbs (albeit ‘atelic’). The second is that, as mentioned
earlier, Pourcel (2005) seems to conflate lexical (i.e. Aktionsarten) and morphological
(i.e. grammatical aspect) representations of the event/activity distinction – for example
in referring to ‘the variable use of the tenses, e.g. the imperfect or present tense for
activities, and the past perfect or simple past … for completed motion events’ (Pourcel
2005: 181, our emphasis). In the next section, we will propose our own conceptual
analysis of motion situations – a term used occasionally by Pourcel (2005: 186) as well,
as a hyperonym for motion events and activities – which we believe does not suffer from
these problems. At the same time, we wish to express our indebtedness to Pourcel (2005)
for helping bring together the ‘motion’ and the ‘situation type’ literatures, something
which has been long overdue.

3 A taxonomy of motion situations

From the perspective of the analysis of (the invariants of) experience – phenomenology
(cf. Husserl 1999 [1907]), motion as such can be defined as the experience of continuous
change in the relative position of an object (the figure) against a background, in contrast
to stasis – where there is no such change – and in contrast to a dis-continuous change, as
when a light suddenly lights up in position A, ‘disappears’ and then appears in position
B. As well-known, however, if the time fragment between the two discrete events is
small enough then an observer will actually see the light as moving from A to B, in a
continuous manner. Thus, motion is ‘in the eyes of the beholder’. Note that ‘continuous’
is here meant to exclude from the definition of motion such events as disappearing at
one place, and reappearing at another, as in a Star Trek case of ‘teleportation’, which may
be in the sphere of the imaginable, but not in the ordinary human lifeworld. It does not
exclude instances of rather abrupt types of motion, e.g. jumping, blinking, breaking or
other similar ‘punctual’ events.
Furthermore, note that motion ‘from A to B’, i.e. relocation (Smith 2003) is not a
necessary characteristic of a motion situation. First, the light could waver around A,
394 LANGUAGE, COGNITION AND SPACE

and then there would be no change in its average position and thus there would be
‘self-contained’ motion in Talmy’s terms. Second, the figure could be moving along a
vector in an open-ended way, for all eternity perhaps – and hence there need not be
any B to relocate to. Third, the figure’s motion can be either spontaneous or caused by
an external source. Thus, we have three different parameters according to which motion
situations can vary, quite independent of their representation in language. These are
described in the rest of this section, concluding with a summary presentation of the
taxonomy, and its (perceived) advantages compared with those of Talmy or Pourcel.

3.1 Translocative vs. Non-translocative motion

We define translocation, which is similar to but more transparent than Talmy’s term
‘translational motion’ (cf. Zlatev and Yangklang 2004) as the continuous change of an
object’s average position according to a spatial frame of reference. As can be seen from
this definition, this is a special kind of motion, which unlike motion in general requires
a spatial frame of reference (FoR):

In the most general sense, a FoR defines one or more reference points, and possibly
also a coordinate system of axes and angles. Depending on the types of the reference
points and coordinates different types of FoR can be defined. (Zlatev 2007: 328)

An influential treatment of the concept FoR, especially within linguistic typology, is

that of Levinson (1996, 2003), who distinguishes between relative, absolute and intrinsic
FoRs. However, this distinction is only based on horizontal static relations, whereas
Zlatev (2005, 2007) extends and generalizes it to involve dynamic relations, i.e. motion,
as well as the vertical plane. The first type can be called Viewpoint-centered, which when
expressed in language involves the perspective of the speaker or hearer as a reference
point, as in examples (7–8).

(7) I turned and went to the right. FoR: Viewpoint-centered, Speaker

(8) Turn and go to the/your right. FoR: Viewpoint-centered, Hearer

The second type is Geocentric, involving the horizontal or vertical plan while relying
on geo-cardinal positions as reference points, as in (9–10).

(9) I drove West. FoR: Geocentric, Horizontal

(10) The balloon went up. FoR: Geocentric, Vertical

Finally, there is the Object-centered FoR, which can involve the position of either the
focused (and possibly moving) object, the figure, or that of an external object, a land-
mark, as in (11–12).3
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 395

(11) I went forward. FoR: Object-centered, Figure

(12) I went to the church. FoR: Object-centered, Landmark

A particular case of translocation can thus be specified according to one or more of these
frames of reference, which provide the reference points allowing us (a) to judge that the
object/figure has indeed changed its average position and (b) to determine its Path or
Direction (see below). Similarly, in order to state that there is no change in the average
position of a moving figure, i.e. non-translocative motion, a FoR needs to be (at least)
presupposed. John’s running in example (1b) is non-translocative with respect to an
Object-centered FoR with the park as a whole as Landmark. But the same state-of-affairs
can be construed as translocative if we, for example, adopt some more specific reference
point, e.g. the viewpoint of an observer situated within the park.
On this basis, example (1c) can be classified as an expression of translocative motion,
while (1a) and (1b) represent non-translocative motion. The FoRs in all three cases are
object-centred, anchored in, respectively, the referents of ‘the room’, ‘the treadmill’, and
‘the park’. Note how essential the choice of a particular FoR is in order to determine the
type of motion. If the same external state-of-affairs described in (1b) was portrayed as
(13), then the (conceptualized) situation would be translocative, involving the change
of the figure’s position with respect to the ‘end of the park’.

(13) John ran to the end of the park and back.

Analogously, the same state-of-affairs can be experienced – and described – quite

differently, depending on the Frame of reference, as in the examples below.

(14) He is going to the top of the hill. Object-centered, Landmark

(15) He is going forward. Object-centered, Figure

(16) He is going uphill. Geocentric

(17) He is going that way. Viewpoint-centered

While all four examples involve translocation, (15–17) do not specify the change of
position in relation to a beginning (Source), middle (Via) or end (Goal) point, but
rather with respect to the figure’s initial position in (15), geo-centric coordinates in
(16) or a deictic center in (17). Thus following the analysis presented in earlier work
(Zlatev 2003, 2005), we state that of these examples only (14) involves the category Path,
understood in the schematic sense (cf. footnote 1), while (15–17) express the related but
different category Direction. In the case of non-translocative motion there is neither
Path nor Direction, since there is no change in the figure’s average position. The crucial
difference is that Path implies bounded motion, whereas Direction implies unbounded
motion, which brings us to the next parameter.
396 LANGUAGE, COGNITION AND SPACE

3.2 Bounded vs. unbounded motion

The boundedness of a process undergone by X implies that it will inevitably (not just
possibly or probably) lead to X undergoing a state-transition (cf. Vendler 1967). This
means that in expressions of bounded motion, X (the figure) will depart from Source,
or pass through a mid-point (Via), or reach a Goal (as in 12–14) – or all three as in
(2). In unbounded motion, nothing of the sort is implied, and in principle – though
not practically – the motion can go on indefinitely, as in the situations described in
examples (7–11). As pointed out above, bounded translocative motion always involves
the category Path, with one or more reference points being defined through the object-
centred, landmark-defined FoR. In the case of unbounded translocative motion, we have
rather the category Direction, specified either as a vector according to one of the other
FoR conditions, or as a trajectory, that can take particular shapes such as AROUND or
ALONG, as in (3) and (4).
Note furthermore, that there is independence between the two parameters discussed
so far. We have seen how translocative situations can be either unbounded, e.g. (7–11)
or bounded e.g. (12) and (13). Non-translocative situations can be similarly either
unbounded, as (1a) and (1b), or bounded – if the motion involved leads to a state-
transition, as in (18) or the Swedish equivalent (19).

(18) The vase broke (in pieces).

(19) Vas-en gick sönder.

vase-DEF go.PAST broken

One might counter that (18) and (19) do not express, but rather presuppose motion,
but since the ‘breaking’ of the vase will typically involve a perception of physical change
(against a stable background) we consider these sentences representations of non-
translocative bounded motion.

3.3 Self-motion vs. caused motion

The final parameter concerns whether the figure is perceived to be moving under the
influence of an external cause or not. As previously stated, the relevant notion of cau-
sality concerns the (naïve) human lifeworld, and not our scientific understanding of
the universe. Thus, the situation described in (20) above is one of ‘self-motion’ even
though the motion of the raindrops is caused by gravity, objectively speaking. On the
other hand, (21) clearly represents a (translocative, bounded) caused motion situation.

(20) Raindrops are falling on my head.

(21) John kicked the ball over the fence.

 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 397

This parameter is likewise independent of the other two, so it is possible to have, e.g.
caused translocative, unbounded motion situations (22), caused non-translocative
bounded ones (23), and caused non-translocative unbounded ones (24). The self-caused
correspondences to these have already been illustrated.

(22) He pushed the car forward.

(23) He tore the paper up.

(24) She waved the flag.

3.4 Summary

The independence of the three parameters yields the 8 types of motions situations
illustrated in Table 1, with schematic representations in English.
Table 1. Illustration of the expression of 8 motion situation types in English; F = Figure, LM = Landmark, A =
Agent, View-C = Viewpoint centred, Geo-C = Geocentric, Obj-C = Object centred Frame of Reference

-CAUSED +CAUSED
+TRANSLOCATIVE F goes to LM A throws F into LM
+BOUNDED

+TRANSLOCATIVE F goes away (View-C) A takes F away (View-C)

-BOUNDED F goes up (Geo-C) A pushes F upward (Geo-C)
F rolls forward (Obj-C) A pushes F forward (Obj-C)

-TRANSLOCATIVE F breaks (up/down) A breaks F (up/down)

+BOUNDED

-TRANSLOCATIVE F waves A waves F

-BOUNDED

The tense in the examples in Table 1, the present simple, is only seldom used with
any of these situation types (constructions) in English, and if so to express habitual
meanings, as in (25).

(25) Marry goes to school at 8 o’clock in the morning.

However, it was intentionally used in the cells in Table 1 in order to highlight the fact
that the different situation types (i.e. specifying the values of the three parameters) can
be expressed through: (a) the lexical semantics of the verb, (b) verb-satellite (particles
or affixes), (c) adpositional phrases and (d) the grammatical construction (e.g. intransi-
tive vs. transitive). While tense and aspect markers can make the distinction between
398 LANGUAGE, COGNITION AND SPACE

e.g. bounded and unbounded situations even clearer, i.e. by rendering the bounded
ones in past simple as in (21), and the unbounded ones in present continuous as in
(15–17), this is not necessary for making the parameter differentiations, at least for
English. In fact, we broadly agree with Durst-Andersen (1992) that morphological
aspect introduces an extra dimension of meaning over and above those expressed
by (a)–(d), by allowing the profiling of situations either as ongoing or as completed
– whether they are inherently bounded or not. Thus, (20) is no less a representation
of a bounded situation (despite ongoing), and (22) no less a representation of an
unbounded one (despite being ‘in the past’ and thus completed). The following three
‘authentic’ examples, taken from the British National Corpus (http://www.natcorp.
ox.ac.uk/), show how fall in the past tense can be used to express unbounded transloca-
tion, despite the fact that the events are being represented as taking place in the past,
and thus as ‘completed’. Grammatical tense-aspect should therefore be distinguished
from motion situation types, and their linguistic expression, pace Pourcel (2005).

(26) She called to Hermione and Joanna and all the girls who had gone already along the paths
she had rejected, called to them to wait for her and place their steady walking boots on solid
earth to catch her. And still she fell and fell.

(27) The wind blew and the snow fell, but it didn’t matter.

(28) … the devaluation of stock as component prices fell.

The conceptual framework described in this section and in particular the contrast
between bounded and unbounded translocative situations is highly relevant for our
empirical studies involving language and translocation described in Section 5. But prior
to describing these, let us first take stock.
We claim that our proposed taxonomy clarifies some of the problematic issues
described earlier. First of all, we believe that we have introduced definitions of (per-
ceived) motion in general, and specific types of motion situations that are more
consistent than those used in (much of) the ‘motion events’ literature. Second, we
consider our taxonomy to be, if not exhaustive, at least better equipped than alter-
natives to serve as a basis for typological investigations in the ‘domain’ of motion.
It allows us to analyse e.g. cases such as those that were discussed in Section 2 in
an unambiguous way. Thus, examples (4)–(6) can be classified as expressions of
translocative unbounded motion situations, together with (3), while (1c) and (2)
are representations of translocative bounded ones. On the other hand, (1a) and (1b)
are neither, but rather expressions of non-translocative unbounded motion. This is
summarized in Table 2. As pointed out in Section 2, examples such as these have
been grouped and termed in various ways in the past.
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 399

Table 2. A classification of the examples discussed in Section 2, on the basis of the presented taxonomy of motion
situations

Examples from Section 2 Motion situation type

(3) The dog is running around the house. + translocative
(4) The dog ran up the street. - bounded
(5) Marc monte les escaliers sur la pointe des pieds.
(6) Marc longe les bords de la rivière.

(1c) John ran into the room. + translocative

(2) The dog ran out of the barn across the field to the house. + bounded

(1a) John ran on the treadmill. - translocative

(1b) John ran in the park. - bounded

Third, we have defined Path as always related to Source, Via or Goal (on the basis of an
Object-centered, Landmark-defined FoR), while unbounded translocative situations
involve Direction, and non-translocative situations involve Location. In this way, we
have sharpened the conceptual apparatus used in the field. One thing this allows us
is to reformulate the famous boundary-crossing constraint (Slobin and Hoiting 1994),
stating that a motion verb expressing manner may not be used if there is a crossing
of a boundary, as follows: a Manner-verb can co-occur with an expression of Direction
or Location, but not with Path in the same clause. Assuming that French, as most
V-languages, generally obeys this constraint, examples (29–32), where the first two are
from Pourcel (2005: 40a-41a), and the latter two from Zlatev and David (2003: 40b-c)
can be straightforwardly explained as follows.

(29) Nous avons marché le long de la plage.

We walked along the beach
MANNER DIR
‘We walked along the beach.’

(30) Nous avons marché dans la pièce.

We walked in the room
MANNER LOC/*PATH
‘We walked inside the room.’
* ‘We walked into the room.’

(31) *Il a couru en entrant dans la maison.

3sg+MASC run+PAST entering in DEF house
MANNER PATH LOC
‘He ran entering the house.’

(32) Il a couru pour entrer dans la maison.

3sg+MASC run+PAST to enter in DEF house
MANNER PATH
‘He ran in order to enter the house.’
400 LANGUAGE, COGNITION AND SPACE

Example (29) does not violate the constraint, since it includes a combination of Manner
and Direction within the clause. In (30), only a non-translocative interpretation of
walking about ‘inside’ the house is possible. Such an interpretation is excluded in (31)
due to the participle en entrant, expressing Path, and the result is ungrammaticality
(uncorrectness), due to semantic factors. Finally, (32) is in contrast a correct French
sentence, since the Manner and Path expressions are in seperate clauses.
The reformulated boundary crossing constraint will play a role in the interpretation
of the results from our experiments, described in Section 5. But prior to that, we briefly
review some of the recent research on how different languages can possibly affect the
experience of motion in a way that ‘colours’ it accordingly.

4 Neo-Whorfian research on the categorization of translocation

If Talmy made ‘motion events’, or as we prefer – translocative situations – into a popular

subject for typology, it was Slobin (1996) who brought the subject to the attention of
neo-Whorfian research on linguistic relativity. According to one of Slobin’s formulations,
it may even be a mistake to look for language-independent taxonomies of situations
such as that presented in the previous section, since:

The world does not present ‘events’ and ‘situations’ to be encoded in language. Rather,
experiences are filtered through language into verbalized events. A ‘verbalized event’
is constructed online, in the process of speaking. (Slobin 1996: 75)

But at the same time, Slobin’s famous ‘dynamic’ formulation of the Whorfian program,
known as thinking for speaking, only concerns the ‘special kind of thinking […] that
is carried out, on-line, in the process of speaking’ (Slobin 1996: 75) and is therefore
different from Whorf ’s (1956) notion of ‘habitual thought’, according to which language
should have much more pervasive effects (cf. Blomberg 2007). Methodologically, Slobin
(1996, 1997, 2003) concentrated on differences in the ‘rhetorical style’ of speakers of
V-languages such as Spanish and S-languages such as English – as something that could
be explained by the languages’ different ways of expressing, above all, the concepts Path
and Manner. For example, due to the optional expression of Manner in V-languages (see
Figure 1), their speakers were found to express Manner less often and preferred to give
more static descriptions in which the Figure’s motion could be inferred from the ‘scene
setting’ and the result of the motion, while S-languages induced descriptions in which
the events were presented more dynamically, with more elaborated representations of
the Path. But, as pointed out by Pourcel (2005), Slobin’s research gives little support
for strong relativistic effects in the categorization of experience as such, i.e. even when
‘thinking for speaking’ is (apparently) not involved.
A number of other studies have attempted to demonstrate such effects using,
among other methods, a classic task for studying categorization in an (apparently)
non-linguistic context: forced-choice similarity judgments. The general method, used
with various modifications, in all of these studies is to use triads of representations of
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 401

motion situations: a target situation is presented along with two alternatives, where one
differs from the target with respect to Path and the other with respect to Manner, and the
subject is asked which of the two ‘is most similar’ to the target. The general reasoning
is that if language impinges on categorization, then speakers of a V-language should
be predisposed to prefer ‘same-Path’ rather than ‘same-Manner’ to a greater extent
than speakers of S-languages, where both components are expressed equally easy (see
Section 2). An exception to this line of reasoning was offered by Papafragou, Masely and
Gleitman (2002), who suggested an alternative basis for a linguistic effect that actually
runs in the opposite direction: since Manner is often expressed in a non-obligatory
constituent in a V-language, when it is expressed, it would be ‘foregrounded’ and thus
achieve more semantic salience (Talmy 1985) than in an S-language where it is expressed
by an obligatory constituent, such as the main verb. Papafragou, Masely and Gleitman
(2002) compared among other things the categorization of triads (using static pictures)
by speakers of Greek (assumed to be a V-language) and English (an S-language) and
despite differences in the linguistic descriptions that followed the predicted patterns
(along the lines of Slobin’s research), they found no bias for either Path or Manner-based
judgments in either group, and thus argued against the presence of any Whorfian effect
on motion event categorization.
However, other studies applying the same method, but using triads of dynamic
(video-clip) representations have given different results. Finkbeiner, Greth, Nicol
and Nakamura (2002) compared English (S-language) with Spanish and Japanese
(V-languages) speakers’ performance, and found a considerably stronger preference
for Manner-based similarity in the English group, and thus support for a degree of
linguistic relativity. Importantly, this effect was present only when the target clip was
presented first, and the alternatives (in parallel) afterwards. When the three clips were
presented simultaneously, the Manner-bias for the English group disappeared, leading
the authors to conclude that ‘the apparently nonlinguistic task used in Experiment 1
actually encouraged the participants to encode the scenes linguistically’ (Finkbeiner
et al: 454).
Gennari, Sloman, Malt and Fitch (2002) compared speakers of the two prototypical
languages for Talmy’s two types, English and Spanish, and established no clear difference
between the groups when the represented situations were not described prior to the
similarity judgments. But when they asked the subjects to provide such a description in
their native tongues prior to their choice, a stronger preference for Path in the Spanish
group was observed. This could be taken as offering support for a version of Slobin’s
thinking-for-speaking.
Pourcel (2005) reports evidence for an effect of language-type in a memory-
based study, but in her categorization study with 15 triads in the form of video-clips
representing people involved in various motion situations, she failed to find any
difference between English and French speakers. Both without and with prior linguistic
description there was a preference for same-Path categorization for both language
groups. An interesting finding, however, was that two types of motion situations,
corresponding to our distinction between bounded and unbounded translocation
described in Section 3 gave different results: there was a strong Path bias for bounded
402 LANGUAGE, COGNITION AND SPACE

motion (‘telic Path’), but this bias was neutralized, and with linguistic description
even replaced with a Manner-bias for the unbounded motion situations (‘atelic Path’)
(cf. Pourcel 2005: 243–245). Finally, an important difference compared to the study
of Finkbeiner et al. (2002) was that all three video-clips in each triad were presented
sequentially (in different orders).
Bohnemayer, Eisenbeiss and Narasimhan (ms), conducted the most extensive study
of this type, in the sense that they contrasted not just two or three languages, but 17
typologically, areally and genetically diverse languages, including Polish (S-framed
with verb-prefixes), German (S-framed with verb-particles), Japanese (V-framed)
and Lao (serial-verb, ‘third type’). The stimuli used by Bohnemayer, Eisenbeiss and
Narasimhan (ms) are identical to those used in our studies described in Section 5,
where we describe them in more detail, but suffice it for now to point out that they
involve an animated, smiling tomato-like figure which ‘jumps’, ‘rolls’, ‘spins’ or ‘slides’
either up/down a ramp, or left/right across a field, either with or without crossing
the boundaries of the Ground objects. While Pourcel (2005) criticizes the animated
‘unnatural’ character of the protagonist, and the fact that it allows a limited scope of
Manners of motion, we would argue that this design – similar to that of Finkbeiner et
al. (2002) – has a considerable advantage: it contrasts Manner and Path (in some cases:
Direction) completely systematically, so that the two choice situations are identical
with the target in each triad, apart from the manipulated variable. Furthermore, given
that even illiterate speakers of languages such as Jukatek living in traditional societies
did not have difficulties interpreting the situations with the ‘animate tomato’ capable
of self-motion suggest that it was not so ‘unnatural’.4
The foremost strength of the study of Bohnemeyer et al. (ms), however, is the large
number and variety of the languages involved. Accordingly, the results showed a wide
variation in the produced biases in the similarity judgment task: from 85% same-Manner
for the Polish group to 43% same-Manner for the Jalonke and Jukatek groups, but no
general pattern for speakers of S-languages preferring Manner more than those of
V-languages. This rather convincingly shows that the binary ‘motion-event typology’
of Talmy is not sufficient to predict categorization preferences (though it may be one
of the factors that play a significant role) and a better conceptual and methodologi-
cal basis is necessary in matching motion (i.e. translocation) typology and linguistic
relativity. Interestingly, Bohnemeyer et al. (ms) also established a language-general
difference between the representations of situations in which the figure moved up or
down (diagonally) on a ramp from the cases when in moved either from-to, or out of-
into a landmark: in the latter case the subjects were more likely to base their similarity
judgment on the basis of Manner than in the first. The authors attempt to explain this
in terms of the greater ‘simplicity’ of the ramp scenes, involving one reference object
(the ramp), rather than two.
But another explanation is possible: in the case of the ‘ramp’ scenario, the situ-
ation was at least ambiguous between unbounded translocation (moving upward or
downward) and bounded translocation (moving to ‘the top’ or ‘the bottom’ of the ramp).
On the other hand, the other two types of situations involved unambiguously bounded
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 403

translocative events, with or without boundary crossing. Thus, as in the study of Pourcel
(2005) the similarity judgments for the bounded and unbounded translocative situations
differed, implying the cognitive relevance of this distinction. However, the biases in
the two studies were converse: stronger preferences for same-Path categorization for
unbounded than bounded situations in the Bohnemeyer et al. (ms) study and stronger
preferences for same-Path categorization for bounded than unbounded motion in that
of Pourcel (2005). The conclusion is therefore that this factor must interact with other
‘variables’ such as the nature of the stimuli (animated vs. non-animate) and/or the nature
of the presentation of the alternatives (sequential vs. parallel). It is possible furthermore
that these factors affect the degree to which language influences the categorization
process.
In sum, the studies of the categorization of motion (translocation) situations by
speakers of different languages over the past few years have yielded different and some-
what contradictory results. What has become clear though is that:

a) the nature of the stimuli – static vs. motion pictures, animated vs. ‘real life’
video-clips, sequential vs. parallel presentation – influences the similarity
judgments;

b) different types of motion situations can yield different categorization prefer-

ences;

c) the role of linguistic description, especially prior to making the similarity

judgment, needs to be more carefully explored;

d) more languages than simply two representatives of the binary typology need
to be taken into consideration.

Our empirical studies using the Event Triads tool of Bohnemeyer et al. (ms) (Section
5.1 and 5.2) and a modification of it (Section 5.3) with speakers of Swedish, French and
Thai address the latter three points. In Section 6, we will offer an interpretation of the
apparently contradictory results, suggesting a coherent explanation.

5 Three empirical studies with event triads

5.1 Study 1

In our initial study we used the original Event Triads elicitation tool, developed at
the Max Planck Institute for Psycholinguistics, Nijmegen (Bohnemeyer, Eisenbeiss
and Narasimhan 2001), which was created to investigate biases for Path or Manner in
forced-choice similarity judgments. First a 5-second long animated film of the moving
tomato-like figure is shown on the whole computer screen, and after one second two
clips – identical to the first but differing with respect to either Path/Direction or Manner
404 LANGUAGE, COGNITION AND SPACE

– are shown in smaller windows in parallel (see Figure 2). The tool includes 72 such
different triads, ‘distributed across 6 randomized presentation lists in a Latin-square
design’ (Bohnemeyer, Eisenbeiss and Narasimhan, ms), where each list was presented
to two participants, in reverse order.

Figure 2. An example triad from the stimulus tool Event Triads. The black outline of the tomato-figure
is added, so that it would be more clearly visible when viewed in a black and white printout. In the
elicitation tool the red color of the tomato contrasts clearly with that (green or white) of the back-
ground and no such outlining is necessary

Thus, the Event Triads tool requires 12 participants for varying the order of presen-
tation, for counterbalancing the left/right position of the Manner-similar and Path/
Direction-similar smaller films in the second segment of the triad, and for trying all
possible combinations of Path/Direction and Manner. Following three practice trials,
each participant was given 50 triads. Of these, only 12 contrasted Path and Manner,
while the other 38 were distracters in which the figure stops at mid-scene, or involve
differences in color, or completely different situations such as one figure throwing an
object to another. The 12 crucial trials can be divided in 3 groups, depending on the
type of motion situation represented in the first segment (large window in Figure 2),
using the terminology introduced in Section 3:

• 4 Bounded translocative situations, from landmark1 to landmark2 (FROM/TO

Path)
• 4 Bounded translocative situations, out of landmark1 into landmark2 (OUT/
INTO Path)
• 4 Unbounded translocative situations, up (or down) (VERTICAL Direction)

As pointed out, in each of these cases the second segment presents a choice between a
situation in which the figure moves according to the same Path or Direction, but differs
in Manner, or has the same Manner, but moves in the reverse Path or Direction. There are
four different types of Manner that can be glossed in English as jumping, rolling, spinning
or sliding. As mentioned in Section 4, these manners of motion are quite perceptually
salient and conspicuous (especially for a ‘tomato’) and it was expected that there would
be a relatively strong Manner bias for the similarity judgments irrespective of language.
Nevertheless, one could expect this bias to be strongest (everything else being equal) for
speakers of S-languages, and weaker for speakers of a V-language (i.e. relatively more
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 405

Path-based choices). As for speakers of serial-verb languages such as Thai, we expected

these to show an intermediary position, given that both Manner and Path are easily
codable, or alternatively equally ‘backgrounded’, in such a language (cf. Section 2).
Participants were 3 groups of 12 monolingual undergraduate students from Lund
University (Swedish group), the University of Poitiers (French group) and Chulalongkorn
University (Thai group). The procedure was the following: each participant was given
three practice trials, followed by the 50 test triads. For the similarity judgment task,
after every triad, the participant had to point to either the left or the right half of
the second segment (cf. Figure 2) which was to serve as the answer to the question
‘Which is most similar to the first film – the left or the right?’ Following this and a
brief pause, there was a verbal description task, in which the participant was asked
to describe 18 video-clips of only the first fragment, representing the three kinds of
translocative situations in the data: 4 Vertical, 4 FROM/TO and 10 OUT/INTO.5 The
results of the similarity judgments task were marked in a coding sheet, and the verbal
description were recorded and transcribed, and both were subsequently subjected to
statistical analysis.
The results for the similarity judgements are presented in Figures 3 and 4. Contrary
to our expectations, it was not the Swedish, but the Thai group that had the largest
proportion of same-Manner choices: the difference between the Thai group on the
one side, and the French and Swedish groups on the other was statistically significant,
chi2(2) = 14.415 (p < .05), while that between the French and the Swedish groups
was not.




















    

Figure 3. Distribution of Manner vs. Path/Direction biased categorization choices for the three
language groups of French, Swedish and Thai. Max = 144 (12 participants * 12 choices) per
language

More interesting, however, were the results when we divided the 12 test triads according
to the three types listed above: FROM/TO, OUT/INTO and VERTICAL. As can be seen
in Figure 4, the classification of the Vertical unbounded translocative situations for the
French group differed significantly from the other two types of situation (chi2(2) =
406 LANGUAGE, COGNITION AND SPACE

6.933, p = 0.031), while there where no such differences for the other two languages.
Given that the total number of choices of this type was 48, the French group actually
displayed a weak Path bias (25 vs. 23) for this type.

Same-Path choices

30

25

20 FROM/TO
15 OUT-OF/INTO
10 VERTICAL

0
French Thai Swedish

Figure 4. Same-Path based choices for the three language groups, divided by situation types:
FROM/TO Path, OUT-OF/INTO Path, and VERTICAL Direction. Max = 48 (12 subjects * 4 choices) per
language

To help interpret this, we analyzed the results of the linguistic description task for the
French group in detail. We asked if there is a correlation between the differences in
the group’s similarity judgments (between the Vertical and the other two types) and
the semantic and grammatical structure of the descriptions of the group. In analyzing
the latter, we had a mini-corpus of 216 descriptions (12 participants * 18 transloca-
tive stimuli). We found indications for two such correlations. Table 3 displays all the
verbs (types and tokens) in the French descriptions, divided by the categories Vertical
Direction, Horizontal Path (FORM/TO + OUT/INTO), Manner and Other. The absolute
number of tokens were actually mostly Manner verbs, which may appear at first hand
surprising, given that French is (supposedly) a V-language, but as Pourcel (2005) and
Pourcel and Kopecka (ms) show, French involves several types of constructions were
Manner is expressed by the main verb (see also below). More relevant for our purposes,
however, was the fact that the Direction verbs, above all monter and descendre were
relatively more frequent than the Path verbs: there were only 4 stimuli (per subject) with
situations that could be described with these, whereas there were 14 stimuli for the Path
verbs (10 INTO and 4 TO). The ratio 8.75 vs. 5.36 in favor of Direction verbs compared
to the Path verbs suggests that Direction was more readily codable than Path, and thus
possibly also attracted relatively more attention than Path, compared to Manner in the
similarity judgment task. But admittedly this is only a tentative suggestion, and it says
nothing about the direction of (possible) causation involved: it is equally possible that
Direction is more easily cognitively ‘processable’ than Path, and therefore received a
higher degree of linguistic coding.
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 407

Table 3. Motion verbs produced by the French group in Study 1, in response to the linguistic task involving 4 Direc-
tion and 14 Path (4 FROM/TO and 10 OUT/INTO) stimuli

DIRECTION PATH MANNER OTHER

monter (ascend): 18 sortir (exit, go out): 23 rouler (roll): 54 aller (go): 81

descendre (descend): 15 rentrer (enter/come pivoter (pivot, revolve): 1 faire un déplacement
gravir (climb, struggle back home): 23 faire des galipettes (make a move): 1
up a slope):1 partir (leave): 14 (somersault): 1 se déplacer (move): 7
dévaler (tumble traverser (cross): 6 tourner (turn, spin): 14 s’arrêter (stop): 8
down): 1 passer (pass, go faire la toupie (move
through): 2 like a spinning top): 2
avancer (move glisser (slide): 14
forward): 2 sautiller (hop, skip): 11
arriver (arrive): 1 sauter (jump, leap): 3
retourner (go back): 2 faire des bonds (leap,
revenir (come back): 2 spring up): 1
bondir (jump, bounce): 1

Stimuli: 4 Stimuli: 14 Stimuli: 18 Stimuli: 18

Verb tokens: 35 Verb tokens: 75 Verb tokens: 102 Verb tokens: 97
Ratio: 8.75 Ratio: 5.36 Ratio: 5.67 Ratio: 5.39

The second correlation could more easily be related to a potential linguistic effect. It
turned out on analysis that in the verbalization of the bounded translocative (Path)
stimuli, only 18 out of 43 Manner expressions were present in the same clause as the
Path verb, while the remaining 25 (58%) occurred in an additional clause. On the other
hand, in the descriptions of the unbounded translocative (Direction) stimuli, in 27 out
of the 28 cases which also included an expression of Manner, the latter was expressed
in the same clause, as in (33). In only one case out of 28 (3.5%) was Manner expressed
in an additional clause.

(33) La tomate monte la montagne en roulant

DEF tomato climb DEF mountain rolling
DIRECTION MANNER

What this could be attributed to is the difficulty of encoding both Path and Manner in
the same clause, as opposed to Direction and Manner, due to the boundary-crossing
constraint (cf. Section 3.4) This would lead to Manner being expressed separately in the
case of bounded translocation, as the main verb of a separate clause, and thus making
it more semantically salient, somewhat along the lines suggested by Papafragou et al.
(2002), mentioned in Section 4, though not in comparison to other languages, but in
comparison to other types of motion situations within the same language.6 The reason-
ing is thus somewhat paradoxical, and called for a further study in order to see if this
correlation and possible explanation could be further supported.
408 LANGUAGE, COGNITION AND SPACE

5.2 Study 2

In this study we replicated Study 1, but using only 12 French speakers, this time of
different ages (24 to 60), and professional/educational backgrounds. The linguistic
descriptions were subjected to more thorough analysis. Since the Swedish group in
Study 1 did not display a bounded/unbounded translocation asymmetry, this study
was not designed as a comparative one.
The results from the similarity judgment task followed the same pattern as in Study
1: a general (though somewhat reduced) Manner bias but a reversal in the case of the
Vertical Direction motion situation: 27 vs. 21 same-Direction choices. Furthermore,
in dividing the Vertical stimuli in two groups depending on the direction of motion
(24 each), it turned out that while in the case of UPWARD motion the ratio between
same-Manner and same-Direction was even, in the case of DOWNWARD motion, there
was a strong preference for same-Direction over same-Manner (15 vs. 9).
The verbal descriptions were this time analyzed differently. Each description
was attributed to one of 5 different types: (i) Path/Direction+Manner in the same
clause, (ii) Path/Direction and Manner in different clauses, (iii) Path/Direction only,
(iv) Manner only and (v) Other, and each one of these was crossed with the four
situation types (OUT/INTO, FROM/TO, Vertical-UP and Vertical-DOWN) – due
to the differences in the similarity judgment task between the latter two, we decided
to treat them separately. The results, displayed in Table 4, showed striking differences
between the situation types. Whereas the most common type of verbal description for
the bounded translocative stimuli, and especially FROM-TO, was that of Manner only,
that for the unbounded translocative ones, and especially Vertical-DOWN was that of
Direction+Manner in the same clause (highlighted in Table 4). Furthermore, taking
together the rightmost two columns in Table 4, we can see that in the large majority
of cases of FROM/TO (81,3%) Path was not expressed at all, and similarly for half of
the OUT/INTO stimuli (49,2%). On the other hand, only a small minority of Vertical
stimuli (16,7% and 20,8%) lacked an expression of Direction. No such conspicuous
imbalance could be observed in the descriptions lacking Manner (the third and the
fifth columns taken together).
Thus, we find a strong correlation between the French speakers’ similarity judg-
ments – the same-Path bias for the Vertical stimuli – and their linguistic descriptions:
more frequent Path/Direction expression, particularly in the same clause. Admittedly
this is again only a correlation, and given that the descriptions were produced after
the similarity judgment task, this could not be a matter of any (direct) causation.
Nevertheless, the correlation was so obvious that it seems paramount to search for
an explanation.
One suggests itself once we realize that the ‘Path’ in the Vertical stimuli was rather
Direction, and that the stimuli represented situations that were more readily interpreted
as unbounded, rather than bounded. According to our redefinition of the ‘boundary-
crossing constraint’ (Section 3.4) it is above all the boundedness of the situation that
makes it difficult to express Manner and Path in the same clause in a V-language, while
there is no such difficulty with respect to Manner and Direction. Thus, given that
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 409

Table 4. Classifying the data from the verbal description task in Study 2: 4 types of motion situations and 5 expres-
sion patterns, with the highest proportions highlighted

Situation \ Path/ Path/ Path/ Manner only Other

Expression Direction Direction & Direction
+Manner Manner (diff. only
(same clause) clauses)

FROM/TO 1 6 2 31 8
(Tot: 48) (2,1%) (12,5%) (4,2%) (64,6%) (16,7%)

OUT/INTO 15 35 11 41 18
(Tot: 120) (12,5%) (29,2%) (9,2%) (34,2%) (15%)

VERT-UP 11 3 6 4 0
(Tot: 24) (45,8%) (12,5%) (25%) (16,7%)

VERT-DOWN 14 3 2 5 0
(Tot: 24) (58,3%) (12,5%) (8,3%) (20,8%)

Manner is perceptually salient – which we know independently to be the case for the
Event Triads stimulus tool – it is more likely to be expressed linguistically in a separate
clause (row 2), or alone (row 4) in verbalizing bounded than unbounded translocative
situations. Furthermore, this could increase the semantic salience of Manner, compared
to the cases where it is ‘conflated’ in the same clause with Direction and thus lead to a
stronger same-Manner bias.
Notice also in Table 4 that Manner was most often co-expressed with Direction
in the case of Vertical-DOWN, and this was also the situation type that produced the
most significant ‘Path’ (i.e. Direction) bias in the similarity judgment task. While this
may be somewhat post hoc, we can interpret the difference between the two kinds
of Vertical motion stimuli in terms of ‘degrees of boundedness’: e.g. rolling down is
more open-ended than rolling-up to the top of a hill, and hence the Vertical DOWN
stimuli represented the least bounded situation in the set. Thus we are lead to a tentative
generalization (and prediction): The more bounded a situation, the more salient Manner
will be for speakers of a V-language.
Pourcel (2005: 149) calls a similar interpretation that Zlatev and David (2004)
offered of these results (though in terms of the concept of telicity) ‘counter-intuitive’,
but we beg to disagree. As pointed out earlier, Bohnemeyer et al. (ms) noted a general
tendency for lower same-Manner bias in the Vertical triads in the 17 languages studied,
and while they did not find a general interaction with language-type, it remains unclear
to what extent all the different languages in their sample abide by the ‘boundary crossing
constraint’. Swedish and Thai do not, and we did not find a bounded/unbounded asym-
metry in their speakers’ similarity judgments, which in the case of French we did. Pourcel
(2005) also found an asymmetry, but in the opposite direction: greater Path salience for
the bounded than for the unbounded situation. However, the design-differences between
the two studies can perhaps be called on for an explanation, cf. Section 6.
410 LANGUAGE, COGNITION AND SPACE

Finally, note that we do not interpret the combined results of Study 1 and Study 2
in terms of a ‘Whorfian’ effect, since the differences in the categorization preferences
between the language groups seems to be due to an interaction between language-
independent differences in the situation types, and the constraints of a particular (type
of) language. To further investigate this possible interaction, we conducted our next
study, which more explicitly contrasts different contexts in which language can be
thought to influence the categorization of motion situations to different degrees.

5.3 Study 3

For the purpose of our third study, we modified the Event Triads elicitation tool so that
two groups of 12 Swedish and two groups of 12 French subjects participated: Group 1
for both languages performed the similarity judgment as in the original Event Triads
tool, whereas for Group 2 there was a break after the first segment and the participant
was asked to ‘describe the film just seen’, after which the second segment was shown and
the participant was asked to make the similarity judgment. Furthermore, the number
of distracters was decreased rather drastically from 38 to 8, leaving the total number of
triads per participant to 20, where each first segment was described by all participants:
for Group 1 after the similarity judgment task was competed, and for Group 2 prior
to each judgment. In this way we could investigate possible correlations between the
descriptions and the choices not only on a type-by-type basis (as in Studies 1 and 2),
but also on a triad-by-triad (instance) basis. The reduction of distracter triads was
necessary, since describing 50 video-clips, most of which are near-identical, would have
been both tiring for the participants and could lead to a sort of ‘habituation’ in which
they would fall into a stereotypical pattern of description that is less likely to reflect
naturalistic language use.
The results were highly interesting. Whereas the similarity judgments for Group 1
(post-choice description) were similar to those in Study 1 and practically identical for
the two languages (chi2(1) = 0.14, p > 0.05), i.e. a preference for same-Manner choices
(albeit a weaker preference, cf. Figure 3), the situation was completely reversed for
Group 2 (pre-choice description), with a surprisingly strong bias for same-Path choices,
as shown in Figure 5 for both the French and Swedish groups. The difference between
Group 1 and Group 2 was extremely significant (p < 0.0001). Furthermore, there was
a stronger Path/Direction bias for SG2 than FG2, which was also significant (chi2(1)
= 4.964, p=0.026). 7
When we divided the 12 test triads according to the three types of situations as
before (IN/OUT, FROM/TO and VERTICAL), we noticed, however, also a difference
between Group 1 and the previous results: in the case of VERTICAL the Manner-bias
was neutralized for both the Swedish and the French speakers (the slight difference
between SG1 and FG1 for Vertical is not statistically significant): see Figure 6 and 7.
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 411

Figure 5. Total results of same-Path/Direction vs. same-Manner preference for French (FG1) and
Swedish (SG1) Group 1 (post-choice description) and French (FG2) and Swedish (SG2) Group 2 (pre-
choice description). Total number of choices is 144 per group

Figure 6. The results for the two Swedish groups, divided by the three different situation types: OUT-
INTO, FROM-TO and VERTICAL. Total number of choices per situation type and group is 48

We coded the linguistic descriptions for the presence of Manner expressions: Manner
verbs such as hoppa and sautille (‘jumps’) and adverbials such as snurrande or en roulant
(‘rolling’), Path expressions such as från or de (‘from’) and till or a (‘to’) and Direction
expressions such as upp (‘up’) or monte (‘climbs’) and ner (‘down’) or descend (‘descends’)
and looked for correlations between the presence of these elements and the choices of
the subjects.
412 LANGUAGE, COGNITION AND SPACE

Figure 7. The results for the two French groups, divided by the three different situation types: OUT-
INTO, FROM-TO and VERTICAL. Total number of choices per situation type and group is 48

Table 5. Correlations of significant value (Pearson’s Correlation, significant at > ± .3 at the .05-level, two tailed)
between elements in the descriptions (Direction, Path, Manner) and corresponding choice for the two groups of
French speakers (FG1 and FG2) and the two groups of Swedish speakers (SG1 and SG2), divided by situation type
(From/To, Out/Into, Vertical) Non-existing or non-significant correlations are marked as ×

Group Type Direction Path Manner

From/To × × ×
FG1 Out/Into × × ×
Vertical × +.308 ×
From/To × × ×
FG2 Out/In × × - 304
Vertical × × -.329
From/To × -.309 ×
SG1 Out/Into -.338 × ×
Vertical -.302 × ×
From/To × -.307 ×
SG2 Out/Into × -.443 ×
Vertical +.674 × ×

Surprisingly, there were few positive correlations: for SG2-Vertical and for FG1-Vertical.
In qualitative terms, this means that if a subject had used a Direction expression, he
was more likely to make a same-Direction than same-Manner choice. We are not sure
how to interpret the negative correlations for SG1, SG2 and FG2. On the face of it, it
seems that e.g. if a French speaker had used a Manner expression (in the pre-choice
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 413

description group, FG2), he was less likely to make a same-Manner choice. Also, it
was surprising that the positive correlation for the French speakers was for FG1, the
post-choice describing group, and it involved Path, rather than Direction expressions.
In other words, the results do not lend themselves to an explanation in terms of Slobin’s
(1996) thinking-for-speaking hypothesis. According to the latter, and the classification of
Swedish as an S-framed and thus Manner-salient language and of French as a V-framed
language, one would have expected a Manner correlation for SG2 a Direction/Path
correlation for FG2. In fact, the only clear correlation was for SG2, and it involved
Direction rather than Manner.
A further indication that the results cannot be explained only on the basis of lin-
guistic differences and their effects is the neutralized Direction-Manner bias in the
case of the translocative unbounded (VERTICAL) situations for both the French and
the Swedish groups (see Figures 6 and 7). Unlike the results from Study 2, this cannot
be explained by a linguistic effect since Swedish does not obey the boundary-crossing
constraint. Thus, the asymmetry in the choices between VERTICAL and the other two
types of stimuli thus corroborate our claim in Section 3 that bounded and unbounded
translocative situations differ (even) pre-linguistically. We may express this by saying
that Direction is conceptually simpler than Path: all that is required is to pay attention
to the vector (or the shape of the trajectory) of translocation, rather than perform an
explicit ‘parsing’ of the translocative event in terms of Source, Via and/or Goal. Like
Manner, Direction seems to be a category that is more perceptually given than conceptu-
ally derived, and thus less subject to the effects of linguistic mediation, as understood
by Vygotsky (1978, 1986).

6 Discussion: from linguistic relativity to linguistic mediation

The Soviet psychologist Lev Vygotsky (1896–1934) distinguished between ‘higher’ and
‘lower’ mental functions, described by Kozulin (1986: xxv) as follows:

Vygosky […] made a principal distinction between ‘lower’, natural mental functions,
such as elementary perception, memory, attention, and will, and the ‘higher’, or
cultural, functions which are specifically human and appear gradually in a course
of radical transformation of the lower functions.

Thus, what is uniquely human, according to Vygotsky, is the ability to use artefacts and
signs, mediating between perception and behaviour, and functioning as ‘psychologi-
cal tools’ for the purpose of reflection and self-regulation: ‘the central fact about our
psychology is the fact of mediation’ (Vygotsky 1933, quoted by Wertsch 1985:15).
The most important kind of signs, and thus psychological tools, are according
to Vygotsky those of language. Like artefacts, linguistic signs are initially social and
interpersonal, but with experience become internalized and thus intra-personal.
Vygotsky argued that such internalization occurs via so-called ‘egocentric speech’ in
414 LANGUAGE, COGNITION AND SPACE

early childhood, and that such speech is highly functional for the child since its presence
increases with the difficulty of the task to be performed.
Applying the notion of linguistic mediation to the triad studies, both our own, and
those described in Section 4, allows us to make sense of most of the results reported in
the literature. First, due to the nature of the task, the similarity judgment task can be
performed either more directly (i.e. using perceptual categorization) or more mediatedly
(i.e. using external or internal speech). This can explain the results of both Gennari et
al. (2002) and Finkbeiner et al. (2002), in which a typologically congruent bias was
observed in the tasks where language was used either overtly or (apparently) covertly,
but not otherwise. On the other hand, if Manner is a category which is (in general)
more perceptually and conceptually simpler than Path, as suggested earlier, then tasks
which induce categorization through less mediated processes, should bias for Manner
rather than Path, and vice versa. We can thus explain the results of Study 3 for both the
Swedish and the French groups through a possible ‘Vygotskyan effect’ of language on
the categorization of (translocative) experience: linguistic mediation yields an explicit
‘parsing’ of the components of a motion situation, and thus attention to more abstract
components such as Path than to more perceptually immediate components such as
Manner (or Direction). Such an effect appeared to be independent of typological differ-
ences between languages. At the same time, this interpretation predicts that if Manner
is expressed in French, it will be more prominent semantically. Whether this would lead
to a cognitive effect, however, is less clear: Study 2 seems to support this, while Study 3
(e.g. the negative correlations for Manner in Table 5 for FG2), did not.
At the same time, if the Manner of motion is of a complex type, such as that in the
stimuli used by Finkbeiner et al. (2002), while ‘Path’ is more a matter of ‘moving left/
right’ and thus Direction, then the opposite tendency should be observed: a greater
same-Manner bias will be observed in the more demanding task, involving sequential
presentation and language-based short term memory, which again was the case estab-
lished in that study.
This can furthermore even help us understand the apparently contradictory findings
in the triad study of Pourcel (2005): In her first experiment with both French and English
participants, the sequential presentation of stimuli possibly already induced the use of
internal speech, resulting in an overall preference for ‘same Path’. The second experiment
used explicit written description, which ‘balanced’ the preferences somewhat, but still
privileged Path. What remains unaccounted for, though, is why ‘same-Manner’ prefer-
ences were higher for the ‘atelic Path’ (unbounded) situations than for the ‘telic Path’
(bounded) situations, while in our studies the asymmetry was in the reverse direction:
a neutralization of the Manner-bias, and thus relatively lower ‘same-Manner’ prefer-
ences for the ‘less bounded’ situations. The divergent results can perhaps be explained
by the marked difference in the nature of the stimuli used: whereas the relevant kinds
of Manner in Pourcel’s experiment were mostly of the ‘default’ kind and thus less per-
ceptually salient, those in our studies were all attention-grabbing, yielding an overall
Manner-bias on categorization (mostly) on the basis of perceptual processes. This bias
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 415

was then reduced for the slope scenes due to the similarly perceptually more immediate
notion of Direction and possibly also due to the greater ease of verbalizing Manner in
the same clause as Direction (as opposed to Path) for French. On the other hand, Path
and Direction were the most relevant aspects in Pourcel’s stimuli, while (simple clause)
verbalization would instead have promoted Manner to higher prominence. In any case,
both studies imply the importance of distinguishing between what we have analyzed
as bounded vs. unbounded translocation, and thus offer support for the taxonomy
presented in Section 3.

7 Summary

In this chapter we have tried to show that ‘motion event’ typology has suffered for
quite some time from conceptual and empirical problems, and despite the indubitable
contributions of scholars such as Talmy and Slobin, it is time that we move on, and
establish a more coherent framework for describing our experiences of motion. Inspired
by the literature on situation types (Vendler 1967), as well as Durst-Andersen (1992)
and Pourcel (2005), we have attempted to provide one such framework through our
taxonomy of motion situations, which, we suggest, are largely independent of the way
different languages ‘lexicalize’ motion.
The second step, which we have only here touched, is to try to establish how as
many (diverse) languages as possible express this experience. Talmy’s binary typology
has clearly outlived its time, but exactly how many different types of languages in terms
of their expression of translocation there are is currently an open question.
In the cases where languages systematically differ in this respect, we can investigate
possible linguistic effects of various sorts and strengths on seemingly ‘non-verbal’ cogni-
tive tasks, and thus contribute to the neo-Whorfian program. We have described three
such studies which suggest at least some effect of the differences between French on
the one hand, and Swedish and Thai on the other, on the categorization of transloca-
tive situations on the basis of the components Path, Direction and Manner, arguing
for the necessity of distinguishing between the first two. The effects have, however,
been attributed to an interaction between language-independent factors and linguistic
constraints, and cannot support a strong version of the Whorfian hypothesis (‘different
languages entail different worldviews’).
We have also argued that we should be open to the possibility that the differences
between languages may be relatively minor compared to their similarities – at least
as far as the categorization of (motion) experience is concerned – and have thus sug-
gested possible ‘Vygotskyan’ rather than ‘Whorfian’ effects, based on the differential role
of linguistic mediation in the different tasks and study designs. Further studies with
(typologically) different languages are likely to shed more light on these issues. Progress
in linguistic typology and psycholinguistics should thus go hand-in-hand.
416 LANGUAGE, COGNITION AND SPACE

Acknowledgements
We wish first of all to thank all the participants in our studies, most of who did so without
any economic compensation, and furthermore the host institutions who made the studies
possible, including the Centre for Research in Speech and Language Processing (CRSLP) at
Chulalongkorn University. We also thank the participants at seminars at the Copenhagen
Business School and at Lund University, an anonymous reviewer and the editors of this
volume for their useful comments. Also Joost van Weijer for help with statistical analysis.
Finally, we express our warmest gratitude to Jürgen Bohnemeyer, Sonja Eisenbeiss and
Bhuvana Naranhimsan for generously allowing us to use (and modify for our purposes)
their Event Triads tool.

Notes
1 Zlatev (2005, 2007) refers to this as the distinction between ‘elaborated’ and a ‘sche-
matic’ concept of Path, and argues for the need to separate the latter from the concept of
Direction, as in the present chapter.
2 The original examples in Pourcel (2005, Chapter 5) are respectively (6), (90) and (92).
3 Note that our use of the term ‘figure’ corresponds to that used by Talmy (2000) and
Levinson (2003), the term ‘trajector’ (Lakoff 1987; Regier 1996; Zlatev 1997) or ‘referent’
(Miller and Johnson-Laird 1976). On the other hand, our use of the term ‘landmark’, is
more specific than that used in much of the cognitive linguistic literature (Langacker
1987), in referring to some physical object, which is typically expressed through a noun
phrase in language (cf. Zlatev 2005, 2007).
4 One thing to be borne in mind, however, is that this design has been shown to give a
general bias towards Manner-based categorization, probably due to the conspicuousness
of the motion of the ‘tomato’ figure, so that the results produced using this stimulus tool
cannot be directly compared with results obtained using another elicitation tool (cf.
Kopecka and Pourcel 2005).
5 This unequal distribution was due to the fact that at the time of our first study we had
not yet realized the importance of distinguishing between the three types.
6 Notice that this also helps explain the high proportion of Manner verbs produced by the
French group, shown in Table 3.
7 However, since the Group 2 data was both compared both with Group 1, and within
the two sub-groups (FG2 and SG2), Bonferroni correction (here, p-value * 2) would be
required, placing the difference between FG2 and SG2 on the border of significance.

References
Bohnemeyer, J., Eisenbeiss, S. and Naranhimsan, B. (2001) Event triads. In S. C.
Levinson and N. Enfield (eds) ‘Manual’ for the field season 100–114. Nijmegen,
Max Planck Institute for Psycholinguistics.
Bohnemeyer, J., Eisenbeiss, S. and Naranhimsan, B. (ms) Ways to go: Methodological
considerations in Whorfian studies on motion events.
 TRANSLOCATION, LANGUAGE AND THE CATEGORIZATION OF EXPERIENCE 417

Blomberg, J. (2006) Rörelse och verbalisering – En studie av språkets påverkan på

kategoriseringen av rörelsesituationer. BA thesis. Centre for Languages and
Literature, Lund University.
Blomberg, J. (2007) Linguistic relativity, mediation and the categorization of motion.
MA thesis. Centre for Languages and Literature, Lund University.
Brown, P. (2004) Position and motion in Tzeltal frog stories: The acquisition of nar-
rative style. In S. Stromqvist and L. Verhoeven (eds) Relating events in narrative:
Cross-linguistic and cross-contextual perspectives 37–58. Mahwath, NJ: Earlbaum.
Durst-Anderson, P. (1992) Mental grammar: Russian aspect and related issues.
Columbus, Ohio: Slavica Publishers, Inc.
Finkbeiner, M., Nicol, J., Greth, D. and Nakamura, K. (2002) The role of language in
memory for actions. Journal of Psycholinguistic Research 31(5): 447–457.
Gennari, S. P., Sloman, S. A., Malt, B. C. and Fitch, T. (2002) Motion events in
language and cognition. Cognition 83: 49–79
Husserl, E. (1999) [1907] The idea of phenomenology. (Trans. Lee Hardy, Collected
Works Vol. 8.) Dordrecht: Klewer.
Itkonen, E. (1991) Universal history of linguistics. Amsterdam: John Benjamins.
Kopecka, A. and Pourcel, S. (2005) Figuring out figures’ role in motion conceptu-
alization. Presentation, International Cognitive Linguistics Conference, Seoul,
Korea. 21 July 2005.
Kozulin, A. (1986) Vygotsky in context. Preface to L.S. Vygotsky Thought and
Language. Cambridge, MA: MIT Press.
Lakoff, G. (1987) Women, fire and dangerous things: What categories reveal about the
mind. Chicago: Chicago University Press.
Langacker, R. (1987) Foundations of cognitive grammar Vol. 1. Theoretical prerequi-
sites. Stanford: Stanford University Press.
Levinson, S. C. (2003) Space in language and cognition: Explorations in cognitive
diversity. Cambridge: Cambridge University Press.
Miller, G.A. and Johnson-Laird, P.N. (1976) Language and perception. Cambridge,
MA: Harvard University Press.
Papafragou, A., Masely, C. and Gleitman, L. (2002) Shake, rattle, ‘n’ roll: The repre-
sentation of motion events in language and cognition. Cognition 84: 184–219
Pourcel, S. S. (2005) Relativism in the linguistic representation and cognitive
conceptualisation of motion events across verb-framed and satellite-framed
languages. PhD Thesis, Department of Linguistics and English Language,
University of Durham.
Pourcel, S. S. and Kopecka, A. (ms) Motion events in French. Typological intricacies.
Regier, T. (1996) The human semantic potential: Spatial language and constrained
connectionism. Cambridge, MA: MIT Press.
Slobin, D. I. (1996) From ‘thought and language’ to ‘thinking for speaking’. In
J. Gumperz and S. C. Levinson (eds) Rethinking linguistic relativity 70–96.
Cambridge: Cambridge University Press.
Slobin, D. I. (1997) Two ways to travel: Verbs of motion in English and Spanish. In
M. Shibatani and S. Thompson (eds) Grammatical constructions: Their form and
meaning 195–220. Oxford: Oxford University Press.
418 LANGUAGE, COGNITION AND SPACE

Slobin, D. I. (2003) Language and thought online: Cognitive consequences of

linguistic relativity. D. Gentner and S. Goldin-Meadow (eds) Language in mind
157–192. Cambridge: MIT Press.
Slobin, D. I. (2004) The many ways to search for a frog: Linguistic typology and the
expression of motion events. In S. Strömqvist and L. Verhoeven (eds) Relating
events in narrative: Cross-linguistic and cross-contextual perspectives 219–257.
Mahwath, NJ: Earlbaum.
Slobin, D. I. and Hoiting, N. (1994) Reference to movement in spoken and signed
languages 487–505. Proceedings of the 20th Annual Meeting of the Berkeley
Linguistics Society.
Smith, V. (2003) Motion at the sugar factory: Is Russian a genuine Manner language?
In Proceedings of the 14th Symposium on LSP. University of Surrey, 18–22 August
2003.
Strömqvist, S. and Verhoeven, L. (eds) (2004) Relating events in narrative: Cross-
linguistic and cross-contextual perspectives. Mahwath, NJ: Earlbaum.
Talmy, L. (1985) Lexicalisation patterns: Semantic structure in lexical forms. In
T. Shopen (ed.) Language typology and syntactic description Vol. 3 57–149.
Cambridge: Cambridge University Press.
Talmy, L. (1991) Path to realisation: A typology of event conflation 480–520.
Proceedings of the Seventeenth Annual Meeting of the Berkeley Linguistics Society.
Talmy, L. (2000) Toward a cognitive semantics: Concept structuring systems Vol. 1.
Cambridge, MA: MIT Press.
Vendler, Z. (1967) Linguistics in philosophy. Ithaca: Cornell University Press.
Vygotsky, L. S. (1978) Mind in society. Cambridge, MA: MIT Press.
Vygotsky, L. S. (1986) Thought and language. Cambridge, MA: MIT Press.
Wertsch, J. V. (ed.) (1985) Vygotsky and the social formation of mind. Cambridge,
MA: Harvard University Press.
Whorf, B. L. (1956) Language, thought and reality: Selected writings of Benjamin Lee
Whorf. (ed. J. B. Carroll.) Cambridge: MIT Press.
Zlatev, J. (2003) Holistic spatial semantics of Thai. In E. Casad and G. Palmer (eds)
Cognitive linguistics and non-Indo European languages 305–336. Berlin: Mouton
de Gruyter.
Zlatev, J. (2005) Semantics of spatial expressions. Encyclopedia of language and
linguistics. (Second edition, Article 00298.) Oxford: Elsevier.
Zlatev, J. (2007) Spatial semantics. In D. Geeraerts and H. Cuyckens (eds) Oxford
handbook of cognitive linguistics 318–350. Oxford: Oxford University Press.
Zlatev, J. and David, C. (2003) Motion event constructions in Swedish, French and
Thai: Three different language types? Manusya 6: 18–42.
Zlatev, J. and David, C. (2004) Do Swedes and Frenchmen view motion differently?
Presentation at the conference Language, Culture and Mind. Portsmouth, 20 July
2004.
Zlatev, J. and Yangklang, P. (2004) A third way to travel: The place of Thai in motion
event typology. In S. Stromqvist and L. Verhoeven (eds) Relating events in nar-
rative: Cross-linguistic and cross-contextual perspectives 159–190. Mahwath, NJ:
Earlbaum.
16 Motion: a conceptual typology
Stéphanie Pourcel

1 Introduction

The domain of motion in space has been extensively studied in cognitive linguistics (e.g.
Talmy 1985, Slobin 2004) and in investigations of the language-cognition relationship
(e.g. Gennari, Sloman, Malt and Fitch 2002, Hohenstein 2005, Papafragou, Massey and
Gleitman 2002, Pourcel 2004, 2005, 2009a, b, Slobin 1996, 2000). Likewise, the domain
has received attention in psychological studies of perceptual and conceptual aspects of
motion information processing (e.g. Mandler 2004). Yet, comparatively few efforts – if
any – have sought to adduce a comprehensive framework for the conceptual definition
of motion variables and motion types. In this chapter, I examine the domain of motion
independently of language in order to design a conceptual typology of motion that will
serve linguistic analyses and applications of these analyses. Importantly, the typology
attempts to categorise types of motion in a way that is concordant with how human
minds categorise motion types in conceptualisation – using cognitive data obtained from
categorisation tasks. In other words, the conceptual typology is not based on language
patterns, as are linguistic typologies, but on conceptual categories. The conceptual
typology therefore offers a classification of events and sub-events pertaining to one
domain, in which the types are isolated on conceptual grounds, rather than on linguistic,
cultural, folk, or otherwise arbitrary grounds. By virtue of being based on conceptual,
rather than symbolic, representations, the typology characterises a fundamentally
human understanding of given domains. The typology should thus inform potentially
universal patterns of classification, rather than language- or culture-specific ones. The
present research tests this possibility only to an extent by using native populations of
typologically diverging languages – French, Polish and English.
Sketching a conceptual typology for the domain of motion is important for sev-
eral reasons. First, motion is a complex domain with dynamic and variable schematic
components. These components do not simply comprise moving figures following
spatial paths. They also comprise physical manners of displacement, force dynamics,
landscapes, locations, causal motivations, goals, resultative endstates, objects (e.g. vehi-
cles, instruments, buildings), and more, which are situated within ideational contexts,
comprising cultural cognitive models, ideologies, emotions, symbolisms, as well as
expectations concerning motion properties and contextual embedding. Each of these
motion component is rich and variable in nature, which entails that motion types are in
turn numerous, rich and complex. Crucially, this suggests that the domain of motion is
unlikely to be conceptualised in a unitary fashion. The specific claim made in this paper
is that this complex and variable range of event realisations reflects conceptually distinct
types of motion. Motion types have consequences for the cognitive representation of

419
420 LANGUAGE, COGNITION AND SPACE

the event in question, e.g. in memory, categorisation, analogy, etc. On this basis, it
should be possible to isolate different types of motion, which may correspond to distinct
conceptual realities.
Besides its rich complexity, motion is a pervasive domain of experience, which is
conceptualised and also expressed in language with high frequency in human daily
life. As a result, the domain of motion has received extensive attention in cognitive
linguistics, for instance, in typological work (e.g. Talmy 1985, 1991, 2000), in lexico-
discursive studies (e.g. Slobin 2004), and in explorations of the language-cognition
relationship (e.g. Gennari et al. 2002, Papafragou et al. 2002, Pourcel 2009a, b,, 2005,
Zlatev and David 2004, 2005). Most studies have faithfully adhered to the usage-based
tradition embraced by cognitive linguistics, and have therefore been comparative in their
treatment of linguistic data. However, the generalisations drawn – though useful – are
typically emergent from the language data and take little account of the conceptual
reality of motion independently of language. This does not mean that existing means
for analysing motion are incorrect, but that they largely remain products of the reality
depicted by the languages observed. These means are therefore anchored in herme-
neutics and correspond to language-embedded categories, rather than to conceptual
categories pertaining to an ‘objective’ reality. To the extent that cognitive linguistics
seeks to commit to an understanding of language that is concordant with the workings
of the human mind, generalisations that are mainly or solely language-based prove
potentially problematic.
In addition, the categories identified in linguistic work are often applied beyond the
realm of linguistic description to questions of conceptual relativism (see Whorf 1956,
Lucy 1992). For instance, a number of studies have investigated whether richer lexical
resources and more systematic means of encoding manners of motion in the grammar
of a particular language render those manners more cognitively salient to their native
speakers (e.g. Gennari et al. 2002, Papafragou et al. 2002, Zlatev and David 2004, 2005,
Pourcel 2009a, b). To guide their assumptions, hypotheses and experimental designs,
these studies have departed from cross-linguistic differences. In other words, they have
adopted the categories found in language to investigate matters of conceptualisation,
which are richer and more complex than the linguistic means used to encode them.
Few of these studies have reached successful conclusions or even consensus across
their respective findings. The outcome of these studies might have proved altogether
different had they considered, from the outset, the conceptual reality of motion – inde-
pendently of language – and had they examined their data relative to conceptually
real categories, rather than to solely linguistically-defined parameters such as path
and manner. In addition, though their methodologies were often closely similar (e.g.
triad-based similarity judgement tasks), research teams have often used different figures
in their motion stimuli. For analytical purposes, therefore, the design of a conceptual
typology of motion with greater attention to figure types and other related schematic
properties is paramount to avoiding linguacentric tendencies in research, which go
against the very stance of cognitive linguistics as a language-analytic enterprise. The
issue of linguacentrism in linguistic, cognitive and behavioural research has long been
the subject of methodological discussions (e.g. Whorf 1956: 162), yet this key pitfall is
 MOTION: A CONCEPTUAL TYPOLOGY 421

still not entirely addressed in contemporary studies investigating the language-cognition

interface (Lucy 2003: 25).
This paper attempts to address the notion of linguacentrism in linguistic and
relativistic research, with special attention to the domain of motion. To do so, it offers
a preliminary sketch of a conceptual typology for the domain of motion, based on
experiential facts and cognitive data, with the types isolated corresponding to important
conceptual differences and, potentially, to actual categories of events. In doing so, it
aims (i) to represent comprehensively the complexity of the domain of motion, (ii)
to offer a ‘language-neutral’ analytic template (or metalanguage) for the comparative
study of motion linguistics, and (iii) to allow for greater discernment in applications
of motion linguistics to questions of cross-linguistic conceptual representations (e.g.
linguistic relativity).
To achieve this, the paper proceeds with an initial illustration of the diversity exist-
ing across types of motion, followed by a proposal to place motion figures at the centre,
or basis, of the conceptual typology. The choice for basing the typology on figure type is
motivated by the fact that the physical properties of figures constrain and determine to
a large extent the types of motion figures undergo, with impacts on types of manners,
paths, causal relations and so on. A detailed outline of figure types and their properties
makes it possible to establish a conceptual typology of motion types. Within these
motion types, special consideration is paid to schematic variables such as paths, man-
ners, causality, agency and intentionality. This proposal, together with the conceptual
distinctions between motion types are substantiated with experimental data drawn from
categorisation tasks implemented with speakers of various languages, including English,
French and Polish. The tasks at hand systematically represented instances of motion in
a visual medium, e.g. film clips and film extracts. The stimuli were then non-linguistic.
This paper presents examples of these motion scenes in the written medium required
by the static publication format. Yet, it is key to understand that the discussion is not
focused on the linguistic examples (here, in English), but instead, on the conceptual
reality encoded by these examples.
The paper ends with a suggestive sketch for a conceptual typology of motion, based
on the figure types identified.

2 Some important distinctions in motion types

A few examples of motion situations suffice to illustrate the variability of possible motion
types and of the internal properties of individual motion schemas. Consider the fol-
lowing scenarios:1

(1) Helen is jogging.

(2) Helen walked to the store.
(3) Tom pushed the pram along the street.
(4) Mum rocked the baby to sleep.
422 LANGUAGE, COGNITION AND SPACE

These four examples illustrate a variety of types of motion. For instance, (1) depicts
a motion activity where the ongoing nature of the motion, together with its manner
of completion are central to conceptualising the event. In (2), the motion type is no
longer an activity, but corresponds more adequately to a motion event, as commonly
discussed in the literature. An important difference between (1) and (2) concerns
the presence of directionality and the reaching of a goal or destination in the second
example. In (2), the directionality, or path endpoint, represents the goal of the figure’s
motion, whereas in (1), the motion itself constitutes the figure’s goal. (1) and (2) thus
contrast instances of activity versus event, respectively (Pourcel and Kopecka 2005,
to appear).
In (3) and (4), the goal of the figure’s motion no longer applies to the motion figure
itself, but to an external entity. For instance, in (3), the figure seeks to alter the spatial
location of an object – here, a pram; and in (4), the figure seeks to alter the state or
condition of another being. In both cases, motion is instrumental in completing these
goals, in that motion is used as a means to an end. The pushing and rocking events
depicted in (3) and (4) are instances of causal motion.
The two situations in (3) and (4) are distinct, however. In (3), the figure’s motion
causes the change of location of another figure (i.e. effectively the pram’s motion along
a path), whereas in (4), the figure’s motion causes the change of state of another figure,
where the rocking causes the baby to fall asleep. A situation like (4) may be characterised
as caused state, that is, the motion of Figure 1, i.e. the mother, causes the change of
state of Figure 2, i.e. the baby. The state is dependent on the motion. In (3), we also
have two figures. The two figures are moving figures which undergo a change of spatial
configuration. The motion of Figure 2, i.e. the pram, is dependent upon the motion of
Figure 1, i.e. Tom. (3) thus represents an instance of caused motion.
In addition, (3) illustrates that moving figures, such as people and objects, may be
agentive or passive. Agentive figures are capable of self-motion, whereas passive figures,
such as objects, are subject to caused motion. This distinction reflects a fundamental dif-
ference in terms of animacy and, often, intentionality. In caused motion, the primary, or
motion-causing, figure is animate. This points to another potential distinction between
animate and inanimate motion.
Additional distinctions are present when contrasting motion instances (2) and (3).
Both are directional motion events. Yet, (2) indicates a clear path endpoint, whereas (3)
does not. In other words, the motion goal is overt in (2), but not in (3). This difference
may be characterised in terms of telicity (Aske 1989, Talmy 1991), and yields another
contrast between telic motion where the motion goal is explicit, e.g. (2), and atelic motion
where the motion goal is not apparent, e.g. (3).
Finally, when contrasting the motion situations in (1)-(4), we also note differences
in terms of the physical properties involved in motion performance. These properties
differ relative to aspects such as degree of control, muscular effort, speed, and general
force dynamics. Types of motion may therefore be distinguished relative to physical
characteristics. For instance, walking – to an average adult human being – involves
relatively little effort or control compared to jogging. In fact, walking is a typical manner
of adult human motion. (2) represents an instance of default motion (Pourcel 2004).
 MOTION: A CONCEPTUAL TYPOLOGY 423

Other manners such as jogging, or limping, waltzing, staggering, and so forth, represent
less typical, or non-default instances of motion. Distinctions linked to the physical
manner of displacement of a figure thus lead to additional types of motion.
In sum, motion types may be classified relative to directional, aspectual, causal,
agentive, or physical properties, and possibly more. The few types of motion reviewed
so far present important conceptual characteristics, despite the fact that they are
mostly encoded in language (here, English) using the same constructional pattern
(Talmy 1985). Indeed, all the above examples illustrate the satellite-framed pattern
of linguistic encoding for motion typical of the English language. This pattern maps
the figure onto the subject position of the sentence, the manner of motion onto the
main verb, the path onto a verb particle, or satellite, and the ground onto a nominal
complement to the satellite. Additionally, in causal cases, the manner verb is followed
by the second motion figure in a direct object. In other words, one main grammatical
formula, or construction type, serves to encode the distinct types of motion outlined
above. Formulaic tendencies in language do not preclude, however, the existence of
significant conceptual distinctions across the events expressed. It is in conceptual,
rather than linguistic, terms that this article aims to analyse the domain of motion. To
do so, it is necessary to unpack and deploy apparent schematic differences between the
types of events that motion may entail, regardless of the linguistic patterns available
for the expression of this domain.

2.1 Directionality

A broad distinction may be drawn between motion types where motion is incidental to
a change of location, i.e. motion events, and motion types where motion is the essence
of the figure’s action, i.e. motion activities. Consider, for instance:

(5) a. Helen is jogging.

b. Helen jogged to the store.

Examples in (5) both involve motion, but (5a) depicts an activity and (5b) an event.
The main difference between activities and events relates to directionality, as briefly
mentioned above. Activities lack overt directionality, whereas events necessitate direc-
tionality by virtue of entailing a change of locational grounds.
Motion activities then describe motion with an emphasis on the type of motion
itself – typically its manner. Activities, like all types of motion, include spatial reference,
i.e. a ground, but the notion of directionality, or path, is not salient. In fact, there may
not be any overt directionality at all in some instances, as in the act of running on a
treadmill. The essential schemas in activities consist of the figure and its actual motion
as characterised by a specific manner. The high salience of the manner schema versus
the low salience of the path schema in motion activities is incidentally examplified in
language, where activities may be expressed without mention of path details, e.g.
424 LANGUAGE, COGNITION AND SPACE

(6) Stop kicking your sister!

(7) Mary swims every morning.

Motion events, on the other hand, describe directional or goal-oriented motion. In this
case, paths are salient conceptual dimensions of motion, whereas manners are merely
instrumental to following the course of the motion path. The core schema of a motion
event is therefore its path (Talmy 1991). To further illustrate the core schematicity of
path and the secondary importance of manner, linguistic mappings of motion events
in English may leave manner information unexpressed, using a neutral motion verb
such as go or a path verb instead, e.g.2

(8) Mary went to the store.

(9) Tom crossed the bridge.

The distinction between event and activity is important as it contrasts the salience of
the manner and of the path of motion, two central elements of motion. It is possible
that this distinction is conceptually real, so that activities and events are systematically
conceptualised as different types of actions. This possibility remains to be tested at
this stage.

2.2 Telicity

Besides directionality, motion involves an aspectual dimension, in that motion may

be ongoing or completed. This distinction is typically discussed in terms of ‘telicity’
(from Greek telos meaning ‘end’) – as briefly mentioned earlier. In the domain of space,
telicity, or event completion, is understood as the reaching of locational goals and
the obvious change in the locational state of the figure. Motion activities are typically
ongoing, uncompleted acts of motion and therefore are atelic. Motion events, on the
other hand, can be either telic or atelic depending on the type of motion path. Indeed,
although motion events involve directionality, the path goal may not necessarily be
apparent and salient. Consider, for instance, up events in which the top of the ground
is not readily at hand, visible, or reached. Likewise, along events entail by definition
an endpoint-free path, e.g.

(10) John walked along the beach for hours.

These types of events are atelic due to the lack of a path endpoint, yet they differ from
activities by virtue of having directionality.
Motion events are most frequently telic, however. They typically involve an endpoint
and a change of location or state. Consider the instance of an up event in which the top
 MOTION: A CONCEPTUAL TYPOLOGY 425

of the ground is readily visible or reached. The trajectory endpoint is highly salient in
this case. Telic events may also involve change of location via the crossing of boundaries,
as in crossing, entering, or exiting events, e.g.

(11) a. The plane flew across the Atlantic.

b. Mary dived into the pool.
c. Jo walked out of the room.

Note that telic events lead to resultative interpretations of motion. By undergoing motion
along a path, the figure reaches a goal, which corresponds to a particular location in
space. This resultative aspect is all the more obvious in cases of caused state events, where
the motion of a primary figure causes the resultative end-state of a secondary figure, e.g.

(12) The attacker stabbed the victim to death.

Aspect is thus an inherent property of motion, and we may distinguish an additional

two types of motion relative to this property, namely telic and atelic motion.

2.3 Causality

A further distinction across motion types is effected by the figure’s relation to its own
motion, to the motion of other figures, or to the endstate of other figures. Indeed, a
figure may instantiate its own motion or that of other figures. Consider,

(13) a. Felix crossed the street.

b. Tom pushed the pram across the street.
c. Mum rocked the baby to sleep.

Example (13a) presents a single figure initiating its own motion via its physical motor
abilities. (13a) is an instance of self-motion. In (13b) and (13c), on the other hand, we
have two interacting figures in each motion scene, i.e. Tom and the pram in (13b),
and Mum and the baby in (13c). As far as the motion is concerned, the two figures are
interdependent in their interaction. A primary figure initiates motion and, in so doing,
causes an alteration in a secondary figure. As a result, the secondary figure may undergo
a change of spatial location, as in (13b), or it may undergo a change of state, as in (13c).
Examples in (13b) and (13c) display instances of causal motion – from the perspective
of the primary figure. (13b) may also be characterised as caused motion, and (13c) as
caused state – these terms reflect the perspective of the secondary figure.
Self-motion then describes the physical motion of a figure which initiates its own
motion independently of external elements, whilst causal motion describes the motion
of a primary figure causing the alteration of a secondary figure. Causal motion may cause
either the motion of a secondary figure, or the change of state of a secondary figure.
426 LANGUAGE, COGNITION AND SPACE

In the case of caused motion, we obtain the motion of two figures with a cause-effect
relation between motion1 and motion2 (see (14) and (15) below). In effect, the motion
of Fig1 causes the motion of Fig2, so that Fig1 moves Fig2. Note that, linguistically,
motion1 characterises the manner of motion and motion2 captures the path of motion
– in English, e.g.

(14) The wind blew the hair into her face.

Fig1 motion1 Fig2 motion2
manner path

(15) Mary drove her sister to the airport.

Fig1 motion1 Fig2 motion2
manner path

Effectively, in (14), the wind causes the hair to go into her face by blowing, and in (15),
Mary causes her sister to go to the airport by driving her there.
In the case of caused state, on the other hand, we obtain the motion of one figure
and the state of a secondary figure with a cause-effect relation between the motion of
the primary figure and the state of the secondary figure. So, the motion of Fig1 results
in the state of Fig2, e.g.

(16) Sara kicked the door shut.

Fig1 motion Fig2 state

In sum, causal motion entails an interdependent relationship between two figures, with
the primary figure engaged in a motion act. Causal motion results either in the change
of location or in the change of state of a secondary figure.
In addition, the causal nature of motion impacts on the agentive value of the motion
path. In the case of self-motion, path is agentive, in that the path is instantiated and
followed by the motion-initiating, or primary, figure, e.g.

(17) Jim dived INTO the pool.

Causal motion, on the other hand, involves resultative paths, as in (14)-(15) above and
(18) below.

(18) Carol poured her drink OVER Gary’s head.

In causal motion then, the path corresponds to the result of Fig1’s motion and refers
to what is effectively happening to Fig2. Fig1 does not necessarily follow Fig2’s path.
Note that it may, as in (15) above.
This outline points to important differences in motion types relative to causal
properties. These differences mainly concern the physical nature of the figure, as well as
the interactions between distinct figures. Causal properties, in addition, determine the
 MOTION: A CONCEPTUAL TYPOLOGY 427

agentive value of the path of motion, which may be either agentive or resultative. These
differences demonstrate the dynamic complexity of motion as a domain of experience,
which may, as a result, engender divergent conceptual representations for the types
observed.

2.4 Animacy

As highlighted above, in causal motion, the secondary figure may be both an object,
e.g. a pram, or a person, e.g. a baby, but the primary figure cannot be an object, such
as a pram. The primary figure needs motor capacities or inherent force dynamics to
initiate its own motion and that of other figures. To fulfill this requirement, the primary
figure needs to be animate. If we take the capacity for self-motion as a defining criterion
for judging animacy, then we may categorise as animate any live creature with motor
abilities (e.g. humans, animals), natural elements (e.g. water, fire), as well as natural
forces in the universe (e.g. magnetism, electricity), and force-animated entities (e.g.
planets, currents), and so on.3 On the other hand, dead organisms, objects, and even
vehicles can only undergo displacement in space when set into motion by animates. It
follows that inanimate figures may only be subject to causal motion. On this basis, we
may distinguish animate (e.g. (19a)-(b)) from inanimate motion (e.g. (19c)).

(19) a. Jayne kicked the ball.

b. The earth rotates around its own axis.
c. The ball rolled down the hill.

Examples in (19) present conceptual differences relating to animacy properties. In terms

of self-motion capacities, only (19a) and (19b) offer figures that may cause their own
motion and that of other figures. In (19c), the figure is in motion presumably because
it was set into motion by a human or animal being, or by a natural element such as the
wind or other. We may therefore distinguish animate motion from inanimate motion.

2.5 Agency

Figure properties seem significant in conceptualising motion types. The foregoing has
defined a distinction in terms of animacy. However, the motion of a live creature versus
that of a natural force seem to present equally important differences. Both types of figures
may be considered animate and capable of self-motion. However, only the former may
be characterised as agentive. That is, only live creatures, e.g. humans and animals, may
instantiate motion as an intentional act. Forces, on the other hand, do not operate on
an intentional basis, but on a purely mechanical one as relating to the laws of physics
in the universe. The intentional dimension of motion present with live beings relates to
agency, and may correspond to a figure’s explicit goals or instinctive impulses for spatial
displacement. Whether to the foreground or background of consciousness, the motion
428 LANGUAGE, COGNITION AND SPACE

of live beings is typically a cognitive response to the world (to the exception of reflex
motion), rather than a purely physical and mechanical one, and is often a volitional
act. Consider, for instance,

(20) The cat pounced on the butterfly.

(21) The tide is coming in.

We may therefore establish a further distinction between agentive and non-agentive

motion, as relating to figure properties. Note, too, that the motion of object figures is
non-agentive. This is also the case for animates undergoing causal motion, in the sense
that the figure, though animate and potentially agentive, is not in this case cognitively
active in engaging the act of motion.

2.6 Force dynamics

In addition, one may distinguish motion types relative to the figure’s physical capacity
for motion. An examination of the force dynamics of motion instances directs us to
manners of displacement. The case is most obvious with human figures, who are highly
flexible and creative in using their bodies and other objects to undergo motion. Consider,

(22) a. Chris walked across the car park.

b. Chris hopped across the car park.
c. Chris skated across the car park.

Examples in (22) demonstrate that one figure may employ several manners of motion
with distinctions present in terms of force dynamics such as speed, level of control,
impacts on physical aspects (e.g. tiredness), muscular effort, use of external objects, and
more. These distinctions may correspond to differing goals on the part of the figure.
Importantly too, a conceptualising observer may have given expectations relating to
motion performance based on the nature of the figure. Manners may be highly variable,
whilst at the same time, they are constrained by the physical properties of the moving
figure, e.g. pigs cannot fly, snakes cannot walk, and so forth. Relative to figure type then,
we may elaborate a broad manner-based classification of motion types, including (a)
typical, or default, motion (e.g. (22a) so long as Chris is a human figure), (b) atypical, or
forced, motion (e.g. (22b)), and (c) motion types requiring a form of support, instrument,
or vehicle, which may be referred to as instrumental instances of motion (e.g. (22c))
(see Pourcel 2004). According to this classification, default motion depicts the typical
manner of displacement for a given figure, e.g.

(23) a. The ball ROLLED across the lawn.

b. The bird FLEW out of the nest.
c. John WALKED home.
 MOTION: A CONCEPTUAL TYPOLOGY 429

Forced motion, on the other hand, involves atypical manners that may require a special
effort, impediment, or high degree of motor control for performance, e.g.

(24) a. The bird HOPPED across the motorway.

b. John LIMPED home.

Finally, instrumental motion depicts manners necessitating vehicles or extra elements

besides the figure’s body for motion, eg

(25) a. Tim SLEDGED down the slope.

b. We SAILED across the Mediterranean.

Interestingly, forced and instrumental motion are more commonly encountered in

instances of human motion, rather than animal, force or object motion.
Given that motion always involves a manner of displacement, we may thus draw an
analytical distinction between default, forced, and instrumental types of motion – based
on the manner of motion.

2.7 Summary

Motion types and the properties of motion-constituting components, such as paths and
manners, display great variability, overall. In fact, this variability is so great that widely
divergent situations are understood as instances of motion. At the same time, however,
not one instance of motion may be thought of as prototypical motion, or motion par
excellence. The rich variability of this domain in large part constitutes its complexity.
It may thus be useful to partition the domain of motion into types of motion, not so
much for the sake of analytical elegance, but because they may reflect distinct conceptual
representations in cognisers’ minds. The point is a simple one: motion is such a vast and
diverse domain that conceptualising motion cannot be a straight-forward and unitary
process, with the same schematic properties, e.g. path, receiving fixed levels of salience
in conceptualisation, regardless of the motion type. The nature of the motion scene must
in itself impact on the conceptualisation of its schematic properties, and this may be,
as suggested above, relative to directional, aspectual, causal, agentive, or force dynamic
properties of the scene, for instance.
Motion properties are not variable in a vacuum however. Instead, these properties
often appear to interact with and constrain other properties. For instance, motion
telicity constrains path types, cause-effect relations impact on path value, causality
necessitates figure animacy, figure type constrains agency as well as manner types,
and so on.
430 LANGUAGE, COGNITION AND SPACE

3 Figures as a basis for typological modelling

One central element of motion that seems to impact on most, if not all, types of motion
and motion schemas consists of the figure performing, or undergoing, the motion. I
will here use figure types as the basis for the elaboration of the conceptual typology,
because figure properties make possible and also constrain types of motion (and motion
properties). This section seeks to demonstrate how, for instance, the ‘existential status’
of figures defines the scope for motion types (e.g. Superman can fly across the Atlantic
in a few minutes but my neighbour cannot), and how figures’ physical properties
limit the range of manners of motion (e.g. sharks can swim but pigs cannot), and also
the types of path followed (e.g. some insects can naturally move on vertical surfaces
but humans cannot normally do so). In addition, only animate figures capable of
self-motion may cause the motion of external figures, so that figure type effectively
impacts on causal potential. In short, I will seek to demonstrate that conceptualisation
of motion, and possibly of other events, seems to be centered around figures, rather
than grounds, paths, manners or causal motivations. By identifying the figure as the
motion component determining other aspects of motion, I thereby suggest that any
conceptual model of the domain of motion should therefore be based on, or centered
around, the figure schema. In this section, I review how the properties of given figures
interact and determine the various other components of motion, including motion
types, as outlined in the previous section.

3.1 On figures and reality

Figure types may be distinguished relative to their ‘existential status’. Indeed, the ‘world
out there’ presents a perceptual reality as well as a fictional reality composed of fictional,
or artificial, entities pertaining to popular and personal imagination. In this world,
therefore, motion figures may be either real or fictional. Real figures, on the one hand,
have physical existence: they are perceptually real and may be physically interacted
with. They range across:
(i) humans, including babies, children, and adults, as well as their body parts;
(ii) animals, e.g. reptiles, birds, quadrupeds;
(iii) objects, e.g. bottles, bicycles, stools;
(iv) natural elements, e.g. water, fire;
(v) forces and currents, e.g. magnetism, electricity;
(vi) force-animated entities, e.g. planets, stars, currents, tornadoes.

Fictional figures, on the other hand, are man-made in the sense that they are created by
human minds. Fictional figures are not perceptually real therefore. Instead, they exist
in the world of fiction and may be found in myths, story books, cartoons, films, and
so on. Fictional figures therefore consist of virtual creations, or objects of popular and
individual imagination. Examples include:
 MOTION: A CONCEPTUAL TYPOLOGY 431

(vii) humans, e.g. Cinderella, Charlie Chaplin, Mary Poppins, the Incredibles;
(viii) human-like figures, e.g. Homer Simpson, Hulk, Hobbits, fairies, Michelin
Man;
(ix) human-like animals, e.g. Donald Duck, Nemo, Chicken Little, Bugs Bunny
(x) animals, e.g. Silvester the Cat, Coyote;
(xi) animal-like humans, e.g. werewolves, centaurs, Spiderman, Catwoman,
mermaids;
(xii) objects, e.g. Star Trek Enterprise spaceship, Cinderella’s mops and buckets;
(xiii) human-like objects, e.g. Thomas the Tank Engine, Christine the car, R2D2;
(xiv) artificial creations, e.g. Thing in the Addams family, Teletubbies, aliens,
monsters, Pacman;
(xv) folk superstition figures, e.g. gods, ghosts, witches, angels.

Numerous fictional figures involve the selective blending of characteristics pertaining

to different types of figures, e.g. human and animal, and to different representational
formats for these figures, e.g. film versus cartoon (Fauconnier and Turner 2002). For
instance, if we take the category of animal-like humans, we may note that mermaids
and centaurs blend the physical physiognomy of both humans and animals in equiva-
lent measure, whereas Spiderman and Catwoman retain the human physiognomy but
integrate physiological properties of animals, such as spider or feline eye sight, claws,
and so forth. Two points are interesting in these blends with regard to motion. First,
the blending of physical properties entails new possibilities for figure motion. These
possibilities may be enabling, such as Spiderman’s ability to climb up vertical structures.
They may also be constraining, such as mermaids’ inability to walk on solid ground. In
other words, the scope for motion types is redefined relative to the blended figure. The
second point of interest here is that within one category of fictional figure, we obtain
different selections of features in the resulting blend, as shown by the contrast between
centaurs and Catwoman. This differential selectivity means that fictional figures are
highly diverse and present a scope for motion types which is absent in the case of real-life
motion as performed by real figures.
In sum, figures capable of undergoing motion in space are wide-ranging and highly
diverse. A preliminary distinction may therefore be drawn between figure types that
are perceptually ‘real’ and those that are man-made, either virtually, or in popular and/
or individual imagination. From this distinction ensues one between real motion and
fictional motion.

3.2 On figures and manners of motion/ force dynamics

One of the most obvious aspects of motion which figure type determines concerns
likely manners of displacement. Real figures are constrained by their intrinsic physical
properties with respect to manners of motion, e.g. humans cannot fly, babies cannot
walk, dogs cannot tiptoe, birds cannot jog, balls cannot limp, books cannot roll, feathers
cannot pounce, winds cannot jump, and so on. Interestingly, human figures appear to
432 LANGUAGE, COGNITION AND SPACE

be the most flexible of all real figures and the most creative ones in terms of the types
of manners they may and do use for motion. Indeed, humans may walk, run, jump,
crawl, swim, dance, limp, and use instruments to drive, cycle, ice-skate, roller-blade, ski,
sail, paraglide, windsurf, fly, sledge, and so on. Animals, on the other hand, appear to
perform a much more limited range of manners of motion. This range typically excludes
instrumental motion and is very restricted in terms of forced types of manners. Even
our primate cousins do not seem capable of performing the multiplicity of manners of
motion humans can perform, and it takes years to train animals to perform controlled
manners, as is apparent from circus displays. Objects, natural elements and forces appear
even further restricted in their potential for manner variability.
Fictional figures, on the other hand, have seemingly infinite scope for man-
ners of motion. Most fictional figures have arguably become cultural icons with
figure-specific potential for manner types, and this potential is at once enabling and
restricting. This is very clear when contrasting figures pertaining to the same type.
Take, for instance, fictional figures that are human, such as Cinderella, Mary Poppins
and Charlie Chaplin. Only Cinderella seems to conform to prototypical real-life
human manners of motion. Mary Poppins, on the other hand, also has the capacity
to fly in the air. Finally, Charlie Chaplin’s manner of motion is characterised by his
duck-like walk, which prevents him from running or going up and down staircases
in a typical adult-human fashion.
In sum, manner types are more restricted and less diverse for non-human real
figures than for human figures, and they are almost infinite in scope for fictional
figures. This point is important to conceptualisation because it entails that cognisers
may have expectations regarding likely manners of displacement, and these expecta-
tions are more or less varied depending on the figure type. This means that motion
conceptualisation in the case of fictional figures, for instance, is likely to be different
in nature to that of real-life figures, as there may be no expectation concerning the
range of possible motion types. Even if there were any expectations, these may be
easily violated, given the unpredictable nature of fictional figures. In fact, one of the
delights and reasons for the success of fictional motion scenes as displayed in films and
cartoons is the very scope of possibilities for motion and the exploration of imaginary
scenarios. For instance, a number of successful characters in cinematography are
characterised by their non-veridical manners of motion, e.g. Superman, Wonder
Woman, Spiderman, the Incredibles, Batman, Peter Pan, Mary Poppins, Jumbo.
In fact, most fiction seems to rely on conceptual blends of figure properties, either
granting human figures animal-like properties, e.g. flying, or giving animals and
inanimates human properties, e.g. speech, bipedal motion, material culture, and so
forth (Turner 1996).

3.3 On figures and animacy

As mentioned in the previous section, types of figures may be further distinguished

relative to their animacy. Given the variety of figures outlined above, we may classify
as animate the following categories of figures:
 MOTION: A CONCEPTUAL TYPOLOGY 433

(xvi) real and fictional human and human-like

(xvii) real and fictional animal and animal-like
(xviii) fictional objects
(xix) fictional creations
(xx) natural elements and forces
(xxi) force-animated entities
Inanimate figures, on the other hand, comprise:
(xxii) real and fictional objects
The crucial distinction between animate and inanimate figures is that animates are
capable of initiating their own motion, and that of other figures, whereas inanimates
cannot perform self-motion. Inanimates may only undergo caused motion.

3.4 On figures and agency

Figure animacy has consequences for agency in motion. Indeed, agentive motion neces-
sitates an animate figure with the capacity for self-motion. This capacity is underpinned
by implicit instincts or explicit goals to either alter one’s location in space or to undergo
motion for the sake of it. In other words, agency entails some level of volition or inten-
tionality in instantiating and undergoing motion. Hence, only figure types (xvi)-(xix)
above may be agentive (e.g. humans, animals, fictional creations), but natural elements,
forces and force-animated entities in the universe do not display agentive motion, in
that sense, even though they may be capable of self-motion.
Given this understanding, non-agentive motion is therefore typical of non-sentient
entities, including inanimates and animates such as natural elements and forces. Note,
however, that non-agentive motion also occurs with animates when caused to move by
an external figure or force. In this case, motion is not intentional and thus non-agentive.
Consider, for instance:
(26) The ball rolled down the hill.
(27) The horse threw the rider flying off the saddle and into the air.

(26) illustrates an instance of inanimate object motion which is non-agentive. Indeed,

the ball rolled down the hill, not of its own agency, but because it was set into motion
by an external figure, such as an agentive human figure, or an external force, such as
the wind, gravity, or other. In (27), on the other hand, we have two animate figures in
motion; yet, only the primary figure, the horse, is agentive, whilst the secondary figure,
the rider, is unwittingly undergoing motion as a result, and is thus non-agentive in its
own motion.
Figure animacy may therefore be considered to correlate with properties of agency
in motion, in that only animate motion may generate agentive motion, whereas inani-
mate motion can only be non-agentive. It does not follow, however, that all animate
motion is agentive, as exemplified in (27). Likewise, the animate motion of natural
434 LANGUAGE, COGNITION AND SPACE

elements and forces is not agentive. In other words, animacy is necessary for agency,
but it is not sufficient. The agentive figure must also consist of a self-moving sentient
entity. Animacy and agency are thus distinct notions constrained by figure properties
and by the nature of the interactions between given figures.

3.5 On figures and causality

Figure animacy and agency have, in turn, consequences for causal properties in motion,
in that agentive figures only may cause the motion or change of state of other figures, as
in example (27) above. Also capable of causal motion are natural elements and forces.
However, non-agentive figures can only undergo caused motion, as in (26). In addition,
causality may be either overt, as in (27), where the primary motion-causing figure is
conceptually salient, or it may be covert, as in (26), where the primary agent is not
conceptually in focus, and yet it must be present for the motion act to take place. Note
that linguistic mappings often eclipse the causal dimension of inanimate motion, as in
(26), yet causality is physically real and is conceptually more or less overt as a result.
Because the agentive nature of the figure type impacts on (and constrains) causality,
the figure type also impacts on path values. In agentive self-motion, the notion of path
has an agentive value and it refers to spatial information, e.g.
(28) Jenny ran DOWN the hill.

The path is agentive in example (28) as it represents the path followed by the sole and
primary agentive figure in this motion act: Jenny effectively descended the hill.
In causal motion, on the other hand, the path becomes resultative and it may refer
to either spatial or stative information, e.g.
(29) The wind blew the napkin OFF the table.
(30) Mary tore the napkin TO SHREDS.

Indeed, in (29) and (30), the blowing and tearing actions exercised on the secondary
motion figure, the napkin, result in a change-of-location path in (29) and a change-of-
state path in (30).
In sum, the animate and agentive nature of the figure has a direct relation to causal
potential in motion, and to path value. Only animate figures may generate causal motion
together with the resultative path of a secondary figure.

3.6 On figures and directionality

As previously mentioned, motion may involve directionality or it may not. In the case
of motion events, the directionality, or path, is conceptually salient. All kinds of figures
undergo directed motion, either of their own doing, or as a result of a causal interaction
with an animate figure, e.g.
 MOTION: A CONCEPTUAL TYPOLOGY 435

(31) Mary walked to church this morning = animate and agentive motion
(32) The earth rotates around its axis = animate and non-agentive motion
(33) The book fell off the shelf = inanimate and non-agentive motion
(34) The door swung open = inanimate and non-agentive motion

On the other hand, non-directed motion, or motion activities, are more typical of
animate figures. Indeed, inanimate figures do not readily undergo motion activities.
Consider,
(35) Mary is driving = animate and agentive motion
(36) The wind is blowing = animate and non-agentive motion
(37) The buoy is floating = inanimate and non-agentive motion
(38) The ball is rolling = inanimate and non-agentive motion

Examples in (37) and (38) display instances of non-directed inanimate motion. These
types of example appear to be few, and to depend importantly on the physical proper-
ties of the moving object. Indeed, non-directed, ongoing motion may be undergone
by rolling items, for instance, but would be ad hoc with objects such as books, spoons,
or baskets. I suggest that activities are atypical of inanimate figures because inanimate
motion is by definition non-agentive, which means that inanimate motion must there-
fore be caused – whether overtly or covertly. As detailed above, caused motion typically
makes salient the path followed by the secondary figure. This path may be either a
change-of-location path, or a change-of-state path. In either case, the presence of a path
is characteristic of motion events, and not activities. In other words, the causality entailed
by inanimate motion leads to expectations of directionality. Conceptually, objects do not
undergo non-directed motion for the sake of it, in the fashion of animates. So, it appears
that motion activities are atypical with inanimate figures, though they are common in
the case of animate figures, whether agentive or not.

3.7 On figures and telicity

Telicity relates to the reaching of endpoints in space or to resultative states. Consider,

(39) Mike drove to the supermarket.
(40) Jane kicked the ball into the net.

In the case of agentive figures, telicity importantly correlates with the figure’s purpose
in undergoing motion. This purpose may be to arrive at a specific location, or to cause a
secondary figure to change its spatial or stative configuration. Given this understanding,
436 LANGUAGE, COGNITION AND SPACE

telic events typically entail agency. However, non-agentive animates may also generate
telic motion events, e.g.
(41) The fire spread across the entire forest.
(42) The wave engulfed the dinghy (under the water).

There is a correlation therefore between telic potential and animacy. Inanimates, on

the other hand, undergo telic motion events by virtue of being set into motion by an
animate figure, as in (40) and (42).

3.8 Summary

This section has argued that figure types determine motion types. The types of figures
identified correspond to properties of animacy and agency, as well as artificiality and
physicality. These fundamental properties constrain motion potential relative to aspects
of manners and force dynamics, causality, path value, directionality, and telicity. This
deterministic understanding of figure characteristics lends support to domain analyses
of motion based on figures, rather than on any other motion-constituting element.
Figures should therefore be the centre of any conceptual modelling of this domain. In
order to demonstrate this point more fully, I now propose to offer cognitive evidence
in support of the centrality of figures in motion conceptualisation.

4 Experimental evidence

Given the above-suggested distinctions concerning the diversity of motion types, the
research question now becomes whether this diversity is conceptually real. That is, do
these motion types correspond to actual conceptual categories of events? The aim of this
section and of the studies it includes is to offer preliminary cognitive evidence in support
of some of these distinctions in motion types. The evidence reviewed demonstrates that
types of figures, paths, manners, and other motion properties (e.g. causality) influence
event conceptualisation, and that they do so in non-random patterns of behaviour across
speakers of distinct languages. In other words, this section offers data in support of a
conceptual typology of motion which is valid regardless of the subject’s native language,
and which may therefore be utilised as a cross-linguistic tool for linguistic and cognitive
research, and as a metalanguage of analysis.

4.1 Method

The aim of the present research is to show that a motion of type x is conceptualised
as distinct from a motion of type y. In effect, the research seeks to show that motion
types x and y constitute distinct conceptual categories, or classes, of motion. Given
 MOTION: A CONCEPTUAL TYPOLOGY 437

this objective, the experimental approach tests event categorisation and does so by
using a sorting task in which subjects categorise motion stimuli together on the basis
of perceived similarity. The stimuli consist of event triads in digital video format. These
triads present a target motion stimulus, such as a man walking up a hill, followed by
two alternate events which resemble the target, yet differ in one variable, such as the
path or the manner of motion, as illustrated in (43).

(43) TARGET: a man walking up a hill

ALTERNATE 1: a man running up a hill (manner variable altered)
ALTERNATE 2: a man walking down a hill (path variable altered)

This triad example displays an animate, real-life, human figure performing instances
of self-motion in which the path is agentive and the manner is a default one for this
type of figure. Should we intend to test for possible effects of figure types on motion
conceptualisation, this type of triad would be used alongside other triads presenting
similar events performed by differing types of figures, such as inanimate or fictional
figures, e.g.

(44) TARGET: a ball rolling up a hill

ALTERNATE 1: a ball bouncing up a hill (manner variable altered)
ALTERNATE 2: a ball rolling down a hill (path variable altered)

The triad in (44) is comparable to the triad in (43) insofar as the figure is real, its
motion is seemingly independent from that of other figures, and the manner is of a
default type for a ball. The main difference between (43) and (44) thus concerns the
nature of the figure, which is human and animate in (43) and object and inanimate
in (44).
By using several triads of types (43) and (44) in one experiment, it is possible to see
whether subjects perform similar association choices for both types of triads (e.g. either
in terms of path or manner), or whether they perform differently – and consistently
so – for each triad type. Should subjects perform similarly, it may then be concluded
that figure type does not cause different conceptualisation of motion. However, should
subjects perform associations differently for the two types of triads, it may then be
concluded that they are conceptualising the two types of motion – human and object
motion – differently. In this latter case, we may infer that figure features cause distinctive
motion conceptualisation.
Likewise, should we wish to test for possible effects of manner types on motion
conceptualisation, triads of type (43) which displays default manners should be con-
trasted with triads displaying non-default manners, e.g.

(45) TARGET: a man limping up a hill

ALTERNATE 1: a man tiptoeing up a hill (manner variable altered)
ALTERNATE 2: a man limping down a hill (path variable altered)
438 LANGUAGE, COGNITION AND SPACE

Again, should default and non-default motion be conceptualised differently, we would

expect subjects to perform different categorisation choices for each triad type.
This experimental set-up was implemented in a number of studies to test the effects
of figures, paths, manners and agentive properties of motion on event conceptualisation.
Table 1 below details the properties of motion examined, together with the types and
sub-types contrasted in the triadic stimuli.
Table 1. Motion properties and types examined

Motion property Type Sub-type Example of motion situation

Figure Real Animate Human person
Inanimate Basket ball
Fictional Animate Virtual tomato

Motion agency Self-motion A person jumping over a wall

Causal motion A person kicking a door open
Path Atelic Along, up, down
Telic Across, into, out
Manner Default Rolling (ball), walking (person)
Non-default Forced Limping (person)
Instrumental Cycling (person)

Experiments were implemented using speakers of different native languages in order

to ensure that performance is independent of linguistic patterns, and to assess the
comparability and reliability of findings across different language populations. In the
exercise of extrapolating a conceptual typology, it was deemed particularly important
to obtain robust findings across language populations, rather than findings subject to
local fluctuations. Should categorisation behaviour be similar across different linguistic
populations, we may suggest that the categories obtained are strongly indicative of
universal trends in the human conceptualisation of motion.

4.1.1 Kopecka and Pourcel (2005, 2006)

Kopecka and Pourcel (2005, 2006) tested the effects of figure type on event conceptu-
alisation with 69 subjects representing the following native populations: Polish (N=24),
English (N=21) and French (N=24). As detailed in Table 1, three distinct figures were
used:

(i) ‘Tomatoman’ [- real] [+ animate] [- human]

(ii) Human person [+ real] [+ animate] [+ human]
(iii) Object ball [+real] [- animate] [- human]

The stimuli consisted of the now-famous virtual tomato designed by the Max Planck
Institute for Psycholinguistics. This tool represents a two-dimensional computer
animation with agentive properties, for instance, the tomato has eyes and it smiles,
and importantly it is capable of self-motion. Tomatoman corresponds to a fictional
 MOTION: A CONCEPTUAL TYPOLOGY 439

figure – though one that most subjects would not be familiar with and may therefore
not have a priori expectations for in terms of motion potential. In addition, the
stimuli included digital triads of a man performing motion instances, and other triads
presented the motion of a real-life plastic ball (yet note that no motion-causing agent
was visible for the performance of the ball’s motion).
Each triad contrasted two default manners of motion relative to the figure prop-
erties and two paths of displacement, as exemplified in triads (43) and (44) above.
Association choices were therefore made either in terms of path or manner similarity.
The results, shown in Graph 1, indicate that all three language groups – Polish, English
and French – performed very similarly in the task. This similarity suggests robust trends
for a universal level of conceptualisation. In addition, the results show that subjects
categorise the motion stimuli differently depending on the nature of the figure. The
virtual tomato, for instance, yields categorisation choices based on the manner variable
to an 80 percent extent. On the other hand, the object figure yields a 40 percent only
preference for manner, whilst the human figure averages a 34 percent preference for
manner. In other words, the fictional tomato figure prompts high manner salience in
conceptualisation, whereas the real-life figures prompt higher path salience, with path
salience being higher in the case of human figures than in the case of object figures
(Mann-Whitney U-test, pE=0.0004, pP=0.0002, pF=0.0002 for Tomato-Human scores;
pE=0.003, pP=0.005, pF=0.003 for Tomato-Ball scores). 4 The three figure types thus
trigger distinctive conceptual categorisation, with the most notable difference between
fictional and real-life figures – over 35 percent. We may therefore suggest that fictional
and real-life motion constitute distinct conceptual categories of events.

90
80
70
60
Polish
50
English
40
French
30
20
10
0
Tomato Ball Human

Graph 1. Proportions of manner association choices in the figure studies

In addition, these results may be explained in terms of the interaction between motion
variables. Indeed, on the one hand, fictional figures unknown to subjects (such as
virtual tomatoes) fail to have default manners of motion, that is, manners known to
and expected by the participant. Any manner of displacement is therefore likely to
be conceptually salient to the cogniser in the case of fictional motion. This salience
is likely to decrease when subjects are conceptualising the motion of known entities,
such as real-life objects and persons. On the other hand, the path of motion informs
us of the destination the figure eventually reaches. In the case of a human figure, that
destination is typically intended and corresponds to the figure’s goal, or purpose of
440 LANGUAGE, COGNITION AND SPACE

motion. In human motion, path thus conceptually correlates with the figure’s inten-
tions. This is so because human figures are animate and agentive. Their actions are
typically not random, but purposeful, and much of human interaction comes down
to deciphering other people’s goals and intentions. This much is part of the subject’s
knowledge and it may explain why subjects found path the most conceptually salient
variable in human motion. This categorisation bias towards the goal-loaded dimension
of motion is likely to decrease in cases where the figure is not intentioned, and where
paths do not correspond to actual goals. This may partly explain the low salience of
path in the conceptualisation of tomato motion (in addition to the non-default nature
of the manners displayed, as just discussed). This suggestion would also predict low
levels of path salience in the conceptualisation of object motion. However, the data
indicate a 60 percent bias towards path in object motion conceptualisation (at least
in the present experiment). In this case, it may be suggested that the real-life nature
of the figure caused the subjects to infer the presence of an agentive force behind the
motion of the objects shown (though that agent was not visible in the stimuli). Indeed,
plastic balls rarely – if ever – bounce across bridges and other types of ground of their
own volition. The inference of a causal element behind the object’s motion may have
rendered the object path relatively salient. In addition, note that the balls used in the
stimuli displayed default rolling and bouncing manners of motion, which might not
be expected to trigger manner salience as a result. These various factors may thus add
up to a relatively low level of manner salience across the findings.
In sum, these preliminary suggestions support the idea of a correlation between
real-life animacy and goal-directed behaviour (see also Mandler 2004). They also show
that there are complex interactions between motion variables in event conceptualisation.
For instance, the default nature of manners seems to decrease attention to manner in
favour of path, whereas unexpected and non-default manners direct attention towards that
variable to the detriment of path. The following study seeks to clarify the validity of these
suggestions by examining path and manner types, as well as causal relations more closely.

4.1.2 Pourcel (2004, 2005)

Pourcel (2004, 2005) tested the effects of path type, manner type and causality on event
conceptualisation with 69 subjects representing the following native populations: English
(N=34) and French (N=35). The stimuli consisted of fifteen video triads using a human
figure to perform motion instances.
As detailed in Table 1, two types of path were displayed in the stimuli:

(i) Atelic paths showing no trajectory endpoint or crossed boundaries,

(ii) Telic paths showing explicit endpoints or crossed boundaries.

Three types of manners were displayed:

(iii) Default manners, such as walking or running casually,

(iv) Non-default manners showing heightened degrees of control for perform-
ance, or an impediment, such as tiptoeing or limping,
 MOTION: A CONCEPTUAL TYPOLOGY 441

(v) Non-default manners showing the use of a vehicle for the performance of
motion, such as a bicycle or a scooter.

Finally, the stimuli showed instances of self-motion and of causal motion:

(vi) Self-motion showing the intentional motion of an agentive figure,

(vii) Causal motion showing the motion of an agentive figure generating the
motion of a non-agentive figure, such as an object.

As in the previous experiments, each triad contrasted two manners of motion and two
paths of displacement. Association choices were therefore made either in terms of path
or manner similarity. The results, shown in Graphs 2, 3 and 4, indicate that the two
language groups performed very similarly in the task. This similarity is again suggestive
of universal trends in motion conceptualisation. In addition, the graphs show that
subjects categorise the motion stimuli differently depending on the nature of the path,
manner, and causal properties displayed in the scenes.
In the case of path properties, Graph 2 illustrates how the presence of a path end-
point encourages path-salient conceptualisation of events (almost 70 percent of choices
were made relative to path similarity). On the other hand, the absence of telicity does
not reverse path salience in favour of manner salience. Instead, it reduces path-salient
conceptualisation, and we obtain mixed performance in conceptualisation in atelic
cases, with other factors likely to interfere with variable salience, such as manner type
and causal relations. What is apparent, then, from Graph 2 is a clear difference in the
conceptualisation of telic and atelic events – a 20 percent difference in associative
performance (Wilcoxon test, pE=0.001, pF<0.001). We may suggest therefore that the
two types of events are conceptually distinct.
80%

60%
English
40%
French
20%

0%
telic atelic
Graph 2. Proportions of path association choices relative to path telicity

In the case of manner properties, Graph 3 shows that default manners of motion prompt
path-salient conceptualisation of events (76 percent of choices overall were made relative
to path similarity). On the other hand, non-default manners, involving either higher
control, impediment, or vehicle, reduce path salience by about 20 percent. This differ-
ence in associative performance is indicative of differential conceptualisation across
default-manner motion and non-default instances (Wilcoxon test, pE=0.001, pF=0.0001
for a comparison of default and forced scores; pE<0.0001, pF=0.001 for default and
442 LANGUAGE, COGNITION AND SPACE

instrumental scores). Note, however, that the distinction is marginal between forced
and instrumental manner types (Wilcoxon test, pE=0.069, pF=0.781). We may thus
posit two broad types of manners that are conceptually distinct, namely default and
non-default types.
0.8
0.7
0.6
0.5
English
0.4
French
0.3
0.2
0.1
0
Default Forced Instrumental

Graph 3. Proportions of path association choices relative to manner type

In the case of causal properties, Graph 4 demonstrates a slight difference in associative

performance. On average, 65 percent of caused motion instances prompt path salience,
in contrast to 58 percent of self-motion instances. We may suggest therefore a tendency
towards path salience when causal relations are apparent between the motion of two or
more figures. In addition, note that this analysis does not sort responses relative to path
and manner types, which suggests that differences between self- and caused motion may
be more pronounced in more controlled conditions, for instance, in conditions where
all manners were default and all paths were telic, and where the sole contrast between
the triads would be relative to causality. Although the score differences are not quite
as marked as in previous tests (Wilcoxon test, pE=0.067, pF=0.029), the preferential
tendency for path salience in caused motion is obvious when we contrast the differences
between path and manner associations in each type of triad. This difference averages
15 percent in self-motion and 30 percent in caused motion.
0.7

0.6

0.5
English - path
0.4 French - path
0.3 English - manner
French - manner
0.2

0.1

0
Self-motion Caused motion

Graph 4. Proportions of path and manner association choices relative to causality

 MOTION: A CONCEPTUAL TYPOLOGY 443

The above analyses do not isolate each motion variable neatly and are therefore likely
to be skewed by the effect of other variables. These graphs are thus suggestive of generic
trends. These trends make it possible to predict conceptual salience for particular types of
motion, based on path telicity, manner defaultness, and motion causality. It has so far been
demonstrated that telicity, default manners and causal relations prompt higher degrees of
path salience, than atelicity, non-default manners and instances of self-motion. We may
therefore hypothesise the following levels of conceptual salience for the following events:

(i) [+telicity] [-causality] [+default] = high path salience

(ii) [+telicity] [+causality] [+default] = high path salience
(iii) [-telicity] [-causality] [+default] = high path salience (though lower
than in (i) and (ii))5
(iv) [+telicity] [-causality] [-default] = mixed path and manner salience
(v) [+telicity] [+causality] [-default] = mixed path and manner salience
(vi) [-telicity] [-causality] [-default] = low path salience

Subject responses were analysed according to these predictions. The analysis supports
these predictions, as shown in Graph 5.

0.9
0.8
0.7
0.6
0.5 Manner
0.4 Path
0.3
0.2
0.1
0
(i) +Telicity (ii)+Telicity (iii)-Telicity (iv)+Telicity (v)+Telicity (vi)-Telicity
-Cause +Cause -Cause -Cause +Cause -Cause
+Default +Default +Default -Default -Default -Default

Graph 5. Proportions of path and manner association choices relative to telicity, causality, and default manner
properties6

Graph 5 presents a more fine-grained analysis of the data, and in so doing, it usefully
illustrates the scope of conceptual variability across different motion types. For instance,
path is more salient than manner to an 85 percent extent in motion events of type (i),
such as a walking-across-a-road event, whereas that salience drops by 50 percent down
to 38 percent in motion events of type (vi), such as a tiptoeing-up-a-staircase event.
We may therefore conclude that the nature of motion properties impacts in significant
ways on the conceptualisation of events.
The conceptual distinctions observed may be explained in similar terms to the
ones highlighted in the discussion of the figure studies by Kopecka and Pourcel (2005,
444 LANGUAGE, COGNITION AND SPACE

2006). That is, differential conceptualisation of events may be explained in terms of

(a) the cogniser’s expectations of likely manners of displacement relative to a given
figure, and (b) the explicitness of the figure’s goals and intentions in performing
motion. Indeed, default manners of motion attract little attention to the manner
variable, as illustrated in motion types (i)-(iii), whilst the overtness of the figure’s
locational goal in telic events renders the path dimension of motion particularly
salient, as exemplified in motion types (i)-(ii) in particular. Likewise, causality entails
an intention on behalf of the agentive figure to achieve a particular result, and as such
it denotes a purposeful action akin to the notion of telicity. Finally, it is interesting
to note that these properties interact in the overall conceptualisation of an event, so
that telic events performed with a non-default manner of motion (e.g. types (iv) and
(v)) yield mixed path-manner biases in conceptualisation. The conceptualisation
of motion events is thus neither straightforward nor unitary, but highly dynamic
and dependent upon discrete properties of motion. Motion conceptualisation is not
random, however, but consistent and predictable, and it may be explained in terms
of perceived intentionality and expectations based on the cogniser’s knowledge of the
world, as suggested above. It is equally noteworthy to recall that performance on the
present cognitive tasks was strikingly similar across all 69 subjects, including subjects
of different native languages. The present conceptual trends may thus be indicative of
universal tendencies in human cognition.

4.1.3 Summary

This section asked whether conceptualisation behaviour supports the distinctions and
categories proposed earlier in this article. The present experiments examined a number
of motion properties, including figure animacy, figure reality, path telicity, manner
defaultness, motion agency and causality. Results indicate that each of these fundamental
property influences motion conceptualisation in significant ways. We may suggest, as a
result, that the motion types discussed constitute distinct conceptual classes of events.
These conceptual classes include 7:
real-life motion
animate motion
inanimate motion
fictional motion
telic motion
atelic motion
default motion
non-default motion
self-motion
causal motion
 MOTION: A CONCEPTUAL TYPOLOGY 445

Path- and manner-based conceptual distinctions in event classification between these

types of motion are summarised in Graph 6.

90%

80%

70%

60%

PATH
50%
MANNER
40%

30%

20%

10%

0%
fictional atelic non- self object causal human telic default
motion motion default motion motion motion motion motion motion
motion

Graph 6. Proportions of path and manner association choices relative to motion type

5 Sketching a conceptual typology of motion

This study has reported preliminary evidence for conceptual distinctions between
real-life and fictional figures
animate and inanimate figures
default and non-default manners of figure motion
telic and atelic events relating to figure goals
self-initiated and caused motion

These motion types are based on properties relating to the figure, including basic exis-
tential properties (e.g. real/ fictional, animate/ inanimate), physical properties (e.g.
bodily capacity for performing given manners, relative strength for causing the motion
of other figures), and cognitive properties (e.g. goals and intentions). It appears that
motion is fundamentally about the figure that performs or undergoes it. Motion may
thus be modelled as a figure-centred domain, and I propose to devise a figure-centred
conceptual typology for the domain of motion, as a result. Table 2 offers a summary of
the possible types of motion, based on figure properties.
446 LANGUAGE, COGNITION AND SPACE

Table 2. Typological modelling of motion, based on figure properties.

FIGURE TYPE MOTION TYPE

Animacy Agency Figure type Motion Path Directionality Telicity Force
dynamics
animate
R Agentive
E
human self / causal agentive events / activities telic / default /
A
atelic non-default
L
animal self / causal agentive events / activities telic / default /
atelic non-default
F
I non-agentive
G natural self / causal/ mechanical/ events / activities telic / default /
U element caused resultative atelic non-default
R human caused resultative events telic default /
E non-default
S animal caused resultative events telic default /
non-default
inanimate
object caused resultative events telic Default

animate
Agentive
F human self / causal agentive events / telic / default /
I activities atelic non-default
C
human-like self / causal agentive events / telic / default /
T
activities atelic non-default
I
animal self / causal agentive events / telic / default /
O
activities atelic non-default
N
A animal-like self / causal agentive events / telic / default /
L activities atelic non-default
non-agentive
F natural elementself / causal mechanical/ events / telic / Default
I resultative activities atelic
G human caused resultative events telic default /
U non-default
R human-like caused resultative events telic default /
E non-default
S
animal caused resultative events telic default /
non-default
animal-like caused resultative events telic default /
non-default
inanimate
object caused resultative events telic Default
 MOTION: A CONCEPTUAL TYPOLOGY 447

6 Conclusion

This research was initially triggered by methodological concerns in relativistic inves-

tigations of the relationship between language and cognition. Paramount among
these concerns is the need to avoid linguacentric analyses of experiential domains.
Motion, one of the most extensively used domain in contemporary relativistic research,
remains to this day largely characterised in linguistic and typological terms. That is,
little linguistic research has sought to acquire a conceptual understanding of this
domain and to develop a neutral metalanguage for the analysis of motion and for
the application of these analyses. This research does not pretend to develop one such
metalanguage. Rather, it seeks to direct awareness towards the necessity to develop
conceptual typologies and analytic metalanguages systematically in answer to calls for
scientific rigour in multi-disciplinary research (e.g. Lucy 1992, 2003). In so doing, it
is hoped that this research will trigger further inquiries into the conceptualisation of
motion, which may in turn be of use to linguistic applications. Such inquiries would
be useful in terms, for instance, of additional experimental research bringing to the
fore new cognitive data to establish the reality of the categorial distinctions suggested.
Indeed, the present research has offered data in support of a certain number of distinc-
tions. More data is needed to substantiate the distinction between event and activity,
for instance. According to the analysis reported in this article, experimental hypotheses
would predict the following in the case of human figures: activities generate manner
salience, whereas events generate path salience. In addition, more data is required to
support the distinction between caused and self-motion events, as well as between
further motion types based on figure characteristics, e.g. animal vs. human motion.
Data is also needed to reach a better understanding of the interactions between motion
variables in conceptualisation. It was suggested that motion variables do not occur
in a vacuum, but instead, interact and constrain other variables in systematic ways,
which are far from transparent at this juncture. Cross-linguistic data is also required
to offer conclusiveness to the conceptual typology proposed, so far only supported
by data from English, Polish and French subjects. Further support is also required
to fully substantiate the suggestion that intentionality, for instance, is an explanatory
factor for path salience in human motion conceptualisation. Finally, it would also be
key to explore domain extensions and their conceptual reality, e.g. fictive motion, or
the domain of time, and so forth.

References
Aske, J. (1989) Path predicates in English and Spanish: A closer look. Proceedings of
the Fifteenth Annual Meeting of the Berkeley Linguistics Society 1–14.
Fauconnier, G. and Turner, M. (2002) The Way We Think: Conceptual Blending and
the Mind’s Hidden Complexities. New York: Basic Books.
Gennari, S. P., Sloman, S. A., Malt, B. C. and Fitch, W. T. (2002) Motion events in
language and cognition. Cognition 83: 49–79.
448 LANGUAGE, COGNITION AND SPACE

Hohenstein, J. (2005) Language-related motion event similarities in English- and

Spanish-speaking children. Journal of Cognition and Development 6: 403–425.
Kopecka, A. and Pourcel, S. (2005) Figuring out figures’ role in motion conceptuali-
sation. Paper presented at the 9th International Cognitive Linguistics Conference,
Seoul, Korea.
Kopecka, A. and Pourcel, S. (2006) Understanding figures in conceptualising
motion. Paper presented at the 2nd Biennial Conference on Cognitive Science, St
Petersburg, Russia.
Lucy, J. A. (1992) Language Diversity and Thought. Cambridge: Cambridge University
Press.
Lucy, J. A. (2003) Semantic accent and linguistic relativity. Manuscript.
Mandler, J. (2004) The Foundations of Mind. Oxford: Oxford University Press.
Papafragou, A., Massey, C. and Gleitman, L. (2002) Shake, rattle, ‘n’ roll: The repre-
sentation of motion in language and cognition. Cognition 84: 189–219.
Pourcel, S. (2004) Motion in language and cognition. In A. Soares da Silva, A. Torres
and M. Gonçalves (eds) Linguagem, Cultura e Cognição: Estudos de Linguistica
Cognitiva Vol. 2 75–91. Coimbra: Almedina.
Pourcel, S. (2005) Relativism in the linguistic representation and cognitive concep-
tualisation of motion events across verb-framed and satellite-framed languages.
Unpublished doctoral dissertation, University of Durham, UK.
Pourcel, S. (2009a) Relativistic application of thinking for speaking. In J. Guo,
E. Lieven, N. Budwig, S. Ervin-Tripp, K. Nakamura and S. Özçalişkan (eds)
Crosslinguistic Approaches to the Study of Language: Research in the Tradition of
Dan Isaac Slobin 493–503. Hove: Psychology Press.
Pourcel, S. (2009b) Motion scenarios in cognitive processes. In V. Evans and S.
Pourcel (eds) New Directions in Cognitive Linguistics 371–391. Amsterdam: John
Benjamins.
Pourcel, S. and Kopecka, A. (2005) Motion expression in French: Typological diver-
sity. Durham and Newcastle Working Papers in Linguistics 11: 139–153.
Pourcel, S. and Kopecka, A. (To appear) Motion events in French: Typological
intricacies. Linguistic Typology.
Slobin, D. I. (1996) From ‘thought and language’ to ‘thinking for speaking’. In
J.J. Gumperz and S.C. Levinson (eds) Rethinking Linguistic Relativity 70–96.
Cambridge: University Press.
Slobin, D.I. (2000) Verbalised events: A dynamic approach to linguistic relativity
and determinism. In S. Niemeier and R. Dirven (eds) Evidence for Linguistic
Relativity 107–138. Amsterdam: John Benjamins.
Slobin, D. I. (2004) The many ways to search for a frog: Linguistic typology and the
expression of motion events. In S. Strömqvist and L. Verhoeven (eds) Relating
Events in Narrative Vol. 2 219–257. Mahwah, NJ: LEA.
Talmy, L. (1985) Lexicalisation patterns: Semantic structure in lexical forms. In
T. Shopen (ed.) Language Typology and Syntactic Description Vol. 3 57–149.
Cambridge: Cambridge University Press.
Talmy, L. (1991) Path to realisation: A typology of event conflation. Proceedings of the
Seventeenth Annual Meeting of the Berkeley Linguistics Society 480–520.
 MOTION: A CONCEPTUAL TYPOLOGY 449

Talmy, L. (2000) Toward a Cognitive Semantics: Typology and Process in Concept

Structuring Vol. 2. Cambridge, MA: MIT Press.
Turner, M. (1996) The Literary Mind. Oxford: Oxford University Press.
Whorf, B.L. (1956) Language, Thought, and Reality. Cambridge, MA: MIT Press.
Zlatev, J. and David, C. (2004) Do Swedes and Frenchmen view motion differently?
Conference Presentation, Language, Culture and Mind, Portsmouth.
Zlatev, J. and David, C. (2005) Motion event typology and categorisation. Conference
Presentation, International Cognitive Linguistics Conference, Seoul.

Notes
1 Note again that the examples here and below seek to represent event situations. The
examples, unless otherwise stipulated, are not linguistic examples. Instead they seek
to prompt a conceptual image of a particular event. It is in this non-linguistic frame of
mind that examples are used and discussed in this article.
2 Note that this point merely serves as an illustration, but is not used as an argument
for the conceptual distinctiveness of motion events vs. motion activities. As outlined
before, a conclusion of this type would be linguacentric, not to mention that it would
also ignore the plethora of satellite-framed languages which cannot do away with
manner verbs, regardless of the conceptual salience of the manner and path schemas in
the said events.
3 Note that the latter two entities may be posited as being animate despite lacking
intentionality. The contrast between animacy and intentionality is discussed further in
the next section on agency.
4 The initial following the p score corresponds to the language group being tested. E for
English, P for Polish and F for French.
5 Note that the present stimuli did not include atelic caused motion. Such motion events
exist nonetheless, e.g. pushing a pram along the pavement. The prediction may be one of
high path salience, though lower than that expected in (ii) for instance, due to the lack
of telicity.
6 Note that due to the high comparability of the French and English data, responses from
the two groups were conflated in this analysis, for clarity of presentation.
7 Note that the list is by no means exhaustive, but is based on the experimental findings
reported in this article. Additional classes may include artificial animate motion, arti-
ficial inanimate motion, real-life animal motion, real-life animal non-default motion
(e.g. motion as performed by animals in circuses), human sports motion (as opposed to
everyday motion), force motion, and so forth.
 Part VIII
The relation between space, time and modality

451
17 Space for thinking
Daniel Casasanto

1 Introduction

How do people think about things they can never see or touch? The ability to invent
and reason about domains such as time, ideas, or mathematics is uniquely human, and
is arguably the hallmark of human sophistication. Yet, how people mentally represent
these abstract domains has remained one of the mysteries of the mind. This chapter
explores a potential solution: perhaps the mind recruits old structures for new uses.
Perhaps sensory and motor representations that result from physical interactions with
the world (e.g., representations of physical space) are recycled to support abstract
thought. This hypothesis is motivated, in part, by patterns observed in language: in
order to talk about abstract things, speakers often recruit metaphors from more con-
crete or perceptually rich domains. For example, English speakers often talk about
time using spatial language (e.g., a long vacation; a short meeting). Cognitive linguists
have argued such expressions reveal that people conceptualize abstract domains like
time metaphorically, in terms of space (see Lakoff and Johnson, 1999; c.f., Evans,
2004). Although linguistic evidence for Metaphor Theory is abundant, the necessary
nonlinguistic evidence has long been elusive; people may talk about time using spatial
words, but how can we know whether people really think about time using mental
representations of physical space?
This chapter describes a series of experiments that evaluate Metaphor Theory as an
account of the evolution and structure of abstract concepts and explore relations between
language and nonlinguistic thought, using the abstract domain of time and the relatively
concrete domain of space as a testbed. Hypotheses about the way people mentally rep-
resent space and time were based on patterns in metaphorical language, but were tested
using simple psychophysical tasks with nonlinguistic stimuli and responses. Results of
the first set of experiments showed that English speakers incorporate irrelevant spatial
information into their estimates of time (but not vice versa), suggesting that people
not only talk about time using spatial language, but also think about time using spatial
representations. The second set of experiments showed that (a) speakers of different
languages rely on different spatial metaphors for duration, (b) the dominant metaphor
in participants’ first languages strongly predicts their performance on nonlinguistic time
estimation tasks, and (c) training participants to use new spatiotemporal metaphors
in language changes the way they estimate time. A final set of experiments extends the
experimental techniques developed to explore mental representations of time to the
domain of musical pitch. Together, these studies demonstrate that the metaphorical
language people use to describe abstract ideas provides a window on their underlying
mental representations, and also shapes those representations. The structure of abstract

453
454 LANGUAGE, COGNITION AND SPACE

domains such as time appears to depend, in part, on both linguistic experience and on
physical experience in perception and motor action.

1.1 Time as an abstract domain

For what is time? Who can readily and briefly explain this? Who can even in thought
comprehend it, so as to utter a word about it?

If no one asks me, I know: if I wish to explain it to one who asketh, I know not.

Saint Augustine, Confessions, Book 11

How long will it take you to read this chapter? The objective time, as measured by the
clock, might depend on whether you’re scrutinizing every detail, or just skimming to
get the main ideas. The subjective time might vary according to physiological factors
like your pulse and body temperature (Cohen, 1967; Ornstein, 1969), psychological
factors like how much the text engages your interest and attention (Glicksohn, 2001;
James, 1890; Zakay and Block, 1997), and some surprising environmental factors like
the size of the room you’re sitting in (DeLong, 1981).
Although subjective duration is among the earliest topics investigated by experi-
mental psychologists (Mach, 1886), the cognitive sciences have yet to produce a com-
prehensive theory of how people track the passage of time, or even to agree on a set of
principles that consistently govern people’s duration estimates. An excerpt from a review
by Zakay and Block (1997) illustrates the current state of confusion:

People may estimate filled durations as being longer than empty durations, but
sometimes the reverse is found. Duration judgments tend to be shorter if a more
difficult task is performed than if an easier task is performed, but again the opposite
has also been reported. People usually make longer duration estimates for complex
than for simple stimuli, although some researchers have found the opposite. (pg. 12)

What makes time perception so difficult to understand? Ornstein (1969) argues that
although we experience the passage of time, the idea that time can be perceived through
the senses is misleading (cf. Evans, 2004):

One major reason for the continuing scattering of [researchers’] effort has been
that time is treated as if it were a sensory process. If time were a sensory process
like vision…we would have an ‘organ’ of time experience such as the eye. (pg. 34)

Although time is not something we can see or touch, we often talk about it as if it were
(Boroditsky, 2000; Clark, 1973; Gruber, 1965; Jackendoff, 1983; Lakoff and Johnson,
1980). Consider the following pair of sentences:
 SPACE FOR THINKING 455

i) They moved the truck forward two meters.

ii) They moved the meeting forward two hours.

The truck in sentence i is a physical object which can move forward through space,
and whose motion we might see, hear, or feel, from the staring point to the ending
point. By contrast, there is no literal motion described in sentence ii. The meeting is
not translated through space, and there is no way to experience its ‘movement’ through
time via the senses. Events that occur in time are more abstract than objects that exist
in space insomuch as we typically have richer perceptual evidence for the spatial than
for the temporal.1
In this chapter, I will argue that (a) the language people typically use to talk about
duration reveals important links between the abstract domain of time and the relatively
concrete domain of space, (b) people use spatial representations to conceptualize time
even when they’re not using language, and (c) although the domains of space and time
provide a particularly useful testbed for hypotheses about the evolution and structure
of abstract concepts, time is only one of many abstract domains of knowledge that
depend, in part, on perceptuo-motor representations built up via experience with the
physical world.

1.2 Metaphor and the problem of abstract thought

The mystery of how people come to mentally represent abstract domains such as time,
ideas, or mathematics has engaged scholars for centuries, sometimes leading to proposals
that seem unscientific by modern standards. Plato (Meno, ca. 380 B.C.E.) argued that we
cannot acquire abstract concepts like virtue through instruction, and since babies are not
born knowing them, it must be that we recover such concepts from previous incarnations
of our souls. Charles Darwin contended that evolution can explain the emergence of
abstract thought without recourse to reincarnation, yet it is not immediately obvious
how mental capacities that would have been superfluous for our Pleistocene forebears
could have been selected for. What selection pressures could have resulted in our ability
to compose symphonies, invent calculus, or imagine time travel? How did foragers
become physicists in an eyeblink of evolutionary time? The human capacity for abstract
thought seems to far exceed what could have benefited our predecessors, yet natural
selection can only effect changes that are immediately useful. The apparent superfluity
of human intelligence drove Alfred Wallace, Darwin’s co-founder of the theory of
evolution by natural selection, to abandon their scientific theory and invoke a divine
creator to explain our capacity for abstract thought (Darwin, 1859/1998, 1874/1998;
Gould, 1980; Pinker, 1997; Wallace, 1870/2003).2
Darwin’s own formulation of evolutionary theory points toward an elegant potential
solution to Wallace’s dilemma: sometimes organisms recycle old structures for new uses.
An organ built via selection for a specific role may be fortuitously suited to perform other
unselected roles, as well. For example, the fossil record suggests that feathers were not
456 LANGUAGE, COGNITION AND SPACE

originally ‘designed’ for flying. Rather, they evolved to regulate body temperature in small
running dinosaurs, and were only later co-opted for flight (Gould, 1991). The process
of adapting existing structures for new functions, which Darwin (1859/1993) gave the
misleading name preadaptation, was later dubbed exaptation by evolutionary biologist
Steven Jay Gould and colleagues (1982). Gould argued that this process may explain the
origin of many biological and psychological structures that direct adaptation cannot.
Are abstract concepts like dinosaur feathers? Can exaptation account for mental
abilities in humans that could not have been selected for directly? If so, how might this
have happened: which adapted capacities might abstract domains be exapted from?
Steven Pinker (1997) sketched the following proposal:

Suppose ancestral circuits for reasoning about space and force were copied, the
copies’ connections to the eyes and muscles were severed, and references to the
physical world were bleached out. The circuits could serve as a scaffolding whose
slots are filled with symbols for more abstract concerns like states, possessions,
ideas, and desires. (pg. 355)

As evidence that abstract domains arose from circuits designed for reasoning about
the physical world, Pinker appeals to patterns observed in language. Many linguists
have noted that when people talk about states, possessions, ideas, and desires, they do
so by co-opting the language of intuitive physics (Clark, 1973, Gibbs, 1994; Gruber,
1965; Jackendoff, 1983; Lakoff and Johnson, 1980; Langacker, 1987; Talmy, 1988). In
particular, words borrowed from physical domains of space, force, and motion, give rise
to linguistic metaphors for countless abstract ideas. For each pair of expressions below,
l illustrates a literal use and m a metaphorical use of the italicized words.

1l a high shelf
1m a high price

2l a big building
2m a big debate
3l forcing the door
3m forcing the issue
4l pushing the button
4m pushing the limit
5l keeping the roof up
5m keeping appearances up

The concrete objects described in the literal sentences (e.g., shelf, building, door, button,
roof) belong to a different ontological category than the abstract entities in the meta-
phorical examples, according a test of what physical relations they can sensibly be said
to enter into. For example, it is sensible to say ‘the cat sat on the shelf / building / door
 SPACE FOR THINKING 457

/ button / roof ’, but it may not be sensible to say that ‘the cat sat on the price / debate /
issue / limit / appearance’. This test is similar to a test of sensible predicates for concrete
vs. abstract entities devised by Fred Sommer (1963; cf., Turner, 2005).
Based on examples like these, linguists have argued that people create abstract
domains by importing structure from concepts grounded in physical experience.
Although anticipated by others (e.g., Lafargue, 1898/1906), this idea appears to have
been first articulated as the Thematic Relations Hypothesis (TRH) in 1965, by Jeffery
Gruber. TRH was later elaborated by Jackendoff (1972; 1983) who wrote:

The psychological claim behind [Gruber’s linguistic discovery] is that the mind
does not manufacture abstract concepts out of thin air…it adapts machinery that
is already there, both in the development of the individual organism and in the
evolutionary development of the species. (1983, pg. 188–9)

Not all theorists agree on the significance of metaphorical language for theories of mental
representation. Gregory Murphy (1996; 1997) raised concerns about both the vagueness
of the psychological processes suggested by linguists and about the limitations of purely
linguistic evidence for metaphorical conceptual structure. Murphy (1996) proposed
that linguistic metaphors may merely reveal similarities between mental domains: not
causal relationships. Across languages, people may use the same words to talk about
space and time because these mental domains are structurally similar, and are therefore
amenable to a common linguistic coding. He argued that in the absence of corroborat-
ing nonlinguistic evidence, his Structural Similarity proposal should be preferred on
grounds of simplicity. His view posits that all concepts are represented independently,
on their own terms, whereas the metaphorical alternative posits complex concepts that
are structured interdependently. It is evident that people talk about abstract domains
in terms of relatively concrete domains, but do they really think about them that way?

1.3 From conceptual metaphor to mental metaphor

The idea that conventionalized metaphors in language reveal the structure of abstract
concepts is often associated with Conceptual Metaphor theory, proposed by linguist
George Lakoff and philosopher Mark Johnson (1980, 1999). Lakoff and Johnson
described ‘conceptual metaphors’ as one of ‘three major findings of cognitive science’
(1999, pg. 3). Yet, their claim that people think metaphorically was supported almost
entirely by evidence that we talk metaphorically. Despite the impressive body of lin-
guistic theory and data that Lakoff and Johnson summarized (and the corroborating
computational models of word meaning), they offered little evidence that the importance
of metaphor extends beyond language. In the absence of nonlinguistic evidence for
metaphorically structured mental representations, the idea that abstract thought is an
exaptation from physical domains remained ‘just an avowal of faith’ among scientists
who believe that the mind must ultimately be explicable as a product of natural selection
(Pinker, 1997, pg. 301).
458 LANGUAGE, COGNITION AND SPACE

The term ‘conceptual metaphor’ is used ambiguously, sometimes to refer to patterns

in language, and other times to nonlinguistic conceptual structures that are hypothesized
to underlie these patterns in language. To avoid this ambiguity, I will refer to patterns
in language as linguistic metaphors and to the hypothesized nonlinguistic metaphorical
structures in the mind as mental metaphors (Casasanto, 2008, 2009a). This termino-
logical shift allows several critical questions to be framed clearly. Part 1 of this chapter
will address the question, ‘Do people use mental metaphors that correspond to their
linguistic metaphors in order to conceptualize abstract domains, even when they’re
not using language?’ Part 2 asks, ‘If so, do people who tend to use different linguistic
metaphors also rely on different mental metaphors?’ and further, ‘Does using different
linguistic metaphors cause speakers of different languages to rely on different mental
metaphors?’ Finally, distinguishing linguistic metaphors from mental metaphors allows
us to pose other questions that lie beyond the scope of this chapter (see Casasanto, 2008,
2009a, 2009b), such as, ‘Are there any mental metaphors for which no corresponding
linguistic metaphors exist?’ This question has received virtually no attention from
linguists or psychologists. This could be due, in part, to the fact that it is nonsensical
when phrased in the traditional terminology: ‘Are there any conceptual metaphors for
which no corresponding conceptual metaphors exist?’ Whereas Conceptual Metaphor
theorists treat patterns in language as a source of evidence that people think metaphori-
cally, the research presented here takes patterns in language as a source of hypotheses
about conceptual structure.

1.3 Experimental evidence for mental metaphors

Boroditsky (2000) conducted some of the first behavioral tests of the psychological
reality of mental metaphors. Her tasks capitalized on the fact that in order to talk about
spatial or temporal sequences, speakers must adopt a particular frame of reference.
Sometimes we use expressions that suggest we are moving through space or time (e.g.,
we’re approaching Maple Street; we’re approaching Christmas). Alternatively, we can use
expressions that suggest objects or events are moving with respect to one another (Maple
Street comes before Elm Street; Christmas comes before New Year’s). In one experiment,
Boroditsky found that priming participants to adopt a given spatial frame of reference
facilitated their interpretation of sentences that used the analogous temporal frame of
reference. Importantly, the converse was not found: temporal primes did not facilitate
interpreting spatial sentences. This priming asymmetry parallels a well established asym-
metry in linguistic metaphors: people talk about the abstract in terms of the concrete
(e.g., time in terms of space) more than the other way around (Lakoff and Johnson,
1980). Based on these results Boroditsky proposed a refinement of Conceptual Metaphor
Theory, the Metaphoric Structuring View, according to which (a) the domains of space
and time share conceptual structure, and (b) spatial information is useful (though not
necessary) for thinking about time. A second set of experiments showed that real-world
spatial situations (e.g., riding on a train, or standing in a cafeteria line) and even imagi-
nary spatial scenarios can influence how people interpret spatiotemporal metaphors
 SPACE FOR THINKING 459

(Boroditsky and Ramscar, 2002). These studies rule out what Boroditsky (2000) calls the
Dubious View, that space-time metaphors in language are simply ‘etymological relics
with no psychological consequences’ (pg. 6).
If people use spatial schemas to think about time, as suggested by metaphors in
language, then do people who use different spatiotemporal metaphors in their native
tongues think about time differently? To find out, Boroditsky (2001) compared perform-
ance on space-time priming tasks in speakers of English, a language which typically
describes time as horizontal, and speakers of Mandarin Chinese, which also commonly
uses vertical spatiotemporal metaphors. English speakers were faster to judge sentences
about temporal succession (e.g., March comes earlier than April) when primed with a
horizontal spatial event, but Mandarin speakers were faster to judge the same sentences
when primed with a vertical spatial stimulus. This was true despite the fact that all of
the sentences were presented in English. In a follow-up study, Boroditsky (2001) trained
English speakers to use vertical metaphors for temporal succession (e.g., March is above
April). After training, their priming results resembled those of the native Mandarin
speakers.
Together, Boroditsky’s studies provide some of the first evidence that (a) people
not only talk about time in terms of space, they also think about it that way, (b) people
who use different spatiotemporal metaphors also think about time differently, and (c)
learning new spatial metaphors can change the way you mentally represent time. Yet,
these conclusions are subject to a skeptical interpretation. Boroditsky’s participants
made judgments about sentences containing spatial or temporal language. Perhaps their
judgments showed relations between spatial and temporal thinking that were consistent
with linguistic metaphors only because they were required to process space or time in
language. Would the same relationships between mental representations of space and
time be found if participants were tested on nonlinguistic tasks?
The fact that people communicate via language replete with anaphora, ambiguity,
metonymy, sarcasm, and deixis seems proof that what we say provides only a thumbnail
sketch of what we think. Most theorists posit at least some independence between
semantic representations and underlying conceptual representations (Jackendoff, 1972;
Katz and Fodor, 1963; Levelt, 1989; cf., Fodor, 1975). Even those who posit a single,
shared ‘level’ of representation for linguistic meaning and nonlinguistic concepts
allow that semantic structures must constitute only a subset of conceptual structures
(Chomsky, 1975; Jackendoff, 1983). Because we may think differently when we’re using
language and when we’re not, well-founded doubts persist about how deeply patterns
in language truly reflect – and perhaps shape – our nonlinguistic thought. According
to linguist Dan Slobin (1996):

Any utterance is a selective schematization of a concept – a schematization that is

in some ways dependent on the grammaticized meanings of the speaker’s particular
language, recruited for the purposes of verbal expression. (pg. 75–76)

Slobin argues that when people are ‘thinking for speaking’ (and presumably for reading
or listening to speech), their thoughts are structured, in part, according to their language
460 LANGUAGE, COGNITION AND SPACE

and its peculiarities. Consequently, speakers of different languages may think differently
when they are using language. But how about when people are not thinking for speaking?
Eve Clark (2003) asserts that:

[When people are] thinking for remembering, thinking for categorizing, or one
of the many other tasks in which we may call on the representations we have of
objects or events – then their representations may well include a lot of material
not customarily encoded in their language. It seems plausible to assume that such
conceptual representations are nearer to being universal than the representations
we draw on for speaking. (pg. 21)

Clark predicts that results may differ dramatically between tests of language–thought
relations that use language and those that do not:

…we should find that in tasks that require reference to representations in memory that
don’t make use of any linguistic expression, people who speak different languages will
respond in similar, or even identical, ways. That is, representations for nonlinguistic
purposes may differ very little across cultures or languages. (2003, pg. 22)

Clark adds:

Of course, finding the appropriate tasks to check on this without any appeal to
language may prove difficult. (2003, pg. 22)

Clark’s skepticism echoes concerns raised by Papafrougou, Massey, and Gleitman (2002)
regarding the difficulty of studying the language–thought interface:

…domains within which language might interestingly influence thought are higher–
level cognitive representations and processes, for instance, the linguistic encoding
of time […] A severe difficulty in investigating how language interfaces with
thought at these more ‘significant’ and ‘abstract’ levels has been their intractability
to assessment. As so often, the deeper and more culturally resonant the cognitive or
social function, the harder it is to capture it with the measurement and categorization
tools available to psychologists. (pg. 191–192)

For the studies reported here, new experimental tools were developed in order to (a)
evaluate Metaphor Theory as an account of the structure and evolution of abstract
concepts, and (b) investigate relationships between language and nonlinguistic mental
representations. The first two sets of experiments used the concrete domain of space and
the relatively abstract domain of time as a testbed for Metaphor Theory, and the final
set extended these findings beyond the domain of time. These experiments used novel
psychophysical tasks with nonlinguistic stimuli and responses in order to distinguish
two theoretical positions, one which posits shallow and the other deep relations between
language and nonlinguistic thought (table 1):
 SPACE FOR THINKING 461

Table 1.

The Shallow View: The Deep View:

i. Language reflects the structure of the mental i. Language reflects the structure of the mental
representations that speakers form for the purpose representations that speakers form for the purpose of
of using language. These are likely to be importantly using language. These are likely to be similar to, if not
different, if not distinct, from the representations overlapping with, the representations people use when
people use when they are thinking, perceiving, and they are thinking, perceiving, and acting without using
acting without using language. language.

ii. Language may influence the structure of mental ii. Patterns of thinking established during language
representations, but only (or primarily) during use may influence the structure of the mental
language use. representations that people form even when they’re not
using language.

iii. Cross-linguistic typological differences are likely iii. Some cross-linguistic typological differences
to produce ‘shallow’ behavioral differences on tasks are likely to produce ‘deep’ behavioral differences,
that involve language or high-level cognitive abilities observable not only during tasks that involve language
(e.g., naming, explicit categorization). However, such or high-level cognitive abilities, but also when subjects
behavioral differences should disappear when subjects are tested using nonlinguistic tasks that involve low-
are tested using nonlinguistic tasks that involve low- level perceptuo-motor abilities.
level perceptuo-motor abilities.

iv. Although the semantics of languages differ, iv. Where the semantics of languages differ, speakers’
speakers’ underlying conceptual and perceptual underlying conceptual and perceptual representations
representations are, for the most part, universal. may differ correspondingly, such that language
communities develop distinctive conceptual
repertoires.

2 Do people use space to think about time?

Do people use mental representations of space in order to mentally represent time, as

metaphors in language suggest they do – even when they’re not using language? The first
six experiments reported here tested the hypothesis that temporal thinking depends, in
part, on spatial thinking (Casasanto and Boroditsky, 2008). In each task, participants
viewed simple nonlinguistic, non-symbolic stimuli (i.e., lines or dots) on a computer
screen, and estimated either their duration or their spatial displacement. Durations
and displacements were fully crossed, so there was no correlation between the spatial
and temporal components of the stimuli. As such, one stimulus dimension served as a
distractor for the other: an irrelevant piece of information that could potentially interfere
with task performance. Patterns of cross-dimensional interference were analyzed to
reveal relationships between spatial and temporal representations. 3
Broadly speaking, there are three possible relationships between people’s mental
representations of space and time. First, the two domains could be symmetrically depend-
ent. John Locke (1689/1995) argued that space and time are mutually inextricable in our
minds, concluding that, ‘expansion and duration do mutually embrace and comprehend
each other; every part of space being in every part of duration, and every part of duration
in every part of expansion’ (p. 140). Alternatively, our ideas of space and time could be
462 LANGUAGE, COGNITION AND SPACE

independent. Any apparent relatedness could be due to structural similarities between

essentially unrelated domains (Murphy, 1996, 1997). A third possibility is that time
and space could be asymmetrically dependent. Representations in one domain could be
parasitic on representations in the other, as suggested by their asymmetric relationship
in linguistic metaphors (Boroditsky, 2000; Gentner, 2001; Gibbs, 1994; Lakoff and
Johnson, 1980, 1999).
These three possible relationships between space and time predict three distinct
patterns of cross-dimensional interference. If spatial and temporal representations are
symmetrically dependent on one another, then any cross-dimensional interference
should be approximately symmetric: line displacement should modulate estimates
of line duration, and vice versa. Alternatively, if spatial and temporal representa-
tions are independent, there should be no significant cross-dimensional interference.
However, if mental representations of time are asymmetrically dependent on mental
representations of space, as suggested by spatiotemporal metaphors in language,
then any cross-dimensional interference should be asymmetric: line displacement
should affect estimates of line duration more than line duration affects estimates of
line displacement.
For Experiment 1, native English speaking participants viewed 162 lines of
varying lengths (200–800 pixels, in 50 pixel increments), presented on a computer
monitor for varying durations (1–5 seconds, in 500 ms increments). Lines ‘grew’
horizontally from left to right, one pixel at a time, along the vertical midline. Each
line remained on the screen until it reached its maximum displacement, and then
disappeared. Immediately after each line was shown, a prompt appeared indicating
that the participant should reproduce either the line’s displacement (if an ‘X’ icon
appeared) or its duration (if an ‘hourglass’ icon appeared), by clicking the mouse to
indicate the endpoints of each temporal or spatial interval. Space trials and time trials
were randomly intermixed.
Results of Experiment 1 showed that spatial displacement affected estimates of
duration, but duration did not affect estimates of spatial displacement (Figure 1a).
For stimuli of the same average duration, lines that travelled a shorter distance were
judged to take a shorter time, and lines that travelled a longer distance were judged
to take a longer time. Subjects incorporated irrelevant spatial information into their
temporal estimates, but not vice versa. Estimates of duration and displacement were
highly accurate, and were equally accurate in the two domains. The asymmetric
cross-dimensional interference we observe cannot be attributed to a difference in
the accuracy of duration and displacement estimations, as no significant difference
in was found.
Experiments were conducted to assess the generality of these results, and to evalu-
ate potential explanations. In Experiment 1, participants did not know until after each
line was presented whether they would need to estimate displacement or duration.
They had to attend to both the spatial and temporal dimensions of the stimulus.
Experiment 2 addressed the possibility that cross-dimensional interference would
diminish if participants were given the opportunity to attend selectively to the trial-
relevant stimulus dimension, and to ignore the trial-irrelevant dimension. Materials
 SPACE FOR THINKING 463

and procedures were identical to those used in Experiment 1, with one exception. A
cue preceded each growing line, indicating which stimulus dimension participants
would need to reproduce. Results of Experiment 2 (Figure 1b) replicated those of
Experiment 1. Participants were able to disregard line duration when estimating
displacement. By contrast, they were unable to ignore line displacement, even when
they were encouraged to attend selectively to duration. The cross-dimensional effect
of space on time estimation in Experiment 1 was not caused by a task-specific demand
for subjects to encode spatial and temporal information simultaneously.
Experiments 3–5 addressed concerns that spatial information in the stimulus may
have been more stable or more salient than temporal information, and that differences in
stability or salience produced the asymmetrical cross-dimensional interference observed
in Experiments 1 and 2. One concern was that participants may have relied on spatial
information to make temporal estimates because stimuli were situated in a constant
spatial frame of reference (i.e., the computer monitor). For Experiment 3, stimuli were
also situated in a constant temporal frame of reference. Temporal delay periods were
introduced preceding and following line presentations, which were proportional to the
spatial gaps between the ends of the stimulus lines and the edges of the monitor. Results
(Figure 1c) replicated those of Experiments 1 and 2.
Experiment 4 addressed the possibility that space would no longer influence par-
ticipants’ time estimates if stimulus duration were indexed by something non-spatial.
For this experiment, a constant tone (260 Hz) accompanied each growing line. Materials
and procedures were otherwise identical to those used in Experiment 2. The tone began
sounding when the line started to grow across the screen, and stopped sounding when
the line disappeared. Thus, stimulus duration was made available to the participant in
both the visual and auditory modalities, but stimulus displacement was only available
visually. Results (Figure 1d) replicated those of the previous experiments. Displacement
strongly influenced participants’ duration estimates, even when temporal information
was provided via a different sensory modality from the spatial information.
Experiment 5 was designed to equate the mnemonic demands of the spatial and
temporal dimensions of the stimulus. Materials and procedures were identical to those
used in Experiment 2, with one exception. Rather than viewing a growing line, subjects
viewed a dot (10x10 pixels) that moved horizontally across the midline of the screen.
In the previous experiments, just before each growing line disappeared participants
could see its full spatial extent, from end to end, seemingly at a glance. By contrast, the
spatial extent of a moving dot’s path could never be seen all at once, rather it had to
be imagined: in order to compute the distance that a dot travelled, participants had to
retrieve the dot’s starting point from memory once its ending point was reached. The
spatial and temporal dimensions of the dot stimulus had to be processed similarly in
this regard: whenever we compute the extent of a temporal interval we must retrieve
its starting point from memory once the end of the interval is reached. Results (Figure
1e) replicated those of previous experiments.
Experiment 6 investigated whether motion or speed affected participants’ time
estimates in Experiments 1–5, rather than stimulus displacement. Materials and pro-
cedures were identical to those used in Experiment 2, with the following exception.
464 LANGUAGE, COGNITION AND SPACE

Rather than growing lines, participants viewed stationary lines, and estimated either
the amount of time they remained on the screen or their distance from end to end,
using mouse clicks. Results replicate those of previous five experiments (Figure 1f),
indicating that stimulus displacement can strongly modulate time estimates even in
the absence of stimulus motion.

Figure 1. Summary of cross-dimensional interference effects for Experiments 1–6. The effect of
distance on time estimation was significantly greater than the effect of time on distance estimation
for all experiments. (1a, Growing lines: difference of correlations = 0.75; z = 3.24, p <.001. 1b, Growing
lines, selective attention: difference of correlations = 0.66; z = 2.84, p < .003. 1c, Growing lines, tempo-
ral frame of reference: difference of correlations = 0.71; z =2.09, p <.02. 1d, Growing lines, concurrent
tone: difference of correlations =0.63; z = 2.60, p <.005. 1e, Moving dot: difference of correlations =
1.45; z = 3.69, p <.001. 1f, Stationary lines: difference of correlations = 0.54; z = 1.62, p <.05.) Figure
reproduced with permission from Casasanto, D. and Boroditsky, L. (2008). Time in the Mind: Using
space to think about time. Cognition, 106, 579–593.

Results of all six experiments unequivocally support the hypothesis that people incorpo-
rate spatial information into their time judgments more than they incorporate temporal
information into their spatial judgments. These findings converge with those of Cantor
and Thomas (1977), who showed that spatial information influences temporal judg-
ments but not vice versa for very briefly presented stimuli (30–70 msecs). Previous
behavioral tests of Metaphor Theory have used linguistic stimuli (Boroditsky, 2000,
2001; Boroditsky and Ramscar, 2002; Gibbs, 1994; Meier and Robinson, 2004; Meier,
Robinson and Clore, 2004; Richardson, Spivey, Barsalou and McRae, 2003; Schubert,
2005; Torralbo, Santiago and Lupiáñez, 2006). While these studies support the psy-
chological reality of mental metaphors, they leave open the possibility that people only
think about abstract domains like time metaphorically when they are using language
(i.e., when they are ‘thinking for speaking’ (E. Clark, 2003; Slobin, 1996)). Experiments
described above used nonlinguistic stimuli and responses, and demonstrated for the
first time that even our low-level perceptuo-motor representations in the domains of
space and time are related as predicted by linguistic metaphors.
 SPACE FOR THINKING 465

Although English speakers describe time in terms of space almost obligatorily

(Jackendoff, 1983; Pinker, 1997), we can also optionally describe space in terms of
time. For example, in English we could say my brothers live 5 minutes apart to indicate
that they live a short distance apart. Thus, the relationship between time and space
in linguistic metaphors is asymmetrical, but not unidirectional. Accordingly, asym-
metrical cross-dimensional interference between space and time was predicted in
these experiments. This prediction does not entail that time can never affect spatial
judgments: only that the effect of space on time estimation should be greater than the
effect of time on space estimation when the effects are compared appropriately. Results
of Experiments 1–6 did not show any significant effect of time on distance estimation,
but such a finding would still be compatible with the asymmetry hypothesis, so long
as the effect of distance on time estimation was significantly greater than the effect of
time on distance estimation.
It is noteworthy that space influenced temporal judgments even for spatiotemporal
stimuli that participants could experience directly. Growing lines are observable, and
are arguably less abstract than entities like the ‘moving meeting’ described in section
0.1. Brief durations could, in principle, be mentally represented independently of space,
by an interval-timer or pulse-accumulator (see Ivry and Richardson, 2002 for review),
yet these data suggest that spatial representations are integral to the timing of even
simple, observable events. Thinking about time metaphorically in terms of space may
allow us to go beyond these basic temporal representations. Mentally representing time
as a linear path may enable us to conceptualize more abstract temporal events that we
cannot experience directly (e.g., moving a meeting forward or pushing a deadline back),
as well as temporal events that we can never experience at all (e.g., the remote past or
the distant future). Metaphorical mappings from spatial paths, which can be traveled
both forward and backward, may give rise to temporal constructs such as time–travel
that only exist in our imagination.
Together, these experiments demonstrate that the metaphors we use can provide
a window on the structure of our abstract concepts. They also raise a further question
about relations between linguistic metaphors and nonlinguistic mental representations:
if people think about time in terms of space (the way they talk about it), then do people
who use different space-time metaphors in their native languages think differently – even
when they’re not using language?

3 Does language shape the way we think about time?

The first set of experiments supports the Deep View of language-thought relations by
showing that temporal representations depend, in part, on spatial representations,
as predicted by metaphors in English – even when people are performing low-level,
nonlinguistic psychophysical tasks (see Table 1, number i). However, it is not clear from
these data whether linguistic metaphors merely reflect English speakers’ underlying
nonlinguistic representations of time, or whether language also shapes those represen-
tations. According to the Shallow View, it is possible that speakers of a language with
466 LANGUAGE, COGNITION AND SPACE

different duration metaphors would nevertheless perform similarly to English speakers

on nonlinguistic tasks. Thus, the first set of experiments leaves the following question
unaddressed, posed by the influential amateur linguist, Benjamin Whorf:

Are our own concepts of ‘time’, ‘space’, and ‘matter’ given in substantially the same
form by experience to all men, or are they in part conditioned by the structure of
particular languages?’ (1939/2000, pg. 138.)

This Whorfian question remains the subject of renewed interest and debate. Does
language shape thought? The answer yes would call for a reexamination of the ‘univer-
salist’ assumption that has guided Cognitive Science for decades, according to which
nonlinguistic concepts are formed independently of the words that name them, and
are invariant across languages and cultures (Fodor, 1975; Pinker, 1994, Papafragou,
Massey and Gleitman, 2002). This position is often attributed to Chomsky (1975), but
has been articulated more recently by Pinker (1994) and by Lila Gleitman and col-
leagues (Papafragou, Massey and Gleitman, 2002; Snedeker and Gleitman, 2004). The
Shallow View proposed here can be considered a variety of the universalist view that
can still plausibly be maintained despite recent psycholinguistic evidence supporting
the Whorfian hypothesis (e.g., Boroditsky, 2001).
Skepticism about some Whorfian claims has been well founded (see Pinker, 1994,
ch. 3, for a review of evidence against the Whorfian hypothesis). A notorious fallacy,
attributable in part to Whorf, illustrates the need for methodological rigor. Whorf
(1939/2000) argued that Eskimos must conceive of snow differently than English speak-
ers because the Eskimo lexicon contains multiple words that distinguish different types
of snow, whereas English has only one word to describe all types. The exact number
of snow words the Eskimos were purported to have is not clear. This number has now
been inflated by the popular press to as many as four hundred. According to a Western
Greenlandic Eskimo dictionary published in Whorf ’s time, however, Eskimos may have
had as few as two distinct words for snow (Pullum, 1991).
Setting aside Whorf ’s imprecision and the media’s exaggeration, there remains
a critical missing link between Whorf ’s data and his conclusions: Whorf (like many
researchers today) used purely linguistic data to support inferences about nonlinguistic
mental representations. Steven Pinker illustrates the resulting circularity of Whorf ’s
claim in this parody of his logic:

[They] speak differently so they must think differently.

How do we know that they think differently?
Just listen to the way they speak! (Pinker, 1994, pg. 61).

Such circularity would be escaped if nonlinguistic evidence could be produced to

show that two groups of speakers who talk differently also think differently in cor-
responding ways.
A series of experiments explored relationships between spatiotemporal language
and nonlinguistic mental representation of time. The first experiment, a corpus search,
uncovered previously unexplored cross-linguistic differences in spatial metaphors for
 SPACE FOR THINKING 467

duration. Next, we tested whether these linguistic differences correlate with differences
in speakers’ low-level, nonlinguistic time representations.4 Finally, we evaluated a causal
role for language in shaping time representations.5

3.1 1-Dimensionsal and 3-dimensional spatial metaphors for time

Literature on how time can be expressed verbally in terms of space (and by hypothesis,
conceptualized in terms space) has focused principally on linear spatial metaphors. But
is time necessarily conceptualized in terms of unidimensional space? Some theorists
have suggested so (Clark, 1973, Gentner, 2001), and while this may be true regard-
ing temporal succession, linguistic metaphors suggest an alternative spatialization for
duration. English speakers not only describe time as a line, they also talk about oceans
of time, saving time in a bottle, and liken the ‘days of their lives’ to sands through the
hourglass. Quantities of time are described as amounts of a substance occupying three
dimensional space (i.e., volume).
Experiment 7 compared the use of ‘time as distance’ and ‘time as amount’ metaphors
across four languages. Every language we examined uses both distance and amount
metaphors, but their relative prevalence and productivity appear to vary markedly. In
English, it is natural to talk about a long time, borrowing the structure and vocabulary
of a linear spatial expression like a long rope. Yet in Spanish, the direct translation of
‘long time’, largo tiempo, sounds awkward to speakers of most dialects.6 Mucho tiempo,
which means ‘much time’, is preferred.
In Greek, the words makris and kontos are the literal equivalents of the English
spatial terms long and short. They can be used in spatial contexts much the way long
and short are used in English (e.g., ena makry skoini means ‘a long rope’). In temporal
contexts, however, makris and kontos are dispreferred in instances where long and short
would be used naturally in English. It would be unnatural to translate a long meeting liter-
ally as mia makria synantisi. Rather than using distance terms, Greek speakers typically
indicate that an event lasted a long time using megalos, which in spatial contexts means
physically ‘large’ (e.g., a big building), or using poli, which in spatial contexts means
‘much’ (e.g., much water). Compare how English (e) and Greek (g) typically modify the
duration of the following events (literal translations in parentheses):

1e long night
1g megali nychta (big night)

2e long relationship
2g megali schesi (big relationship)

3e long party
3g parti pou kratise poli (party that lasts much)

4e long meeting
4g synantisi pou diekese poli (meeting that lasts much)
468 LANGUAGE, COGNITION AND SPACE

In examples 1g and 2g, the literal translations might surprise an English speaker, for
whom big night is likely to mean ‘an exciting night’, and big relationship ‘an important
relationship’. For Greek speakers, however, these phrases can also communicate dura-
tion, expressing time not in terms of 1-dimensional linear space, but rather in terms of
3-dimensional size or amount.
To quantify the relative prevalence of distance and amount metaphors for duration
across languages, the most natural phrases expressing the ideas ‘a long time’ and ‘much
time’ were elicited from native speakers of English (long time, much time), French
(longtemps, beaucoup de temps), Greek (makry kroniko diastima, poli ora), and Spanish
(largo tiempo, mucho tiempo). The frequencies of these expressions were compared in
a very large multilingual text corpus: www.google.com. Each expression was entered as
a search term. Google’s language tools were used to find exact matches for each expres-
sion, and to restrict the search to web pages written only in the appropriate languages.
The number of google ‘hits’ for each expression was tabulated, and the proportion of
distance hits and amount hits was calculated for each pair of expressions, as a measure
of their relative frequency. English and French, distance metaphors were dramatically
more frequent than amount metaphors. The opposite pattern was found in Greek and
Spanish (Figure 2).
Although all languages surveyed use both distance and amount metaphors for
duration, the relative strengths of these metaphors appears to vary across languages.
This simple corpus search by no means captures all of the complexities of how time
is metaphorized in terms of space within or between languages, but these findings
corroborate native speakers’ intuitions for each language, and provide a quantitative
linguistic measure on which to base predictions about behavior in nonlinguistic tasks.

Figure 2. Results of Experiment 7. Black bars indicate the proportion of Google ‘hits’ for expressions
meaning long time, and white bars for expressions meaning much time in each language.
 SPACE FOR THINKING 469

3.2 Do people who talk differently think differently?

Do people who use different spatiotemporal metaphors think about time differently –
even when they’re not using language? Experiments 8 and 9 explored the possibility that
speakers who preferentially use distance metaphors in language tend to co-opt linear
spatial representations to understand duration, whereas speakers who preferentially
use amount metaphors tend to co-opt 3-dimensional spatial representations. Speakers
of two languages surveyed in Experiment 7 (i.e., English and Greek) performed a pair
of nonlinguistic psychophysical tasks, which required them to estimate duration while
overcoming different kinds of spatial interference (i.e., distance or amount interference).
If people’s conceptions of time are substantially the same universally irrespective of the
languages they speak, as suggested by the Shallow View, then performance on these
tasks should not differ between language groups. On the Deep View, however, it was
predicted that participants’ performance should vary in ways that parallel the metaphors
in their native languages.
The ‘distance interference’ task was modeled on the ‘growing line’ task described
in Experiment 2. English participants in the previous growing line studies may have
suffered interference from distance during duration estimation, in part, because distance
and duration are strongly conflated in the English lexicon. Would the same confusion
be found in speakers of other languages? It was predicted that native English speakers
would show a strong effect of distance on time estimation when performing the growing
line task, whereas speakers of Greek would show a weaker effect, since distance and
duration are less strongly associated in the Greek language .
A complementary ‘amount interference’ task was developed, in which participants
watched a schematically drawn container of water filling up gradually, and estimated
either how full it became or how much time it remained on the computer screen, using
mouse clicks as in the growing line tasks. Spatial and temporal parameters of the stimuli
were equated across tasks. Behavioral predictions for the Filling Tank task were the
mirror image of predictions for the Growing Line task: speakers of Amount Languages
like Greek should show a strong influence of ‘fullness’ on time estimation, whereas
speakers of Distance Languages like English should show a weaker effect.
Results showed that effects of spatial interference on duration estimation followed
predictions based on the relative prevalence of distance and amount metaphors for
time in speakers’ native languages. English showed a strong effect of line length but a
weak effect of tank fullness on duration estimation; Greek speakers showed the opposite
pattern of results (Figure 3). A 2 x 2 ANOVA compared these slopes with Language
(English, Greek) and Task (distance interference, amount interference) as between-
subject factors, revealing a highly significant Language by Task interaction, with no
main effects (F(1,56)=10.41, p=.002).
470 LANGUAGE, COGNITION AND SPACE

Figure 3. Results of Experiments 8 and 9. Black bars indicate the slope of the effect of line displace-
ment on duration estimation. White bars indicate the slope of the effect of tank fullness on duration
estimation. The relationship between the effects of distance and volume on time estimation was
predicted by the relative prevalence of distance and amount metaphors in English and Greek (see
figure 2).

The observed differences in the effects of spatial distance and amount on duration esti-
mation cannot be attributed to overall differences in performance across tasks or across
groups. Within-domain performance (i.e., the effect of target duration on estimated
duration, and the effect of target distance or fullness on estimated distance or fullness)
was compared across tasks and across groups: no significant differences were found
between correlations or slopes, even in pairwise comparisons.
One difference between the Growing Line and Filling Tank tasks was that the
lines grew horizontally, but the tanks filled vertically. To determine whether the spatial
orientation of the stimuli and responses gave rise to the observed cross-linguistic differ-
ences in performance on the Growing Lines and Filling Tank tasks, an Upward Growing
Lines task was administered to speakers of English and Greek. No significant difference
was found in the effect of vertical displacement on time estimation across languages,
suggesting that the orientation of stimuli cannot account for the between-group differ-
ences observed in Experiments 8 and 9.
Overall, Experiments 7–9 show that the way people talk about time correlates
strongly with the way they think about it – even when they’re performing simple
nonlinguistic perceptuo-motor tasks – as predicted by the Deep View of language-
thought relations. (See Table 1, ii.- iv.) Much of the literature on temporal language has
highlighted crosslinguistic commonalities in spatiotemporal metaphors (e.g., Alverson,
1994). The studies presented here begin to explore some previously neglected crosslin-
guistic differences, and to discover their nonlinguistic consequences. The corpus search
reported in Experiment 7 provides one measure of how frequently different languages
use distance and amount metaphors for duration; the relative frequencies of long time
and much time expressions across languages proved highly predictive of performance
on nonlinguistic duration estimation tasks. Often, however, spatial metaphors describe
events rather than describing time, per se. Preliminary data from a questionnaire study
 SPACE FOR THINKING 471

suggest that English consistently prefers distance metaphors for describing both time
(e.g., a long time) and events (e.g., a long party), whereas Greek consistently prefers
volume metaphors for time (e.g., poli ora tr.‘much time’) and for events (e.g., parti pou
kratise poli tr. ‘party that lasts much’), corroborating the results of the corpus search.
Ongoing studies seek to characterize these crosslinguistic differences more fully, and
to specify which features of language correspond to ‘deep’ differences in nonlinguistic
mental representations of time.

3.3 How might perceptual and linguistic experience shape abstract thought?

How do people come to think about time in terms of space? How do speakers of different
languages come to conceptualize time differently? Turning to the first question, some
mappings from concrete to abstract domains of knowledge may be initially established
pre-linguistically, based on interactions with the physical world (Clark, 1973). For
example, people are likely to track the kinds of correlations in experience that are
important for perceiving and acting on their environment; they may learn associations
between time and space by observing that more time passes as objects travel farther,
and as substances accumulate more. This proposal entails that although time depends
in part on spatial representations, time can also be mentally represented qua time,
at least initially: in order for cross-dimensional associations to form, some primitive
representations must already exist in each dimension. Primitive temporal notions,
however, of the sort that we share with infants and non-human animals, may be too
vague or fleeting to support higher order reasoning about time. Grafting primitive
temporal representations onto spatial representations may make time more amenable
to verbal or imagistic coding, and may also import the inferential structure of spatial
relations into the domain of time (Pinker, 1997).
If metaphorical mappings are experience-based, and are established pre-linguis-
tically, what role might language play in shaping abstract thought? Since the laws of
physics are the same in all language communities, prelinguistic children’s conceptual
mappings between time, distance, and amount could be the same universally. Later, as
children acquire language, these mappings are adjusted: each time we use a linguistic
metaphor, we activate the corresponding conceptual mapping. Speakers of Distance
Languages then activate the time-distance mapping frequently, eventually strengthening
it at the expense of the time-amount mapping (and vice versa for speakers of Amount
Languages). Mechanistically, this could happen via a process of competitive associative
learning.
Did language experience give rise to the language-related differences in performance
reported for the Growing Line and Filling Tank experiments? A perennial complaint
about studies claiming effects of language on thought is that researchers mistake cor-
relation for causation. Although it is difficult to imagine what nonlinguistic cultural
or environmental factors could have caused performance on Experiments 8 and 9 in
English and Greek speakers to align so uncannily with the metaphors in these languages,
the data are nevertheless correlational. Using crosslinguistic data to test for a causal
472 LANGUAGE, COGNITION AND SPACE

influence of language on thought is problematic, since experimenters cannot randomly

assign subjects to have one first language or another: crosslingusitic studies are neces-
sarily quasi-experimental.
For Experiment 10, a pair of training tasks (i.e., true experimental interventions) was
conducted to provide an in principle demonstration that language can influence even the
kinds of low-level mental representations that people construct while performing psy-
chophysical tasks, and to test the hypothesis that language shapes time representations
in natural settings by adjusting the strengths of cross-domain mappings. Native English
speakers were randomly assigned to perform either a Distance Training or Amount
Training task. Participants completed 192 fill-in-the-blank sentences using the words
longer or shorter for Distance Training, and more or less for the Amount Training task.
Half of the sentences compared the length or capacity of physical objects (e.g., An alley
is longer / shorter than a clothesline; A teaspoon is more / less than an ocean), the other
half compared the duration of events (e.g., A sneeze is longer / shorter than a vacation;
A sneeze is more / less than a vacation). By using distance terms to compare event dura-
tions, English speakers were reinforcing the already preferred source-target mapping
between distance and time. By using amount terms, English speakers were describing
event durations similarly to speakers of an Amount Language (see Greek examples
in section 2.1), and by hypothesis, they were activating the dispreferred volume-time
mapping. After this linguistic training, all participants performed the nonlinguistic
Filling Tank task from Experiment 9. We predicted that if using a linguistic metaphor
activates the corresponding conceptual mapping between source and target domains,
then repeatedly using amount metaphors during training should (transiently) strengthen
participants’ nonlinguistic amount-time mapping.
Consistent with this prediction, the slope of the effect of amount on time estimation
was significantly greater after amount training than after distance training (difference
of slopes = 0.89, t(28) = 1.73, p<.05; Figure 4). Following about 30 minutes of concen-
trated usage of amount metaphors in language, native English speakers’ performance
on the Filling Tank task was statistically indistinguishable from the performance of
the native Greek speakers tested in Experiment 9. By encouraging the habitual use
of either distance- or amount-based mental metaphors, our experience with natural
language may influence our everyday thinking about time in much the same way as
this laboratory training task.
These findings help to resolve apparent tensions between the proposal that percep-
tuo-motor image schemas underlie our abstract concepts and the notion of linguistic
relativity. Johnson (2005) defines an image schema as ‘a dynamic recurring pattern of
organism-environment interactions’ (pg. 19). Presumably, people from all language
communities inhabit the same physical world and interact with their environment using
the same perceptuo-motor capacities, therefore the image schemas they develop should
be universal. Yet, even if we all develop similar image schemas initially, based on our
physical experiences, Experiments 8–10 suggest the way we deploy these image schemas
depends on our linguistic experiences. Duration can be mentally represented both in
terms of distance and in terms of amount. The extent to which each of these conceptual
space-time mappings is activated in a given speaker or community of speakers varies
 SPACE FOR THINKING 473

with the strength of the corresponding linguistic metaphors. The structure of abstract
concepts like duration appears to be shaped both by perceptuo-motor experience (which
is plausibly universal) and by language use (which is culture-specific).

Figure 4. Results of Experiment 10. Bars indicate the slope of the effect of tank fullness on dura-
tion estimation after training with distance metaphors (left), amount metaphors (right), or with no
training (middle) prior to performing the Filling Tank task. The cross-dimensional effect of amount on
time estimation was significantly greater after training with amount metaphors than with distance
metaphors.

4 Beyond space and time: Spatial representation of musical pitch

Time and space provide a model system for exploring connections between abstract
and concrete mental representations, but time is just one among many domains that
we spatialize in language; time may be just one of many abstract domains that import
their structure or content, in part, from the domain of space. In Experiment 11, the
psychophysical tasks that were developed to investigate space and time were adapted
to explore relationships between space and musical pitch.7
Like time, pitch is often described in English using linear spatial terms. Unlike time,
pitch tends to be described using vertical rather than horizontal metaphors. Pitches can
be high or low, and can rise, fall, soar, or dip below the staff. Yet, the fact that we talk about
pitch in terms of vertical space doesn’t necessarily mean that we think about it that way.
One possibility is that pitch is mentally represented on its own terms, and is only coded
into the same words that we use to describe space as a matter of convenience: domains
that share structural similarities may be amenable to common linguistic description,
obviating multiple domain-specific vocabularies. Alternatively, the spatialization of
pitch in language may serve as a clue that leads us to a fuller understanding of how
pitch is mentally represented.
The ‘growing line’ task described in Experiment 2 was modified for a nonlinguistic
test of the hypothesis that our mental representations of musical pitch depend, in part,
on spatial representations. Nine displacements ranging from 100 to 500 pixels (in 50
pixel increments) were fully crossed with nine different pitches ranging from middle
474 LANGUAGE, COGNITION AND SPACE

C4 to G#4 (in semitone increments). For each trial, participants heard a constant pitch
while watching a line grow up the screen from bottom to top (for half of the subjects)
or across the screen from left to right (for the other half of the subjects). Before each
stimulus, participants were informed whether they would need to estimate distance
or pitch, to encourage them to attend to the trial-relevant stimulus dimension and, if
possible, to ignore the trial-irrelevant dimension. Participants estimated line displace-
ments using mouse clicks, as in previous experiments. To estimate pitch, participants
used the mouse to adjust a probe tone until it matched the remembered target pitch.
Watching vertical lines significantly modulated subjects’ pitch estimates: tones of
the same average frequency were judged to be higher in pitch if they accompanied lines
that grew higher on the screen (effect of actual distance on estimated pitch: slope=.37;
r2=.77, p<.003). By contrast, watching horizontal lines did not significantly modulate
pitch estimates. This finding is consistent with the occurrence of vertical but not hori-
zontal metaphors for pitch in English. Further analyses showed that whereas vertical
displacement affected estimates of pitch, pitch did not significantly influence estimates
of vertical displacement. Thus, the relation between nonlinguistic mental representa-
tions of space and pitch appears to be asymmetrical, as predicted by the directionality
of space-pitch metaphors in language.
While these results support the claim that musical pitch is mentally represented in
part metaphorically, in terms of vertical space, they are agnostic as to the direction of
causation between language and thought. Further studies (such as those described in
sections 2.1–2.3) are needed to investigate whether linguistic metaphors merely reflect
the spatial schemas that partly constitute pitch representations, or whether the way we
talk about pitch can also shape the way we think about it.

5 Conclusions

Direct evidence that spatial cognition supported the evolution of abstract concepts may
forever elude us, because human history cannot be recreated in the laboratory, and
the mind leaves no fossil record. However, the studies reported here demonstrate the
importance of spatial representations for abstract thinking in the mind that evolution
produced. For decades, inferences about the perceptual foundations of abstract thought
rested principally on linguistic and psycholinguistic data. These psychophysical experi-
ments show that even nonlinguistic representations in concrete and abstract domains
are related as linguistic metaphors predict: we think in mental metaphors.
Together, the experiments described in this chapter suggest that people not only
talk about abstract domains using spatial words, they also think about them using
spatial representations. Results are incompatible with the Shallow View of language-
thought relations, and provide some of the first evidence for the view that language has
Deep influences on nonlinguistic mental representation (see table 1). Experiments 1–6
show that people use spatial representations to think about time even when they’re not
producing or understanding language. Experiments 7–9 show that people who talk
differently about time also think about it differently, in ways that correspond to their
 SPACE FOR THINKING 475

language-particular metaphors. Experiment 10 shows that language not only reflects the
structure of underlying mental representations, it can also shape those representations in
ways that influence how people perform even low-level, nonlinguistic, perceptuo-motor
tasks. Experiment 11 shows that these findings extend beyond the ‘testbed’ domains
of space and time.
These findings are difficult to reconcile with a universalist position according to
which language calls upon nonlinguistic concepts that are presumed to be ‘universal’
(Pinker, 1994, pg. 82) and ‘immutable’ (Papafragou, Massey and Gleitman, 2002, pg.
216). Beyond influencing thinking for speaking (Slobin, 1996), language can also influ-
ence the nonlinguistic representations we build for remembering, acting on, and perhaps
even perceiving the world around us. It may be universal that people conceptualize time
according to the spatial metaphors, but since these metaphors vary across languages,
members of different language communities develop distinctive conceptual repertoires.
The structure of abstract domains like time depends, in part, on both perceptuo-motor
experience and on experience using language.

Acknowledgments
Thanks to Lera Boroditsky, Herb Clark, and the citizens of Cognation for helpful discus-
sions, and to Webb Phillips and ‘Smooth’ Jesse Greene for their help with programming and
data collection. Thanks also to Olga Fotakopoulou and Ria Pita at the Aristotle University
of Thessaloniki, Greece for sharing their time and expertise. This research was supported in
part an NSF Graduate Research Fellowship and an NSF dissertation grant to the author.

Notes
1 Like our mental representations of time, some of our spatial representations may also
be quite abstract. For example, our conception of the Milky Way galaxy’s breadth is no
more grounded in direct experience than our conception of its age.
2 Cultural evolution alone cannot explain our capacity for abstract thought because, as
Wallace noted, members of ‘stone age’ societies who were given European educations
manifested abilities to similar those of modern Europeans: the latent capacity to read, to
perform Western art music, etc. was present in the minds of people whose cultures had
never developed these abstract forms of expression.
3 Experiments 1–6 are described in full in Casasanto, D. and Boroditsky, L. (2008). Time
in the mind: Using space to think about time. Cognition 106: 579–593.
4 A preliminary report on Experiments 7–9 appeared in Casasanto, D., Boroditsky, L.,
Phillips, W., Greene, J., Goswami, S., Bocanegra-Thiel, S., Santiago-Diaz, I., Fotoko-
poulu, O., Pita, R. and Gil, D. (2004). How deep are effects of language on thought? Time
estimation in speakers of English, Indonesian, Greek, and Spanish. Proceedings of the
26th Annual Conference of the Cognitive Science Society, Chicago, IL.
5 A preliminary report on Experiment 10 appeared in Casasanto, D. (2005) Perceptual
foundations of abstract thought. Doctoral dissertation, MIT.
476 LANGUAGE, COGNITION AND SPACE

6 Native speakers of European and South American Spanish report that largo tiempo is
only used in poetic contexts (e.g., the Peruvian national anthem) to mean ‘throughout
the length of history’. By contrast, some bilingual North American Spanish speakers
report that largo tiempo can be used colloquially, much like long time, perhaps because
the construction is imported from English.
7 A preliminary report on Experiment 11 appeared in Casasanto, D., W. Phillips and L.
Boroditsky, Do we think about music in terms of space: Metaphoric representation of
musical pitch. Proceedings of 25th Annual Conference of the Cognitive Science Society,
2003. Boston, MA.

References
Alverson, H. (1994) Semantics and experience: Universal metaphors of time in English,
Mandarin, Hindi, and Sesotho. Baltimore: Johns Hopkins University Press.
Boroditsky, L. (2000) Metaphoric structuring: Understanding time through spatial
metaphors. Cognition 75(1): 1–28.
Boroditsky, L. (2001) Does language shape thought? Mandarin and English speakers’
conceptions of time. Cognitive Psychology 43(1): 1–22.
Boroditsky, L. and Ramscar, M. (2002) The roles of body and mind in abstract
thought. Psychological Science 13(2): 185–189.
Cantor, N. and Thomas, E. (1977) Control of attention in the processing of temporal
and spatial information in complex visual patterns. Journal of Experimental
Psychology: Human Perception and Performance 3(2): 243–250.
Casasanto, D. (2005) Perceptual foundations of abstract thought. Doctoral disserta-
tion. MIT, Cambridge.
Casasanto, D. (2008) Similarity and proximity: When does close in space mean close
in mind? Memory & Cognition 36(6): 1047–1056.
Casasanto, D. (2009a) When is a linguistic metaphor a conceptual metaphor? In
V. Evans and S. Pourcel (eds) New directions in cognitive linguistics 127–145.
Amsterdam: John Benjamins.
Casasanto, D. (2009b) Embodiment of abstract concepts: Good and bad in right- and
left-handers. Journal of Experimental Psychology: General 138(3): 351–367.
Casasanto, D. and Boroditsky, L. (2008) Time in the mind: Using space to think
about time. Cognition 106, 579–593.
Casasanto, D., Boroditsky, L., Phillips, W., Greene, J., Goswami, S., Bocanegra-Thiel,
S., et al. (2004) How deep are effects of language on thought? Time estimation
in speakers of English, Indonesian, Greek, and Spanish. Paper presented at the
Cognitive Science Society, Chicago.
Casasanto, D., Phillips, W. and Boroditsky, L. (2003) Do we think about music in
terms of space: Metaphoric representation of musical pitch. Proceedings of 25th
Annual Conference of the Cognitive Science Society. Boston, MA.
Chomsky, N. (1975) Reflections on language. New York: Norton and Company.
Clark, E. (2003) Languages and representations. In D. Gentner and S. Goldin-
Meadow (eds) Language in mind 17–23. Cambridge: MIT Press.
Clark, H. H. (1973) Space, time, semantics and the child. In T. E. Moore (ed.)
Cognitive development and the acquisition of language 27–63. New York:
Academic Press.
 SPACE FOR THINKING 477

Cohen, J. (1967) Psychological time in health and disease. Springfield: Charles C.

Thomas.
Darwin, C. (1859/1998) The origin of species by means of natural selection or the pres-
ervation of favored races in the struggle for life. New York: The Modern Library.
Darwin, C. (1874/1998) The descent of man. Amherst: Promethius Books.
DeLong, A. (1981) Phenomenological space-time: Toward an experiential relativity.
Science 213(4508): 681–683.
Evans, V. (2004) The structure of time: Language, meaning and temporal cognition.
Amsterdam: John Benjamins.
Fodor, J. (1975) The language of thought. Cambridge: Harvard University Press.
Gentner, D. (2001) Spatial metaphors in temporal reasoning. In M. Gattis (ed.)
Spatial schemas and abstract thought 203–222. Cambridge: MIT Press.
Gibbs, R. W., Jr. (1994) The poetics of mind: Figurative thought, language, and under-
standing. Cambridge: Cambridge University Press.
Glicksohn, J. (2001) Temporal cognition and the phenomenology of time: A multipli-
cative function for apparent duration. Consciousness and Cognition 10: 1–25.
Gould, S. and Vrba, E. (1982) Exaptation – a missing term in the science of form.
Paleobiology 8: 4–15.
Gould, S. J. (1980) Natural selection and the brain: Darwin vs. Wallace. In The
panda’s thumb 47–58. New York: Norton.
Gould, S. J. (1991) Not necessarily a wing. In Bully for bronntosaurus 139–151. New
York: Norton and Co.
Gruber, J. (1965) Studies in lexical relations. Cambridge: MIT Press.
Ivry, R. and Richardson, T. (2002) Temporal control and coordination: The multiple
timer model. Brain and Cognition 48: 117–132.
Jackendoff, R. (1972) Semantic interpretation in generative grammar. Cambridge:
MIT Press.
Jackendoff, R. (1983) Semantics and cognition. Cambridge: MIT PRess.
James, W. (1890) Principles of psychology. New York: Rinehart and Winston.
Johnson, M. (2005) The philosophical significance of image schemas. In B. Hampe
(ed.) From perception to meaning: Image schemas in cognitive linguistics 15–33.
Berlin: Mouton de Gruyter.
Katz, J. and Fodor, J. (1963) The structure of a semantic theory. Language 39(2):
170–210.
Lafargue, P. (1898) Ursprung der abstrakten ideen. Die Neue Zeit, XVIII(2).
Lakoff, G. and Johnson, M. (1980) Metaphors we live by. Chicago: University of
Chicago Press.
Lakoff, G. and Johnson, M. (1999) Philosophy in the flesh: The embodied mind and its
challenge to western thought. Chicago: University of Chicago Press.
Langacker, R. (1987) An introduction to cognitive grammar. Cognitive Science 10:
1–40.
Levelt, W. (1989) Speaking: From intention to articulation. Cambridge: MIT Press.
Locke, J. (1689/1995) An essay concerning human understanding. Amherst:
Promethius Books.
478 LANGUAGE, COGNITION AND SPACE

Mach, E. (1896/1897) Contributions to the analysis of sensations. Chicago: Open

Court Publishing Company.
Meier, B. and Robinson, M. (2004) Why the sunny side is up: Associations between
affect and vertical position. Psychological Science 15(4): 243–247.
Meier, B., Robinson, M. and Clore, G. (2004) Why good guys wear white: Automatic
inferences about stimulus valence based on color. Psychological Science 15(1):
84–87.
Murphy, G. (1996) On metaphoric representation. Cognition 60: 173–204.
Murphy, G. (1997) Reasons to doubt the present evidence for metaphoric representa-
tion. Cognition 62: 99–108.
Ornstein, R. (1969) On the experience of time. Hammondsworth: Penguin.
Papafragou, A., Massey, C. and Gleitman, L. (2002) Shake, rattle, ‘n’ roll: The repre-
sentation of motion in language and cognition. Cognition 84: 189–219.
Pinker, S. (1989) Learnability and cognition: The acquisition of argument structure.
Cambridge, MA: MIT Press.
Pinker, S. (1994) Mentalese. In The language instinct 55–82. New York: Harper.
Pinker, S. (1997) How the mind works. New York: Norton.
Price-Williams, D. R. (1954) The kappa effect. Nature 173(4399): 363–364.
Pullum, G. (1991) The great Eskimo vocabulary hoax. Chicago: University of Chicago
Press.
Richardson, D., Spivey, M., Barsalou, L. and McRae, K. (2003) Spatial representa-
tions activated during real-time comprehension of verbs. Cognitive Science 27:
767–780.
Schubert, T. (2005) Your highness: Vertical positions as perceptual symbols of power.
Journal of Personality and Social Psychology 89(1): 1–21.
Slobin, D. (1996) From ‘thought and language’ to ‘thinking for speaking’. In J.
Gumperz and S. C. Levinson (eds) Rethinking linguistic relativity 70–96.
Cambridge: Cambridge University Press.
Snedeker, J. and Gleitman, L. (2004) Why is it hard to label our concepts? In Hall and
Waxman (eds) Weaving a lexicon. Cambridge: MIT Press.
Sommer, F. (1963) Types of ontology. Philosophical Review 72: 327–363.
Talmy, L. (1988) Force dynamics in language and cognition. Cognitive Science 12:
49–100.
Torralbo, A., Santiago, J. and Lupiáñez, J. (2006) Flexible conceptual projection of
time onto spatial frames of reference. Cognitive Science 30(4): 745–757.
Turner, M. (2005) The literal versus figurative dichotomy. In S. Coulson and B.
Lewandowska-Tomaszczyk (eds) The literal and nonliteral in language and
thought 25–52. Frankfurt: Peter Lang.
Wallace, A. (1870/2003) Contributions to the theory of natural selection. In
Contributions to the theory of natural selection 400. London: Routledge.
Whorf, B. L. (1939/2000) The relation of habitual thought and behavior to language.
In J. B. Carroll (ed.) Language, thought and reality: Selected writings of Benjamin
Lee Whorf 134–159. Cambridge, Massachusetts: MIT Press.
Zakay, D. and Block, R. (1997) Temporal cognition. Current Directions in
Psychological Science 6(1): 12–16.
18 Temporal frames of reference
Jörg Zinken

1 Introduction

Do people understand time in the same way across languages and cultures, or is our
understanding of time culturally specific? On the one hand, anthropologists have often
emphasised differences between the ways cultures interpret time (see Gell, 1992, and
Munn, 1992, for reviews of the literature). On the other hand, some of the problems that
conceptions of time address must be addressed by humans in all environments: human
life is finite all around the globe, and humans live in groups which need to coordinate
their activities. Maybe then there is some cognitive bedrock of thinking about time that
is the same across languages and cultures?
One way of finding out is to look for universals in the way people across languages
talk about time (Bloch, 1989). But although the anthropology of time is a vast research
field with a long history, a systematic linguistic anthropology of time is less developed
than one might expect (Levinson, 2004). This chapter discusses possibilities of making
one aspect of such a linguistic anthropology of time more systematic. In particular, I
will discuss possible heuristic contributions that typologies of spatial frames of reference
might make to typologies of temporal frames of reference.
The observation that in English and many other languages the vocabulary used
to talk about the location of objects in space is also used to talk about the location of
events in time has attracted considerable interest (Clark, 1973; Fillmore, 1997 [1971];
Jackendoff, 1983; Lakoff, 1993). More recently, the universality of such vocabulary
sharing has been hypothesised within the framework of Conceptual Metaphor Theory
(Lakoff and Johnson, 1999). Within this framework, cross-linguistic studies assess the
presence in the studied language(s) of metaphorical models such as TIME IS SPACE
(Radden, 2003), TIME AS SPACE (Yu, 1998), or TIME PASSING IS MOTION (Ahrens
and Huang, 2002).
These studies have begun to provide semantic evidence for universals in the cog-
nition of everyday time to supplement the abundant anthropological evidence for
diversity in time cognition in ritual contexts (Bloch, 1989; Senft, 1996). However, these
studies have also highlighted methodological problems. Global models such as TIME
IS SPACE or TIME PASSING IS MOTION need to be qualified and specified before
they can be appropriate frameworks for typological research, but such qualifications
and specifications have not been systematically made. These models need to be quali-
fied because, as they stand, they might suggest that abstract English concepts such as
time, space, or motion are universally relevant, which they clearly are not. The contexts
in which the word ‘time’ is used by speakers of English are diverse; although some of
these contexts might be universally relevant, others are unlikely to be (Evans, 2004).

479
480 LANGUAGE, COGNITION AND SPACE

Furthermore, these models need to be specified, because they are so general that con-
structions with diverse functions can be used as evidence for, for example, a TIME IS
SPACE model. Taken together, the cross-linguistic irrelevance of terms such as ‘time’,
‘space’ and ‘motion’, and the generality of the proposed models, can lead to research
that mirrors the unfortunate model of research into the universals of ‘colour’ terms:
constructions with diverse functions are forced into a framework that has no validity for
the languages studied (Saunders, 1995; Wierzbicka, 1996). As far as research on ‘spatial
time’ is concerned, this global approach has indeed led to the formulation of universals
on a rudimentary empirical basis. After all, nearly all of the languages in which the
polysemy of spatiotemporal lexemes has been studied are spoken by urban speakers in
industrialised societies (such as Chinese, English, Japanese, Turkish). Rare exceptions are
Malotki’s (1983) study of Hopi time, Moore’s (2000) study of spatial metaphors for time
in Wolof, and Nuñez and Sweetser’s (Núñez and Sweetser, 2006) study of Aymara. The
anthropological literature, in turn, contains many studies of time in non-industrialised
cultures (Munn, 1992). However, the linguistic descriptions provided in these studies
are usually not very detailed.
Models such as TIME PASSING IS MOTION might be inappropriate as a frame-
work for a linguistic anthropology of ‘spatial time’. Some framework, of course, is neces-
sary if we want to draw any (cross-linguistic) generalisations. The more detailed our
framework, the better our chances that we describe genuine cases of conceptualisation
rather than researcher-induced artefacts (Lucy, 1997). With this in mind, I want to
suggest here that existing typologies of spatial frames of reference can help in making
useful conceptual distinctions for a semantic typology of ‘spatial time’.
In the remainder of this introduction, I will briefly describe a distinction between
two kinds of time commonly used in the philosophy of time. Further, a typology of
spatial frames of reference will be briefly introduced. It will then be the aim of the main
body of this chapter to bring the two together in developing a typology of temporal
frames of reference that is detailed enough to serve as a framework for cross-linguistic
investigation and generalisation.

The A-series and the B-series of time

What is time, anyway? In order to make sense of diversity across languages and cultures,
we first need to have a good grasp of what we assume to be the universally experienced
aspect of the world that English speakers refer to when they employ the word ‘time’.
Philosophical answers to this question can be categorised into two broad groups: The
‘A-series’ view of time and the ‘B-series’ view of time. This classification can be traced
back to the philosopher McTaggart (1908), and it has been taken up more recently by
Gell (1992) in his anthropology of time. The brief discussion in this section is based
on Gell’s work.
Time can be thought of as a series of events. But what exactly is it about events that
gives them a temporal quality? Some philosophers argue that events constantly change
their status, from belonging to the future to belonging to the present to belonging to the
 TEMPORAL FRAMES OF REFERENCE 481

past. Time is this constant change in the status of events. The series of events constituting
time conceived of in this way is referred to as A-series time, and theorists arguing that
time is the flux of events from futurity through presentness into the past are referred
to as A-series theorists. Other philosophers argue that events never change their status;
they do not ‘become’ and ‘fade’, but simply ‘are’, like beads strung together on a necklace.
Time, on this view, is the set of relations of anteriority and posteriority holding between
events. The series of events constituting time conceived of in this way is referred to as
B-series time, and theorists arguing that time is a never-changing network of anteriority/
posteriority-relations are referred to as B-series theorists.
We hence end up with two kinds of time: the time of our subjective experience, for
which future events have a different meaning than past events (the A-series), and the
network of events as they objectively occur, quite independently of our interest or lack
of interest in them (the B-series). Philosophers debate which of these characterisations
reveals the ontological reality of time. A-series theorists argue that the A-series captures
the ontological reality of time: futurity is an intrinsic property of future events, and
pastness is an intrinsic property of past events. B-series theorists argue that the B-series
is ontologically real: events occur when they occur; futurity and pastness are assessments
which we bring to events due to our active orientation towards the world.
The distinction between an A-series and a B-series of time originates from a meta-
physical debate, i.e. the question which ‘kind’ of time is ‘basic’ and ontologically real. Of
course, the aim in this chapter is not to enter into metaphysical debates, but to provide
a framework for comparing the semantics of everyday time reference across contexts.
The distinction between A-series and B-series is only useful in the current context if it
can be translated into different types of everyday time reference.
Intuitively, it does seem that we make a distinction between the two kinds of time
– the A-series and the B-series – in our everyday life. The A-series is what we experi-
ence as we coordinate our everyday activities and grapple with the finiteness of our
existence – or of our time until the next deadline. The B-series is the real-world founda-
tion for a culture’s inventory of event-types embodied in calendars. Furthermore, the
future-present-past stream of the A-series and the before-after chain of the B-series are
expressed using different vocabularies, both of which are also employed for talking about
spatial relations. The A-series is the kind of time grammaticalised in many languages in
the category of tense, which in many languages is marked by morphemes derived from
motion verbs corresponding to English ‘come’ and ‘go’ (Bybee, 1994; Traugott, 1978).
Also, consider the following expressions of A-series time:
(1) a. I have a fun afternoon in front of me.
You have a hard week behind you.
b. I am looking forward to tomorrow.
I look back at my childhood.

In (1a), events are marked as being in the experiencer’s future or past by placing them
in front of the experiencer or behind him, respectively.1 In (1b), the experiencer’s active
orientation towards events in the future or in the past is expressed using perception
482 LANGUAGE, COGNITION AND SPACE

verbs: In the perceptual field in front of the experiencer, future events can be anticipated,
in the field behind the experiencer, past events can be scrutinised.
Notions of front and back are also used to talk about anteriority/posteriority rela-
tions (the B-series of time), however, using different expressions in English:

(2) a. The 21st April is before the 22nd April.

Thursday comes after Wednesday.
b. We’ll meet in the week following Easter.
Tuesday is ahead of Wednesday.

It is an unchanging quality of the 21st of April that it occurs before the 22nd of April (within
the year), and it is an unchanging quality of Thursday that it occurs after Wednesday
(within the week). The time at which I make the statements in (2a) does not matter for
the interpretation of the temporal reference – i.e. it is a reference to B-series time. In these
examples, a form historically expressing the spatial relation front is used to express the
temporal relation anteriority, and a form historically expressing the spatial relation back
is used to express the temporal relation posteriority. While before and after express static
relations, the same conceptualisation of ‘spatial time’ can be conventionally expressed
in English with terms expressing relations in motion events: In (2b), the posteriority
relation of the meeting to Easter is expressed by locating it ‘behind’ Easter using the
form following, and the anteriority relation of Tuesday to Wednesday is expressed by
locating it ‘in front of ’ Wednesday using the form ahead of.
Intuitively, and on the basis of some suggestive data as discussed above, it seems
that the distinction between two kinds of time is cognitively real for speakers of English.
For the purpose of this chapter, it will be assumed that both the experiencer-centred
understanding of time as a series of future, present, and past events, and the experiencer-
independent understanding of time as a series of before/after relations between events
are universal temporal experiences. Furthermore, we have seen that in both contexts
concepts of front and back are involved (at least historically) in temporal conceptualisa-
tion in English. The distinction between A-series and B-series might therefore be of
value for a typology of temporal frames of reference. However, the characterisations
provided in the philosophical and anthropological literature to explain how people
make temporal sense of these event-series are not precise enough for our purposes.
A-series time is characterised as a stream of events going past the experiencer. B-series
time is characterised as a static chain of events (Gell, 1992). While these metaphors
are suggestive, they are hardly a good basis for cross-linguistic comparison. We need
a more precise language to address our question: Where does the association between
the ideas of front and future in the case of A-series reference, and front and (temporal)
anteriority in the case of B-series reference come from? More specifically: what exactly
is the analogy between locating objects in space and locating events in time? To answer
these questions, we first need to find out what the logic of the reference systems is which
are used to locate objects in space.
 TEMPORAL FRAMES OF REFERENCE 483

Spatial frames of reference

Three frames for locating objects and places in space are commonly used across lan-
guages: The intrinsic or ground-based frame of reference, the absolute or field-based
frame of reference, and the relative or projector-based frame of reference (e.g., Levinson,
2003; Talmy, 2000).2 The brief description in this section is based mainly on the work
of Levinson (1996a; 2003).
Spatial frames of reference are constituted by three logical entities: the object to be
located (the figure), an object with a known location which is used to locate the figure
(the ground), and an object which determines the search space to be projected from
the ground (the origin of the coordinate system).
In the intrinsic frame of reference, ground and origin are conflated: the ground object
is also the origin of the coordinate system. For example, The computer is in front of me
locates the computer using an intrinsic frame of reference. ‘I’ am the referential ground,
and the asymmetry of my body also determines what is to be understood by the relator
front, i.e. how the search space is to be projected from me. Asymmetric inanimate objects
are also often thought of as having intrinsic fronts and backs. An utterance such as The
bike is in front of the house can be understood in this way. The front of an inanimate
object is often the side that people canonically interact with. In the case of houses, the
front side would typically be the side facing the street, where the door to the house is
located. Intrinsic frames of reference take diverse forms across languages, but the logic
of an intrinsic frame of reference seems to be universally used to locate objects in space.
The reason for this might be that intrinsic frames of reference are relatively simple:
because ground and origin are conflated, reference within an intrinsic system requires
the understanding of only a binary relation (Levinson, 1996a,b).
In the absolute frame of reference, the environment in which the ground object is
located provides a field which is organised in such a way that it can be used to deter-
mine a search space; the environment here constitutes the origin of the coordinate
system. Familiar examples are the cardinal points north, west, south, and east. The
utterance Hamburg is north of Bielefeld is comprehensible because the cardinal direc-
tions provide a grid running across the globe (and through Bielefeld, the referential
ground). But the environment used for absolute reference can also be more concrete
and localised. For example, a bowling lane can provide an absolute origin. Suppose a
team of weak bowlers have only managed to toss the bowls about half-way towards the
pins. Bowls lying still do not have intrinsic fronts, so I cannot use an intrinsic frame
of reference to locate a particular bowl in relation to another one. Still, I can refer
to the blue bowl lying behind the red bowl, meaning that it is further away from the
pins. Due to its directedness, the lane can serve as the field (or origin) of reference.
Finally, absolute origins can also be temporary: in the utterance John is behind Mary
in the queue, the directedness of the queue determines how the relator behind is to
be understood (Talmy, 2000).
In the relative frame of reference, an observer constitutes the origin of the coordi-
nate system. The speaker’s coordinates front, back, left and right are projected onto the
484 LANGUAGE, COGNITION AND SPACE

referential ground. The details of this projection differ across and within languages. For
example, the speaker’s coordinates can be ‘reflected’ from the ground, as if the ground
object was another observer ‘facing’ the actual observer. The utterance The ball is in front
of the tree is understood in this way: the ball is between the tree and the observer, the
tree’s front is the side ‘facing’ the observer. However, in other contexts, the projection
involves not reflection but translation, where the orientation of the observer is ‘carried
over’ onto the ground object: A ball to the left of the tree is to the left from the observer’s
point of view, not from the point of view of an observer ‘reflected’ in the tree.

2 Temporal frames of reference

Do analogous temporal frames of reference exist? Can the technical terms as elaborated in
work on spatial conceptualisation be of heuristic value in the description of space-time
analogies used for temporal reference across languages? In this section, I aim to develop
a typology of everyday (spatio-)temporal frames of reference on the background of the
philosophical distinction between the A-series and the B-series of time.

Locating events in A-series time

‘A-series’ time is the subjective experience of a constant change in the status of
events, from their futurity to presentness, to pastness. The futurity or pastness of an
event can conventionally be expressed in English by placing the event in front of or
behind the experiencer respectively, as in (1) above, repeated here:

(1) I have a fun afternoon in front of me.

You have a hard week behind you.

As we can now see, these utterances locate events within an intrinsic frame of reference.
The defining feature of intrinsic frames of reference is their binary structure (Levinson,
2003): a figure entity is located in relation to a ground entity, and the ground is also
the origin of the coordinate system. From meetings, through afternoons and years, to
a whole life, events of varying regularity and temporal scope can conventionally be
referred to as lying in front of us or behind us in English. It seems that for speakers of
English, large-scale time intervals (such as days, seasons, the duration of the world) are
abstracted from the actual environment as an additional, imaginary ‘landscape’ on
which events can be (quasi-)visualised. 3 The conceptualisation of large-scale temporal
intervals as a landscape affords the use of a viewer-centric, relative frame of reference for
the localisation of events. The relative frame of spatial reference locates an object with
respect to the ground from the point of view of an observer. With respect to the location
of events in time, some authors have proposed that expressions such as the day after
tomorrow are understood in a relative frame of reference (Radden, 2003: 12). Radden
illustrates the spatial logic of this expression with the following figure:
 TEMPORAL FRAMES OF REFERENCE 485

PAST PRESENT FUTURE

D-3 D-2 D-1 O D+1 D+2 D+3

D-3 – D+3: Days

O: Location of the observer at day 0
Figure 1. A vision-based understanding of temporal relations (adapted from Radden, 2003)

The blocks in Figure 1 symbolise days, and the day after tomorrow is the one ‘behind’
the one the observer is ‘looking’ at, tomorrow. In other words, tomorrow is the (pri-
mary) referential ground, the speaker’s now is the origin of the coordinate system. The
deictic nature of tomorrow surely supports a relative reading in Radden’s example, but
the present argument should apply also to the day after Tuesday. This relation can be
understood in a relative, quasi-visual manner, if it is understood that I am talking about
a particular future Tuesday.4
The function of using a relative frame of reference to locate events in A-series
time might be that it allows more precision when talking about large-scale time inter-
vals beyond the now than the intrinsic frame of reference and deictic expressions do.
When talking about plans for the immediate future, we are more likely to use a deictic
expression without a frame of reference: We would say I’ll send this e-mail in a moment
rather than, e.g., I’ll send this e-mail 20 seconds before a minute has passed. However,
in time-scales which go beyond the now (a border that is itself likely to vary across
cultures, within and across language communities), simple deictics are not very useful:
I’ll get some crisps before the match is a more relevant information than the deictic I’ll
get some crisps in 3 hours’ time.

Locating events in B-series time

The B-series of events is the time of anteriority/posteriority relations. While events arise
in and fade from the field of our experience, their temporal relations to all other events
never changes. In English, the words ‘before’ and ‘after’ express relations of anteriority
and posteriority respectively. Let’s take a closer look at the example just used:

(3) I’ll get some crisps before the match.

Example (3) can be understood in a relative, quasi-visual manner, as discussed in the

preceding section. When a future ground event (‘match’) and the observer ‘face’ each
other as in a canonical encounter, a figure event (‘getting crisps’) between the ground
event and the observer is ‘in front of ’ the ground event. The temporal relation between
crisps-getting and match in (3) might therefore be understood in a way that is analogous
to the spatial reference in The ball is in front of the tree.
486 LANGUAGE, COGNITION AND SPACE

However, in a relative frame of reference the origin of the coordinate system is the
observer. A spatial scenario anchored in an observer is incoherent with the observer-
independent nature of relations in B-series time. If it was the case that we could under-
stand expressions such as I’ll get some crisps before the match only in an observer-centred
manner, it would mean that we could not express the observer-independent nature of
B-series time. But this is implausible: we can easily be aware of the unchanging anterior-
ity of the crisps-getting event relative to the match. The question is therefore whether
the temporal relation between getting crisps and watching a match in (3) is necessarily
understood in a relative manner.
In their work on the use of spatial frames of reference, Levinson and colleagues have
employed rotation tasks to distinguish between different frames of reference (Levinson,
2003). Would it be possible to use an analogue of such rotation tasks to find out what
kind of coordinate system is used in temporal reference? In order to do this, we would
need to refer to the same figure-ground relation from the opposite temporal perspective.
Suppose, then, that I want to refer to the figure event in (3) in relation to the same
ground event a day later. ‘Looking back’ onto the same events, I now say: ‘Yesterday, I got
crisps _____ the match’. If we understood the temporal relations between these events
in an observer-anchored manner, the correct word to fill the gap would now have to be
‘after’ or another expression of behind-ness, because the figure event (getting crisps) is
now no longer between the match-watching event and me. However, the correct word
to fill the gap remains ‘before’. 5
This suggests that temporal relations between past events are understood in English
employing a coordinate system that is independent of the observer, and while temporal
relations between future events can be understood in a relative manner, they, too, should
be understandable in an experiencer-independent way. In other words, the spatial logic
of the temporal reference in I’ll get some crisps before the match might be ambiguous.
While deictic cues such as future tense and adverbials such as ‘tomorrow’ might prompt
a relative understanding, non-finite expressions (I always get crisps before a match) might
prompt an experiencer-independent understanding.
But how do speakers locate events ‘in front of ’ (‘before’) other events in an observer-
independent frame of reference? One possibility is that speakers make use of an intrinsic
frame of reference in these contexts (Bender, Bennardo and Beller, 2005; Yu, 1998).
When locating objects in space in English, expressions of front (‘in front of ’) and back
(‘behind’) can be used in this way. For example, the spatial reference in He is sitting
in front of the TV in most situations is intended in its intrinsic, rather than relative
interpretation. Spatial reference in an intrinsic frame of reference is independent of
the position of the observer: the referential ground object constitutes the origin of the
coordinate system.
In order to locate events in time in an intrinsic frame of reference where the
observer’s now is not part of the referential scene, we would need to be able to identify
the search interval in which the figure event (getting crisps in (3)) takes place on the
basis of intrinsic features of the ground event (watching a match). Can an event have
an intrinsic front? The fact that this sounds like a funny idea should not deter us from
entertaining the possibility. After all, the notion of intrinsic front is problematic also
 TEMPORAL FRAMES OF REFERENCE 487

when applied to objects: The front side of a TV is not really intrinsic to the physical
object, but determined by the way people canonically interact with TVs (Levinson,
2003).
It could be that the conceptualisation of events as ‘moving’ suffices to assign a front
side to them. Consider symmetrical objects. Like events, balls do not have intrinsic
fronts. Nevertheless, when a ball is rolling, we easily assign a front and back based on
the direction of motion (Fillmore, 1997 [1971]; Svorou, 1994); thus, football players
run behind a ball. Similarly, events in B-series time might be conceptualised as a train,
with each carriage representing an event (Yu, 1998). Irrespective of the position of an
observer, the first cabin, defined by the direction of motion, will always be in front of
the second one; similarly, anterior events always remain ‘in front of ’ posterior ones.
This account suggests that by invoking the idea of ‘moving events’ we can understand
the space-time analogy in an expression like I always get crisps before a match within a
binary figure-ground frame, i.e. in an intrinsic frame of reference. Ultimately, intrinsic
temporal reference might be based in quite literally spatial front/back relations: The
sun moves across the sky ahead of the moon. Day comes before night, night comes after
day, and one day comes after the other might be the clearest cases of such motion-based
intrinsic temporal reference.
But is it plausible to assume that speakers of English conceive of the events in I
always get crisps before a match as moving? This seems counterintuitive, and, to be sure,
it is not evident from linguistic data. While it is conventional to speak of calendaric
event types (Christmas, spring) and other event types that are part of a natural cycle
(the evening, the morning) as coming and going by, the same is much less felicitous when
applied to singular events (?The match is coming). A more prudent account might be
to suggest that I always get crisps before a match is understood in an absolute frame of
reference, with the day as the origin of the coordinate system. The before relation means
that the crisps-getting event is closer to the beginning of the day, the implicit second-
ary reference interval, than the match, the primary reference event. Conventionalised
intervals, such as the day, provide a directed field for such absolute reference, in analogy
to people in a queue (Talmy, 2000). Although the directedness of a queue ultimately
derives from the canonical movement towards the goal of this queue, it maintains its
directedness even when there is no motion. Similarly, events throughout a day can be
thought of as ‘adding up’ one after the other at their specified dates6 along the temporal
field much like people forming a queue do, rather than as moving like bowls rolling one
behind the other on a bowling lane. Such a ‘motionless’ account is advantageous also
because some languages do not seem to use the idea of objects moving through space
to think about temporal relations between events (Bohnemeyer, 1997). However, we
would not want to deny speakers of such languages the ability to speak or think about
unchanging relations in B-series time.
Absolute reference to sequentiality relations does not require the speaker/hearer to
specify a particular directedness of the field along which events are located. However,
such directedness is necessary when communicating visually, rather than vocally,
about (B-series) time, e.g., in co-speech gesture. In cultures with a writing system, the
direction seems to be imported from the relevant conventions of using visual media,
488 LANGUAGE, COGNITION AND SPACE

such as written language or comics. For example, speakers of Spanish assume by default
that events displayed on the left side of a computer screen happened earlier than events
displayed on the right side of a computer screen (Santiago, 2005). Arabic speakers asked
to arrange objects representing a day’s activities on a plane arrange these from right to
left (Tversky, Kugelmass and Winter, 1991). Speakers of Mandarin produce downwards
gestures when talking about a time in the afternoon (irrespective of whether it is the
afternoon of the same, a future, or a past day), and upwards gestures when talking
about the morning.7 With respect to the question of the metaphoricity of temporal
understanding, it is important to bear in mind that such figurative specifications of
temporal ‘directions’ are not part of the conceptual structure employed in thinking
for speaking, but of that employed in thinking for gesturing, i.e. in a visual medium of
communication, which is by necessity one big ‘spatial metaphor’.8

3 Temporal frames of reference and other generalisations

Probably the most widely used generalisations in the study of space-time analogies across
languages are the Moving Time and Moving Ego metaphors, first introduced by Clark
(1973) and Fillmore (1997 [1971]). These two models have been reformulated in various
ways, as TIME PASSING IS MOTION OVER A LANDSCAPE, TIME PASSING IS
MOTION OF AN OBJECT, and the further generalisation TIME PASSING IS MOTION
(Lakoff, 1993). In the Moving Time model, time is viewed as a ‘highway consisting of a
succession of discrete events’ that are ‘moving past us from front to back’. In the Moving
Ego model ‘we are moving along [time], with future time ahead of us and the past
behind us’ (Clark, 1973: 50). Both of these models thus describe the A-series of time. In
terms of the frames of reference introduced here, both models involve an intrinsic (or
possibly relative) frame of reference, and combine this with the idea of motion – either
the motion of events, or the motion of the experiencer. However, while such models
might indeed be operative for speakers of English, it is usually not possible to conclude
this from linguistic data. Temporal reference often employs motion constructions, as
for example in the utterance the evening is coming. This is a deictic reference which does
not employ a frame of reference. Temporal reference also often does employ frames of
reference, for example the intrinsic one in the utterance I have a great evening in front
of me. But it is not common to talk of time as moving past the experiencer from front
to back (?A great evening is coming in front of me). In cross-linguistic research, it would
be more prudent to treat these two examples as different types of temporal reference
rather than as evidence for one general TIME PASSING IS MOTION model. Such
generalisations are better treated as complex models, i.e. as combinations of several
more fundamental conceptualisations. Evidence for such complex models must be
sought in non-verbal data.
Some authors have proposed distinguishing two frames of reference used for locat-
ing events in time: an ego-based or ego-reference-point (ego-RP) frame and a time-based
or time-reference-point (time-RP) frame (Moore, 2000; Núñez and Sweetser, 2006).
This terminology is somewhat unclear in so far as there are two reference points (or
 TEMPORAL FRAMES OF REFERENCE 489

reference intervals) in temporal reference: a primary one, the ground, and a secondary
one, the origin of the coordinate system (Talmy, 2000, Levinson, 2003). The explication
in Núñez and Sweetser (2006) suggests that the reference point they have in mind is the
primary reference point, or ground of reference. If the RP in the suggested distinction
between ego- and time-RP is to be understood as the primary reference point, the
English examples discussed in this chapter should be classified as follows:
Table 1. The primary reference point as the basis for classification

ego-RP (ego=primary RP) time-RP (event=primary RP)

I have a fun afternoon in front of me The 21st April is before the 22nd April
- Wednesday is after Tuesday
One day comes after the other
- The day after tomorrow
- I’ll get some crisps before the match
- I always get crisps before a match

This classification seems wrong. It is the explicit aim of Núñez and Sweetser (2006) to
separate reference to subjective past or future from reference to anteriority/posteriority
relations. In this respect, Wednesday is after Tuesday, which describes sequentiality,
should not be in the same category with The day after tomorrow, which refers to the
speaker’s future.
It seems to me that what Núñez and Sweetser (2006) actually have in mind is a
distinction that is similar to the one between the A-series and the B-series in the philoso-
phy and anthropology of time. If this is correct, the reference point in question would
be the secondary reference point, or the origin of the coordinate system. The English
examples discussed in this chapter would then fall into the two categories as follows:
Table 2. The secondary reference point as the basis for classification

ego-RP (ego=secondary RP) time-RP (event=secondary RP)

I have a fun afternoon in front of me The 21st April is before the 22nd April (RP=month)
The day after tomorrow Wednesday is after Tuesday (RP=week)
One day comes after the other (RP=day)
I’ll get some crisps before the match I always get crisps before a match (RP=day)

The distinction between the A-series and the B-series, or between types of secondary
reference points as conceived in table 2, is an important one, the lack of which has led
to a confusion of (A-series) past with (B-series) anteriority, and future with posteriority
in earlier research (see the next section). However, as a typology of systems for locating
events in time it is less precise than the distinction between the intrinsic, relative, and
absolute frame of reference. In this classification, I have a fun afternoon in front of me and
490 LANGUAGE, COGNITION AND SPACE

I’ll get some crisps before the match are grouped together as the same type of reference.
However, they do differ in the way they locate events in A-series time. The relative frame
of reference allows more specific reference to the temporal location of events in A-series
time beyond the now. The price for this is an increase in cognitive complexity: while
the intrinsic relator ‘in front of ’ specifies a binary relation, the relative relator ‘before’
specifies a ternary relation. Similarly, the statement that Wednesday is after Tuesday is
only true within the absolute frame of the week, whereas the reference one day comes
after the other probably makes use of an intrinsic frame of reference.
Ultimately, it seems that frameworks that are based on the quality of a particular
reference point run into problems. In this chapter, I have argued that it might be better
to use a typology that is based on types of coordinate systems. In sum, the classification
that I propose looks like this:
Table 3. Temporal frames or reference

A-series B-series
I have a fun afternoon in Coordinates: intrinsic One day comes after the other Coordinates: intrinsic
front of me Origin: speaker Origin: day
PRP: speaker PRP: day

You have a tough week Coordinates: intrinsic Wednesday is after Tuesday Coordinates: absolute
behind you Origin: addressee Origin: week
PRP: addressee PRP: Tuesday

I’ll get some crisps before Coordinates: relative I always get crisps before a match Coordinates: absolute
the match Origin: speaker Origin: day
PRP: match PRP: match

The day after tomorrow Coordinates: relative

Origin: speaker
PRP: tomorrow

The distinction between experiencer-centred (A-series, ego-RP) and experiencer-

independent (B-series, time-RP) time together with a typology of frames of reference that
are used to construct these ‘kinds’ of time provide a reasonably fine-grained framework
for the systematic exploration of universals and diversity in space-time analogies.

4 Universals and diversity in spatial time

We are now in a position to discuss how the distinctions made in this chapter can
help us to integrate existing data, ask new questions, and formulate hypotheses about
universals of spatial time.
Forms expressing spatial relations of front and back regularly express anteriority
and posteriority across languages. Furthermore, it seems that, as in English, expressions
 TEMPORAL FRAMES OF REFERENCE 491

of front always express anteriority, and expressions of back always express posteriority
(Haspelmath, 1997).9 A few examples are presented in (4–6).

(4) Kwaio (Keesing, 1991)

(a) na’o-na mae i Gwee’abe
‚before the battle at Gwee’abe’, literally, front-of battle’
(b) buri-na mae i Gwee’abe
‚after the battle at Gwee’abe’, literally ‘back-of battle’

(5) Hopi (Malotki, 1983)

pam put hihin a-pyeve tìi-ti-wa
that that somewhat he-before10 child-CAUS-PASS.PERF
‘He was born a little bit before him’

(6) Wolof (Moore, 2000)

Ci gannaaw la ñów.
LOCPREP back/behind NONSUBJ.FOC.3 come.
‘At back she came.’
‘She came afterwards’

Temporal relations of sequentiality (B-series time) using the relators front and back can
be understood in an absolute frame of reference, and possibly in an intrinsic frame of
reference, if speakers think of events as ‘moving’. As in the domain of space, the linguistic
data alone do not allow us to decide whether the expressions in (4–6) are understood
absolutely or intrinsically. We need additional data sources to answer this question.
Unfortunately, the use of rotation tasks, which make it possible to distinguish between
spatial frames of reference, has its limits in the domain of time. Alternatively, co-speech
gesture might be a valuable source of data, which can answer the question, e.g., whether
speakers habitually think of events as moving.
Similarly, it seems premature on the basis of our current knowledge to be too
sure that all languages use the relators front and back to express sequential relations.
Some languages might not at all explicitly mark anteriority and posteriority relations,
relying instead on context and iconicity: the event mentioned earlier happened earlier
(Bohnemeyer, 1997). Furthermore, languages which prefer an absolute frame for spatial
reference based on the movement of the sun might use the same vocabulary to talk
about sequentiality in time (the morning is east of the evening).
Absolute temporal reference requires that temporal intervals, which provide the
secondary ground, i.e. the origin of the coordinate system, be understood as bounded
entities, or ‘fields’. The fundamental space-time analogy is that between the beginning
of the unfolding of a temporal interval and a field’s front. The reason for the possible
universality of absolute temporal reference might be that relations within an absolute
system remain the same when the ‘viewpoint’ of the observer changes. In a relative frame
of reference, relations between figure and ground change when the observer’s viewpoint
492 LANGUAGE, COGNITION AND SPACE

changes. Of course, for human experiencers it is quite impossible to hold their temporal
‘viewpoint’ onto the world constant. An absolute frame of reference might therefore be
the only viable system for talking about unchanging anteriority/posteriority relations.
Events in A-series time can be located in an intrinsic frame of reference. Could
the particular association of an experiencer’s front with his or her future, and an expe-
riencer’s back with his or her past, be universal? There is some ground for entertain-
ing such an assumption. Anthropologists of time maintain that thinking about the
immediate future, i.e. the work at hand, the time that is still part of the now rather
than a then, is the fundamental context from which societal organisations of time
arise (see Gell, 1992). The immediate future is constantly apprehended and enacted
in (spatial) practice, e.g., manual work and gaze. It seems plausible enough to think
that speakers universally might use their front space symbolically, e.g., for gesture-
supported planning of imminent tasks, but we know too little about temporal cognition
across cultures to say. In this context, it is questionable whether we are dealing with
a metaphorical association between front space and immediate future. After all, the
idea that we do actually perceive the immediate future has been discussed at least
since Husserl introduced the notion of protentional consciousness, which anticipates
what lies just at the boundaries of the now, the current time interval. In so far as gaze
and manipulation are relevant at all for protention, this form of consciousness will be
directed to the body’s front space.
However, not only the immediate future which is still part of the now is conceptu-
alised as being in front of the experiencer in English. Large intervals of subjective time,
such as the future, also ‘lie’ in front of us. The orientation towards large time-‘scales’,
such as the relatively abstract English concept of future is very different perceptually,
conceptually, and linguistically from thinking about immediate future – it involves an
imaginary ‘leap’, the abstraction of conventional time intervals as an additional dimen-
sion. With relation to such ‘larger-scale’ temporal concepts, the association of front
with future is not universal. Several authors have claimed that in particular languages
and cultures, subjective future is conceptualised as lying behind the speaker, whereas
past events are in the observer’s visual field on a temporal landscape (Alverson, 1994;
Clifford, 2004; Dahl, 1995; Klein, 1987; Miracle and Yapita Moya, 1981). The linguistic
analyses supporting such arguments are often sketchy, and seem at times to have been
misguided due to conceptual confusions between A-series and B-series time (for criti-
cal reviews, see Moore, 2000; Núñez and Sweetser, 2006; Shinohara, 1999). However,
the analysis by Miracle and Yapita Moya in relation to Aymara has been supported by
converging evidence from co-speech gesture research (Núñez and Sweetser, 2006).
Núñez and Sweetser found that Aymara speakers would produce hand gestures forward
from their body when talking about past events in the community’s history, but would
produce gestures towards their back when explicating the meaning of the word future.
It seems plausible that the grade of figurativity of a temporal conceptualisation is
related to its cultural specificity, such that the more figurative the analogy between spatial
and temporal relations, the more restricted it is across cultures. Metaphoric ‘leaps’ seem
to require a strong cultural scaffolding to be successful (see also Evans and Wilkins,
2000). A cultural factor that might contribute to the gestures of Aymara speakers is the
 TEMPORAL FRAMES OF REFERENCE 493

emphasis that is placed on being precise about the source of one’s knowledge, which
is grammaticalised in Aymara in the category of evidentiality (see Aikhenvald, 2004).
When Aymara speakers make a predication, their grammar requires them to mark
whether they have seen the reported event themselves or not (Miracle and Yapita Moya,
1981). Since predications about future events are necessarily predictions, they cannot
have been eye-witnessed, which might contribute to their being conceptualised as lying
behind one’s back.
Finally, events in A-series time can be located in a relative frame of reference in
English and related languages. Whether this also occurs more widely across languages is
impossible to say, because, as in English, the linguistic form alone might not be sufficient
to tell whether an utterance is understood in a relative or an absolute frame of reference.
Relative frames of spatial reference differ across cultures in the way in which the
observer’s coordinates are mapped onto the referential ground, as described earlier. In
Hausa, this mapping of coordinates involves translation whereas it involves reflection
in English. The ball is on front of the tree under a ‘translational’ understanding means
that it is on the other side of the tree from where the speaker is standing. Temporal
reference in a relative coordinate system seems to show analogous diversity (Bender,
Bennardo and Beller, 2005; Hill, 1978). In Hausa, speakers ‘view a later day of the week
as gaba da ‘in front of/before’ an earlier one, an earlier day as baya da ‘in back/ of after’
[sic] a later one’ (Hill, 1978: 528). Hill does not specify whether Wednesdays are always
‘in front of ’ Tuesdays (in which case it would be an expression of ‘B-series’ time, and
as such independent from any experiencer, i.e. it could not be an instance of a relative
frame of reference at all), or whether this applies only to days in Ego’s future, in which
case it is a relative (translational) expression of relations in A-series time. Hill’s analysis,
though, suggests that the latter is the case.
Why is the future beyond the now located in front of us in English and related
languages? One factor might be the importance that is placed on the precise planning
of one’s (future) time in our culture. The quasi-visual, if only imagined, access that the
relative frame of reference imposes on temporal conceptualisation of future supports
such planning by providing a ‘space’ that can be used for planning in imagination, and
for temporal reference in language and in co-speech gesture. The use of a relative frame
of reference converts relations in B-series time (relations of anteriority and posterior-
ity between events) into an imagined space that is subject to personal planning and
time-‘reckoning’.
Languages might differ not only in terms of the overall repertoire and the precise
characteristics of temporal frames of reference, but also in terms of the preferred frame
of reference for a given context. Such differences might exist even between closely
related languages. For example, it seems that speakers of German prefer an absolute
frame of reference where speakers of English frequently use a relative frame of reference.
When asked to disambiguate the sentence The meeting planned for next Wednesday
has been moved forward two days, some speakers of English interpret forward to mean
later, as would be expected when using a relative (translational) perspective, whereas
others interpret forward to mean earlier, as would be expected when using an absolute
perspective, with the (beginning of the) week as the origin of the coordinate system
494 LANGUAGE, COGNITION AND SPACE

(see McGlone and Harding, 1998). Speakers of German, in a separate experiment,

consistently chose the absolute solution (Bender, Bennardo and Beller, 2005).
These differences might be indicative of a more general preference for viewer-
centred time reference in the case of English, and event-centred time reference in the
case of German. Stutterheim, Carroll, and Klein (2003) found that speakers of English
predominantly chose a viewer-centred strategy when re-telling the events in a short
film, in which the film is retold as if it was playing again before the mind’s eyes, with
new events introduced with a ‘and now I see’ phrase. Speakers of German, on the other
hand, predominantly chose a strategy which meant that they seemed to arrange the
events ‘like a string of pearls’ (Stutterheim, Carroll and Klein, 2003: 108) and mark the
posteriority of a new event with an ‘and then’ phrase. Stutterheim, Carroll, and Klein
(2003) relate these differences to the grammaticalisation of an ‘ongoing’ aspect in the
English progressive –ing form, which is absent in German.

5 Conclusion

In this chapter, I have suggested some conceptual distinctions that might be useful for
systematic data collection and analysis in cross-linguistic research on ‘spatial time’. The
suggested framework integrates previous data and opens a range of new questions: Are
anteriority/posteriority relations always understood in an absolute frame of reference,
or can they be understood in an intrinsic frame of reference? Are the A-series and the
B-series universal kinds of time, or do contexts of temporal reasoning exist which are
constructed as ‘A-series’ by using a relative frame of reference in one language, but are
constructed as ‘B-series’ by using an absolute frame of reference in another language?
What are the relations between types of time intervals (cyclic vs. non-cyclic; ‘punctual’
moments vs. longer events; events in the immediate future vs. events in the further
future vs. events in the past) and the use of different frames of reference? Systematic
data from a more varied sample of languages and cultures are needed before we can
attempt empirically grounded conclusions about possible universals in the domain of
space-time analogies.
The implicit or explicit (Bloch, 1989) assumption in the anthropology of time has
been that time reference in everyday contexts, as opposed to ritual contexts studied
intensively in the anthropological literature, might display many universalities across
languages and cultures. However, everyday life is a complex beast, and to make sure that
we compare like with like across languages, we need to distinguish not only experiencer-
centred (A-series, ego-RP) time from experiencer-independent (B-series, time-RP)
time, but also frames of reference and the contexts in which they operate. We might
find that a relativised ‘view’ of a temporal landscape stretching out into the future in
front of us is not so much a universal and natural feature of human mind, but rather
an exotic development in cultures that have developed a strong interest in ‘reckoning’
and ‘telling’ time.
 TEMPORAL FRAMES OF REFERENCE 495

Acknowledgements
I am grateful to Alan Costall, Kevin Moore, Vera da Silva Sinha, Chris Sinha, and an
anonymous reviewer for their valuable comments on an earlier draft. Work on this chapter
was supported by grants from the British Academy and the European Union’s 6th framework
programme ‘What it means to be human’.

Notes
1 This chapter discusses only one aspect of the complex ways in which the temporal
structures of events are communicated, and the example utterances are kept simple to
avoid some of these complexities. In terms of Klein’s (1994) approach to the commu-
nication of temporal relations, we deal here only with temporal relations between time
intervals in TSIT, the ‘time of the situation’ talked about. The complex relations between
TSIT, ‘topic time’ (TT) and ‘temporal anchor’ (TA) of the speech event are beyond the
scope of this chapter. But these complexities at least need to be acknowledged. Thus, in
an utterance like ‘Once, I had a great future in front of me’, we might still say that ‘front’
expresses futurity within TSIT, but tense and the temporal adverbial ‘once’ establish TT
as lying in the past relative to TA, the time of the speech event.
2 Levinson and his research group distinguish between intrinsic, absolute, and relative
frames of reference, while Talmy has introduced the distinctions between ground-
based, field-based, and projector-based reference. These terms overlap to a large extent
and can, for present purposes, be treated as synonymous. To minimize confusion I will
employ the terminology of intrinsic, absolute, and relative frames of reference.
3 It may be that in cultures where vision is not conceptualised as the most central modal-
ity in the acquisition of knowledge (see Evans and Wilkins, 2000) concepts of temporal
intervals as a ‘landscape’ are less relevant.
4 The quasi-visual conceptualisation of future time in English is further illustrated by the
use of visual perception verbs in conventional expressions such as I’m looking forward to
the time after Easter.
5 Of course, we might argue that the spatial meaning of ‘before’ is not relevant in this
case, and that the relevant meaning is, say, ‘earlier than X ground ’. However, the next
question then becomes: why has ‘before’ acquired the general meaning ‘earlier than
X ground ’ rather than ‘temporally between now and X ground)’?
6 The term ‘date’ as used here derives from the philosophy of time (see Gell, 1992). It
refers to the real-world spatio-temporal coordinates of an event, and does not imply the
existence of a calendar, as the everyday use of the word ‘date’ does.
7 I am only aware of anecdotal evidence for this so far. However, the association between
up-down relations and anteriority-posteriority relations in Mandarin is also evidenced
in conventional expressions, such as ‘shang-ban-tian’, literally upper-half-day, meaning
‘morning; forenoon’ (Yu, 1998, p. 110).
8 The claim that temporal relations between events beyond immediate future can be
understood in an absolute as well as a relative manner is currently based on intuition
and linguistic data. It should be possible to obtain independent evidence by operational-
ising co-speech gesture. Speakers of English and related languages tend to make left-to-
right gestures when talking about sequences in B-series time, but they produce forward
496 LANGUAGE, COGNITION AND SPACE

gestures when talking about sequences in A-series time. If event sequences which are
part of a speakers time plan are conceptualized in a ‘visualised’, relative way (such as
the tasks on a given day), but event sequences not immediately relevant to personal time
planning (such as, maybe, the seasons) are located in a field-based frame of reference,
speakers’ co-speech gestures should differ across these two contexts.
9 Haspelmath (1997) provides examples of adverbials expressing both temporal and
spatial anteriority and posteriority from a sample of 55 languages. He states that ‘almost
all cases’ (p. 56) follow this path, but he does not provide an example of a different case.
10 Malotki (1983) points out that the morpheme -pyeve ‘before’ itself is related to the
locative suffi x –ve, meaning ‘before a (moving) object’. According to Malotki (p. 92–93),
the antonymic suffi xal element –ngk ‘after a (moving) object’ means temporally after in
a sequence. However, Malotki does not provide an example for this use.

References
Ahrens, K. and Huang, C.-R. (2002) Time passing is motion. Language and
Linguistics 3(3): 491–519.
Aikhenvald, A. Y. (2004) Evidentiality. Oxford: Oxford University Press.
Alverson, H. (1994) Semantics and experience: Universal metaphors of time in English,
Mandarin, Hindi, and Sesotho. Baltimore and London: John Hopkins University
Press.
Bender, A., Bennardo, G. and Beller, S. (2005) Spatial frames of reference for tempo-
ral relations: A conceptual analysis in English, German, and Togan. In B. G. Bara,
L. Barsalou and M. Bucciarelli (eds) Proceedings of the Twenty-seventh Annual
Conference of the Cognitive Science Society 220–225. Mahwah, NJ: Lawrence
Erlbaum.
Bloch, M. (1989) The past and the present in the present. In M. Bloch (ed.) Ritual,
history and power: Selected papers in anthropology 1–18. London and Atlantic
Highlands, NJ: The Athlone Press.
Bohnemeyer, J. (1997) Yucatec Mayan lexicalization patterns in time and space. Paper
presented at the CLS opening academic year 1997/1998.
Bybee, J., Perkins, Revere and Pagliuca, W. (1994) The evolution of grammar: Tense,
aspect, and modality in the languages of the world. Chicago: University of Chicago
Press.
Clark, H. H. (1973) Space, time, semantics, and the child. In T. E. Moore (ed.)
Cognitive development and the acquisition of language 27–63. New York:
Academic Press.
Clifford, J. (2004) Traditional futures. In M. S. Phillips and G. Schochet (eds)
Questions of tradition 152–168. Toronto: Univeristy of Toronto Press.
Dahl, O. (1995) When the future comes from behind: Malagasy and other time
concepts and some consequences for communication. International Journal of
Intercultural Relations 19(2): 197–209.
Evans, N. and Wilkins, D. (2000) In the mind’s ear: The semantic extensions of
perception verbs in Australian languages. Language 76(3): 546–592.
Evans, V. (2004) The structure of time: Language, meaning and temporal cognition.
Amsterdam, PA: John Benjamins.
 TEMPORAL FRAMES OF REFERENCE 497

Fillmore, C. (1997 [1971]) Lectures on deixis. Stanford, CA: CSLI Publications.

Gell, A. (1992) The anthropology of time. Cultural constructions of temporal maps and
images. Oxford: Berg.
Haspelmath, M. (1997) From Space to time: Temporal adverbials in the world’s
languages. München: LINCOM.
Hill, C. A. (1978) Linguistic representation of spatial and temporal orientation. Paper
presented at the Proceedings of the Fourth Annual Meeting of the Berkeley
Linguistic Society.
Jackendoff, R. (1983) Semantics and cognition. Cambridge, MA: MIT Press.
Keesing, R. M. (1991) Time, cosmology, and experience. Unpublished manuscript.
Klein, H. E. M. (1987) The future precedes the past: Time in Toba. Word 38: 173–185.
Klein, W. (1994) Time in language. London: Routledge.
Lakoff, G. (1993) The contemporary theory of metaphor. In A. Ortony (ed.)
Metaphor and thought 202–251. (Second edition.) Cambridge: Cambridge
University Press.
Lakoff, G. and Johnson, M. (1999) Philosophy in the flesh: The embodied mind and its
challenge to Western thought. New York: Basic Books.
Levinson, S. C. (1996a) Frames of reference and Molyneux’s question: Cross-
linguistic evidence. In P. Bloom, M. Peterson, L. Nadel and M. Garrett (eds)
Language and space 109–169. Cambridge, MA: MIT Press.
Levinson, S. C. (1996b) Language and space. Annual Review of Anthropology 25:
353–382.
Levinson, S. C. (2003) Space in language and cognition. Explorations in cognitive
diversity. Cambridge: Cambridge University Press.
Levinson, S. C. (2004) Time for a linguistic anthropology of time. Current
Anthropology 43(Supplement): 122–123.
Lucy, J. A. (1997) The linguistics of ‘color’. In C. L. Hardin and L. Maffi (eds) Color
categories in thought and language 320–346. Cambridge: Cambridge University
Press.
Malotki, E. (1983) Hopi time: A linguistic analysis of the temporal concepts in the Hopi
language. Berlin: Mouton.
McGlone, M. S. and Harding, J. L. (1998) Back (or forward?) to the future: The role
of perspective in temporal language comprehension. Journal of Experimental
Psychology: Learning, memory, and cognition 24: 1211–1223.
Miracle, A. M., Jr. and Yapita Moya, J. d. D. (1981) Time and space in Aymara. In M.
J. Hardman (ed.) The Aymara language in its social and cultural context 33–56.
Gainesville, FL: University Presses of Florida.
Moore, K. E. (2000) Spatial experience and temporal metaphors in Wolof: Point
of view, conceptual mapping, and linguistic practice. University of California,
Berkeley.
Munn, N. D. (1992) The cultural anthropology of time: A critical essay. Annual
Review of Anthropology 21: 93–123.
Núñez, R. E. and Sweetser, E. (2006) With the future behind them: Convergent
evidence from Aymara language and gesture in the crosslinguistic comparison of
spatial construals of time. Cognitive Science 30: 1–49.
498 LANGUAGE, COGNITION AND SPACE

Radden, G. (2003) The metaphor TIME AS SPACE across languages. In C. B.

N. Baumgarten, M. Motz and J. Probst (eds) Uebersetzen, Interkulturelle
Kommunikation, Spracherwerb und Sprachvermittlung – das Leben mit mehreren
Sprachen. Festschrift fuer Juliane House zum 60. Geburtstag. (Vol. Zeitschrift fuer
Interkulturellen Fremdsprachenunterricht 8(2/3): 226–239.
Santiago, J. (2005) Time also flies from left to right. Paper presented at the First UK
Cognitive Linguistics Conference, October 2005, University of Sussex.
Saunders, B. (1995) Disinterring basic color terms: A study in the mystique of cogni-
tivism. History of the Human Sciences 8(4): 19–38.
Senft, G. (1996) Past is present – present is past: Time and the harvest rituals on the
Trobriand Islands. Anthropos 91: 381–389.
Shinohara, K. (1999) Typology of space-time mappings. Manuscript.
Stutterheim, C. von, Carroll, M. and Klein, W. (2003) Two ways of construing
complex temporal structures. In F. Lenz (ed.) Deictic conceptualisation of space,
time and person 97–133. Amsterdam, PA: John Benjamins.
Svorou, S. (1994) The Grammar of space. Amsterdam: John Benjamins.
Talmy, L. (2000) Toward a cognitive semantics. Cambridge, MA: MIT Press.
Traugott, E. C. (1978) On the expression of spatio-temporal relations in language.
In J. H. Greenberg (ed.) Universals of human language 369–400. Stanford, CA:
Stanford University Press.
Tversky, B., Kugelmass, S. and Winter, A. (1991) Cross-cultural and developmental
trends in graphic productions. Cognitive Psychology 23: 515–557.
Wierzbicka, A. (1996) The meaning of colour terms and the universals of seeing. In
A. Wierzbicka (ed.) Semantics: Primes and universals 287–334. Oxford: Oxford
University Press.
Yu, N. (1998) The contemporary theory of metaphor: A perspective from Chinese.
Amsterdam, PA: John Benjamins.
19 From mind to grammar: coordinate systems,
prepositions, constructions
Paul Chilton

1 Introduction

Suppose I want to pick up the pencil on my desk. How do I do it? Here are some basic
elements of what is a complex process. Imagine there is a set of lines relating the object
to its image on my retina, and a set of lines centred on my hand. My eyes and my hand
are bits of me. Somehow my brain has to relate these two ‘perspectives’ on the pen, so
that I can execute the reaching and grasping movement.
Let us consider prepositions. My student has lost her pen. I say to my student: ‘The
pen is in front of the computer’. I can say this no matter where I am standing and prob-
ably she will fix on the same location for the pen, one that arises because we give a front
and a back to objects like computers and because pens are relatively small. Suppose I say:
‘The pen is in front of the waste basket’. Because my waste basket is roughly cylindrical
and has no orienting features, this sentence will probably convey that the pen is located
in a spatial region between me and the waste basket (or between the addressee and the
waste basket). Some uses of some prepositions fix locations of objects relative to other
objects; others fix them relative to the speaker’s position. It’s a question of viewpoint.
Take a more abstract linguistic structure, the counterfactual conditional construc-
tion. One may say: ‘If John had gone to the party he would have seen Sarah’. Here we
have, within the two parts of the sentence, affirmative clauses. But the sense is negative:
John did not go to the party and he did not see Sarah. Conversely, one may say: if John
had not gone to the party, he would not have seen Sarah. Here the clauses are lexically
negative, but John did go to the party and he did see Sarah.
What I aim to do in this paper is to demonstrate the plausibility of making connec-
tions between all three of the above scenarios. That is to say, I want to explore the way
in which attested neural operations may motivate linguistic structures, at two different
levels of abstraction.

2 Egocentric and allocentric operations in visual perception and

spatial orientation

The separation of dorsal and ventral pathways in schematic form has been the object
of research for some years (for summary see Hartley and Burgess 2003). It is now well
established that input form the eyes is fed to the visual cortex (V1), but subsequently
is routed to diverse areas of the cortex – various regions in the parietal lobe (broadly

499
500 LANGUAGE, COGNITION AND SPACE

called the dorsal stream) and various regions of the temporal cortex including the hip-
pocampal and parahippocampal structures (the ventral stream) (Milner and Goodale
1995). This kind of separation is not of course arbitrary but has a functional basis. The
different functions have been characterised in terms of the brain and body’s relationship
to physical space. Geometrically, the dorsal and ventral streams correspond to egocentric
coordinate systems and allocentric systems, respectively (cf. Goodale and Milner 2005,
Klatzky 1998, Rolls 1999, among others).
The broad function of the dorsal stream is related to action, specifically actions
of reaching and grasping. This implies that the representations in the parietal areas
utilise egocentric coordinate systems in order to locate objects the organism is about
to act upon. This is why the dorsal stream has been called the ‘where’ stream. These
egocentric coordinates are actually centred (have their origin at) various bodily parts
including the retina, the hands, the mouth, the feet. It follows also that there must be
geometric transformation between these egocentric coordinate systems. The nature and
localisation of these transformation systems is at present not fully understood (though
Burgess 2002 indicates posterior parietal cortex, Brodman’s area 7a).
The broad function of the ventral stream is object recognition. It is the ‘what’
stream. It is possible for individuals to have lesions in the temporal area, and thus be
unable to recognise or name objects, while still being able to perform spontaneous
motor activities upon them (reaching, navigating, etc.). It is the ventral stream and its
complex connections that enable the organism to understand a scene and the objects
in it: that is, to know the categories and properties of objects as well as to know how
they relate to one another in scene-based (i.e. allocentric) coordinates (cf. Goodale
and Milner 2005: 101). In addition, the ventral system is connected to systems in
the hippocampus that enable navigation through physical space by using allocentric
landmarks within a reference frame. It has been shown that areas CA1 and CA3 of
the hippocampus (anatomically similar to that of primates including humans) contain
‘place cells’, which fire differentially in response to real world locations (O’Keefe and
Nadel 1978). Structures linked to the hippocampus (mamillary bodies, presubiculum,
anterior thalamus) contain head direction cells, which represent the individual’s head-
ing relative to landmarks in the environment. The hippocampus also makes it possible
to lay down long-term (episodic, i.e. individual) memories of locations and of locations
in relation to events Now for this to be possible, there must be a transformational
geometric to-and-fro between the egocentric parietal processing and the allocentric
temporal-hippocampal processing. Again the exact nature and localisation of such
process is still being researched.
The existence of these two interrelated brain systems has already excited the interest
of some linguists. Givón sees a correspondence between the ventral stream and lexical
semantics on the one hand and between the dorsal stream and propositional information
about states or events (Givón 1995: 408–410). Similarly, but in more detail, Hurford
(2003) argues for roughly the following: that the ventral and dorsal streams are the
evolutionary basis of predicate-argument structure, the ventral stream being the basis of
argument concepts, the dorsal the basis of predicate concepts. In the present paper I am
arguing something different. I am arguing that it is the distinction between egocentric
 FROM MIND TO GRAMMAR: COORDINATE SYSTEMS, PREPOSITIONS, CONSTRUCTIONS 501

frames and allocentric frames that is of interest. I am further arguing that it is the
geometric transformations between the two frames that is important. They are important
because the neuroscience and experimental psychology research clearly indicates that
geometric coordinate systems are instantiated neurologically and behaviourally (cf.
Gallistel 1999). And it is also indicates that geometric transformations from ego- to
allocentric coordinate systems is neurologically and behavioural instantiated. It surely
behoves cognitive linguists to take account of this evidence. The claim I want to pursue
here is that the transformations I have alluded to should be encountered in lexical mean-
ing. This is perhaps not controversial, since the study of spatial propositions has long
since been couched in terms of coordinate systems. It should however be noted that I
am not here concerned with the cross-linguistic and Whorfian issues raised by the work
of Levinson and others (Levinson 2003). Whatever the cross-linguistic differences in
coding, egocentric and allocentric spatial cognition, alongside transformation between
the two types of system, is a property of human brains and thus universal.
A more speculative claim – and the one that I shall outline in more detail below – is
that the egocentric and allocentric spatial frames, together with their transformation
operations, are found in grammatical constructions. Indeed, I want to suggest that many
syntactic, semantic and pragmatic phenomena that have hitherto received unconvincing,
superficial or controvertible expositions can be naturally explained in a motivated way,
in a framework that uses an abstract discourse space representation that is derived
from the physical spatial representations, the evidence for the existence of which is
well attested, as noted above.

3 Spatial prepositions

It is widely recognised that spatial adpositions across languages exploit three-dimen-

sional coordinate systems whose axes correspond to the sagittal, vertical and lateral
axes of the human body.1 That such coordinate systems are also transformed in various
ways is widely acknowledged: the most explicit account in the linguistics domain is
Levinson (2003). The present account draws attention to the neurobiological evidence
that the human brain alternates between, and integrates, egocentric and allocentric
coordinate systems (see above). Egocentric systems locate one or more objects by
a position vector in a coordinate system with origin at the speaker/self, oriented
by the heading of the speaker/self. This happens when the speaker is the reference
location or landmark with any proposition (e.g. ‘X is over me’), but also in ‘relative
frames’ where the speaker’s orientation is projected onto an object without intrinsic
orientation (e.g. ‘X is to the left of the tree’, cf. Levinson 2003: 43–47). Allocentric
systems relate one object to another object, either in coordinates centred on the
landmark (‘X is over Y’, etc.) or by reference to an environmental feature (as in the
Tzeltal equivalent of ‘X is uphill of Y’ for horizontal plane location: Levinson 2003).
Levinson’s tripartite approach downplays the underlying cognitive operations of
egocentric and allocentric operations, while the present approach highlights them,
as suggested by the table below.
502 LANGUAGE, COGNITION AND SPACE

Table 1

Type of reference frame Egocentric coordinate system Allocentric coordinate

system
Verticality, determined by canonical ‘X is over me’ ‘X is over Y’
position of objects but predominated (3-dimensional coordinates, with (3-dimensional coordinates, with
by gravitational field origin at me) origin at Y)

Intrinsically orientated coordinate ‘X is in front of me’ ‘X is in front of the horse/chair, etc’

system (I have intrinsic front-back (horse and chair, etc. have ‘front’
orientation) and ‘back’)

Relative orientation of coordinate ‘X is in front of the horse/chair, etc’

system (between me and the horse)

‘I am in front of the tree’

(i.e. my oriented axes are reflected ‘John is in front of the tree’
in the non-intrinsically oriented (i.e. John’s oriented axes are
object, so that this object now has reflected in the non-intrinsically
a ‘front’) oriented object, so that this object
now has a ‘front’)

Absolute orientation of coordinate ‘X is north of me’ ‘X is north of Y’

system (coordinate system is fixed
geophysically)

This table makes clear the obvious fact that egocentric (or deictic) and allocentric
conceptualisations can be made within oriented and non-oriented coordinate systems
with origins at various locations. The orientational prepositions in front of/behind are of
interest because they sometimes give rise to egocentric, sometimes to allocentric rep-
resentations. Indeed, in a non-contextualised sentence they can give rise to conceptual
oscillation, a little like an optical illusion. Thus

(1) John is in front of the tree

can make us think either Figure 1 or Figure 2.

J J
S S

Figure 1. Speaker-centred axis system Figure 2. Allocentric axis system

 FROM MIND TO GRAMMAR: COORDINATE SYSTEMS, PREPOSITIONS, CONSTRUCTIONS 503

Figure 1 shows the conceptualisation in which the coordinates are centred on the speaker,
S, i.e. John is between S and the tree. This conceptualisation also seems to involve the
‘occlusion’ understanding discussed by Evans (Chapter 9 of the present volume). These
are egocentric coordinates. Alternatively, sentence (1) can evoke something like Figure
2: the coordinates are now translated away from the speaker onto John. These are
allocentric coordinates. However, the analysis needs to be slightly more complicated,
since ‘is X in front of Y’ should mean that it is Y’s front that is the landmark – here,
that it is, so to speak, the tree’s front that is the landmark with respect to which John
is being located. We can describe this geometrically as a reflection transformation, as
shown in Figure 3.2

J
S

Figure 3. Reflection of axis system

The reflection transformation maintains John’s left-right directedness. To say a locandum

is to the left of the tree then means that it has the same coordinate on the tree’s lateral
axis as it does on John’s. If the landmark is human or humanoid, then it is possible to
conceptualise, and to linguistically encode, a 180 degree rotation that transfers the
speaker’s right to the landmark’s left. For example, if John is facing S, then S may say,
‘the tree is to John’s left’. If John is not facing S, then there will be a translation of S’s
coordinates.
The meaning of (1) will flip between the egocentric and the allocentric conceptu-
alisation. In real utterances the denoted location of John will vary: in the egocentric
conceptualisation John will always be between S and the landmark, while in the allocen-
tric one John may be anywhere on the circumference of a circle with centre at the tree
and the frontal axis its radius. If the landmark has its own ‘intrinsic’ frontal orientation
(e.g. it’s a cat, bus, the town hall …), then there is an additional possible allocentric
conceptualisation, but this does not affect the main point being made – that egocentric
and allocentric representations are recruited by spatial prepositions.
Once we introduce coordinate systems it is a natural step to introduce vectors –
mathematical objects drawn as arrows that have direction and magnitude. In a coordinate
system the position of a point can be given by the length and direction of a vector from
the origin to the point. Giving the coordinates on the axes of the system is equivalent
504 LANGUAGE, COGNITION AND SPACE

to specifying the vector. This approach can be used for explicating the denotation of
certain prepositions, including ‘in front of ’, by specifying a vector space in which all
vectors have the same origin in some coordinate system (Zwarts 1997, O’Keefe 1996).
This space will be the ‘search domain’ within which an object can be said to be, for
example, ‘in front of John’.

4 Discourse space

The claim that I want to make is that the coordinate transformations described above
for spatial prepositions are also found in grammatical constructions. To see this we have
to move to an abstract (or metaphorical) space. In principle, this is not a controversial
move within Cognitive Linguistics. But the notion of an abstract discourse space is new
and needs a cautious introduction (see also Chilton 2005).
We call this new abstract space, the Discourse Space (DS), and diagrams of this space
are called discourse space models (DSMs). The DS has three scalar axes: d (discourse
distance), t (time) and m (modality), as in Figure 4. This is the base space of the speaker
(self, subject), S. Other coordinate systems of the same type can be set up at points
other than S. The t-axis points in two directions from time 0, the time of utterance. The
d-axis points in one direction and allows us to represent geometrically the foreground
and background distinction (figure-ground separation) in discourse, i.e. the difference
between what is made grammatically salient and what is not. The m-axis also points in
one direction and represents epistemic modality. The point maximally distant from S
on the m-axis is irrealis or counterfactual. The m-axis points in one direction only (i.e.
has no negative half-line) because it models modality in terms of distance from S, who
is at the point of maximal certainty, coinciding with present time on t and maximum
salience on d. The m-axis has an obvious mid-point corresponding to conceptualisations
of ‘possible’ and ‘if ’.
The DS is thus not a direct analogue of the three-dimensional systems for physi-
cal space discussed in section 2. The way the DSM is drawn should not be taken to
correspond to up/down, left/right, front/back axes. The claim is that it is the mini-
mum space need to account for a significant number of grammatical and discourse
phenomena in an insightful way that is linked to the cognitive motivation outlined
in sections 1 and 2.
This does not mean that three dimensions are going to be sufficient to model all
phenomena of discourse meaning. On the contrary, discourse processing certainly
includes many dimensions.3 It is possible, however, that three-dimensionality has a
special part in human cognition. Be that as it may, the aim here is show how even modest
dimensionality can yield insightful modelling of lexical and grammatical phenomena,
precisely because the geometry enables us to model transformations of coordinate
systems relative to a speaker.
 FROM MIND TO GRAMMAR: COORDINATE SYSTEMS, PREPOSITIONS, CONSTRUCTIONS 505

-t
S
+t

Figure 4. Basic discourse space axis system

In the discourse space referents are ‘located’ as points; they may be real to S (m=0),
hypothetical or counterfactual, or, as we shall see, in some embedded axis system with
origin at some location in the base system. Coordinates keep track of anaphoric relations.
‘Locations’ are of course abstract locations that express concepts in grammar, broadly as
outlined in ‘localist’ theory (Anderson 1972). Relationships between discourse referents
are vectors, postulated as unit vectors unless specified otherwise in the context. Vectors
are interpreted in various ways standard physical applications, and these applications
are followed here. Thus force vectors enable us to model causal relations as directed
force, translation vectors allow us to model movement (physical and abstract), position
vectors locate referents with respect to other referents.

5 Grammatical constructions and axis transformations

In the rest of this paper I shall treat discourse referents as points with coordinates in
the DS, and relations between entities as unit vectors, whether spatial relations or not.
But the focus will be on the coordinate systems in which vectors are located, and the
transformational relationships between coordinate systems. Some of the constructions
that were once predominantly described in generative syntax as ‘transformations’ can
also be so described in the theoretical framework I am outlining here. But there is a big
difference. The present framework has simultaneously a cognitive and mathematical
motivation, and possibly, as suggested in section 1, a neurological one too. The next
three subsections demonstrate three different constructions that can be elucidated by
working with three-dimensional DSMs with axis transformations.

5.1 Reflection: active and passive

Passive and active constructions have the function in discourse of foregrounding an

entity that has undergone some action and the resultant state. The resultant state can
be a physical location, but states, like properties, are also treated in DST as conceptual
locations (cf. Anderson 1971, and examples like ‘in a broken condition’). The two
506 LANGUAGE, COGNITION AND SPACE

constructions were treated in early generative grammar as transformations in an idi-

osyncratic sense of the term. Here we treat them as related by a transformation that has
its standard sense in coordinate geometry. Within DST the effect of the transformation
is, as required, to bring one discourse entity conceptually ‘closer’ in the discourse to S.
The example in Figure 5 also illustrates the way the discourse space described in
section 3 is used to represent verb semantics. In the sentences
(2) John broke the vase
(3) John moved the vase

both verbs are analysed as having two component vectors, one a force vector, the other a
translation vector. The force vector causes the translation. The verb ‘move’ is represents
a physical change of place. The verb ‘break’ represents a change of physical state, with
states analysed as (abstract) locations. Sentences (2) and (3) thus have parallel structure.
This shows that, under localist assumptions, we can understand at least two types of
transitive verb in the DST framework.4 With these preliminaries, we can now show
how an account of the active-passive relation can fall out naturally from the theory.
First, consider the active construction depicted in Figure 5. The vector v1i s force
vector with tail at the discourse referent John (realis for S) and whose impact on the
referent ‘the vase’ causes a ‘translation’, v2, to a new state, ‘broken’. And let us assume a
context that has this resultant state (i.e. the vector v3 mapping onto itself) continuing
to t=0. The geometric formalism allows us to show John’s causing act as foregrounded,
i.e. ‘closer;’ on the d-axis than the other parts of the event structure.

broken, v3

translation v2 broken
change of state

vase
force v1
causation

John

ti
v1+ v2= u1 S

Figure 5. Event structure as vectors: John broke the vase

It is now a simple matter to use a transformation of axes – specifically, a reflection

transformation – to represent the passive construction, as shown in Figures 6 and 7.
The effect is now to background John and his causing action. In ‘the vase was broken by
 FROM MIND TO GRAMMAR: COORDINATE SYSTEMS, PREPOSITIONS, CONSTRUCTIONS 507

John’ it is the referent, together with its change of state and ongoing state that are fore-
grounded. This is precisely what the reflection transformation produces. Furthermore,
an explanation of the use of the prepositional phrase also falls into place.
In the English passive, the agent, if it is expressed, is expressed in a prepositional
phrase with ‘by’. Why is this spatial preposition used? All we need to do to see the
answer is to look at Figure 7 and interpret v1as a location vector. The event v2 + v3
(the breaking of the vase) is ‘located at’ the backgrounded referent ‘John’.5 What this
geometric analysis shows clearly is change of perspective. The speaker can bring into
focus either the agent or the undergoer of in an event, just as spatial prepositions can
alternate between viewpoints, e.g. John is in front of the tree, the tree is in front f John.

v3
v2 broken/

vase
v1

John

ti

Figure 6. Passive as reflection of axes: The vase was broken by John

by John
location vector
v1

v2 vase

v3 broken

ti

Figure 7. Reflected axes of Figure 6 in normal view

508 LANGUAGE, COGNITION AND SPACE

5.2 Translation: factive verbs and epistemic verbs

Just as language reflects the cognitive ability to adopt alternate viewpoints on a single
physical-spatial relation (cf. section 2), and just as alternate viewpoints can be taken
on events (cf. section 5.1), so language structure corresponds to a cognitive ability to
adopt the ‘point of view’ of another human agent’s mental state. 6 Certain lexical items
produce semantic constructions that identify S’s epistemic state with that of another
human (or humanised) agent; others attribute mental states to human agents different
from that of S. So-called factive verbs typify the former, as in (4), while epistemic verbs
such as believe typify the latter, as in (5):

(4) John knows that Mary wrote the report

(5) John believes that Mary wrote the report

Representing the core proposition ‘Mary wrote the report’ as points and vectors,
the difference between (4) and (5) can be represented in terms of transformation
of axes. The DST framework naturally incorporates the presupposition triggered by
the factive verb in (4), viz. that Mary did indeed write the report. This presupposi-
tion I interpret as S’s belief, i.e. mental representation, for which the DSM gives the
fundamental scaffolding. In Figure 8, which models (4), ‘know’ can be understood as
a function (transformation) that creates a secondary coordinate system with origin
located at the point for the discourse referent ‘John’ (propositions are ‘located’ in a
mind: cf. Anderson 1972, Lyons 1977). The coordinate for the new origin is m=0 in
S’s base system. This means that what John, S´, knows is also epistemically identical
with what S knows, which is the same as saying that (4) presupposes the truth of the
complement clause for S.
In (5) there is no presupposition, which in effect simply means that S does not
accept, and does not communicate that she or he accepts, that p, i.e. that Mary did
indeed write the report. However, S does hold it to be true that John exists and that
John believes that p. As in the Figure 8, the main verb produces secondary axes by
translation, and its origin has a coordinate on d at ‘John’. However, verbs like believe
(as distinct from know) create a secondary axis system with origin at m>0. The value
of m depends on contextual factors. For example, S might think that p may or may not
be true, in which case the origin of the new set of axes will be located at the mid-point,
as in Figure 9; alternatively, S might think that p is in fact counterfactual, in which case
the origin of the new system will be at the extreme of m. Thus in this DSM, proposition
p is simultaneously true for John, which is what S asserts, but not necessarily true for
S, as the interpretation of (5) requires.
 FROM MIND TO GRAMMAR: COORDINATE SYSTEMS, PREPOSITIONS, CONSTRUCTIONS 509

report
write

Mary

John
know Sƍ
ti

Figure 8. Translation of axes: John knows Mary wrote the report

write

report

Mary
Sƍ
John believe

ti

Figure 9. Translation of axes: John believes Mary wrote the report

5.3 Conditional sentences

A number of puzzles relating to counterfactual conditionals have been discussed

(Fauconnier 1994, and Dancygier and Sweetser 2005). One puzzle that has not been
much explored is the following. In a lexically ‘positive’ counterfactual conditional sen-
tence, the meaning is ‘negative’: in (6) S is communicating that John did not go to the
party. On the other hand, in a lexically ‘negative’ sentence, S is communicating that
John did go to the party.

(6) If John had gone to the party, he would have seen Sarah
(7) If John had not gone to the party, he would not have seen Sarah
510 LANGUAGE, COGNITION AND SPACE

It is possible to give an account of this form-meaning relationship within the framework

of DST. In both Figure 10 and Figure 11 John, Sarah and the party are real for the S and
for S´ coordinate systems. Notice that the party and Sarah have the same coordinates
(in both systems). This reflects the cognitive processing of the sentences: there is a basic
inference that Sarah was at the party.7 The predicates go and see are modelled as vectors
relating John and the party in the protasis, likewise John and Sarah in the apodosis. The
precise characterisation of the relationship between the two clauses is not our concern
here, but this relationship has a temporal component, as indicated.
The word if is a transformation projecting a reflection of the base system axes,
around the mid point on m, giving a new origin 0´ at the distal end of m. Note that this
holds constant the time relationships in both the ‘real’ world of S and the counterfactual
world of S´; the same goes for the discourse perspective given by the d coordinates.
Now, in the positive case (6), the positive verb ‘had’ indicates that the vector relating
John and the party and the vector relating John and Sarah are realis in the reflected
counterfactual coordinates, i.e. located at m´ = 0´. Simultaneously, these vectors are at
the negative end of m in the base system of S. As shown in Figure 10, the reflection
transformation economically represents the linguistic-cognitive facts: we have a ‘positive’
sentence with a ‘negative’ meaning.

go
see

(i) party
(ii) Sarah

John

ti

Figure 10. Counterfactual (positive): If John had gone to the party, he would have seen Sarah

The converse is true in the case of (7), modelled in Figure 11. The negative verb places
the predicates go and see at the negative end of the reflected axis system, which means
that it is simultaneously realis for the base system, as required. It is a simple matter to
demonstrate that the model provides the right analysis of sentences with combinations
of negated and non-negated protasis and apodosis.
This result emerges naturally from the framework we have adopted. The simple geo-
metrical properties we have used are inherent in the standard geometrical framework.
 FROM MIND TO GRAMMAR: COORDINATE SYSTEMS, PREPOSITIONS, CONSTRUCTIONS 511

(i) party
go (ii) Sarah
see
John

ti

Figure 11. Counterfactual (negative): If John had not gone to the party, he would not have seen Sarah

6 Conclusion

Only three grammatical constructions have been demonstrated, but they are all important.
What has been discovered in the DST approach is that the concept of transformation,
as standardly defined in geometry, appears to be relevant to explaining grammatical
constructions, once we have hit on the notion of an abstract discourse space with a ‘modal’
axis. Transformations involving the other two axes can be used to model constructions not
considered here. What we have seen is that coordinate systems, vectors and transformation
of axes appear to be descriptively powerful at the level of mind, spatial semantics and
grammatical constructions. It is remarkable that the rather simple geometric concepts
we have used yield models that integrate linguistic form and conceptualisation. How far
this claim can be extended is, however, a matter for further investigation.
It is important to note that the geometrical axis system (frames of reference) used
in DST is not the same as that used for describing the workings of spatial prepositions.
Certain kinds of frame of reference serve for relations in physical space. The discourse
space of DST is an abstract space based on linguistically relevant concepts. It is an
abstract space but one that is deictically anchored on the speaker S. It is not meant
to be a description of all aspects of discourse processing but only of the fundamental
‘scaffolding’ on which discourse rests. I do want, however, to suggest that geometrical
principles are appropriate to describe it precisely because geometrical principles are
appropriate to describe the physical space that prepositions refer to. I am also sug-
gesting that the human mind is using spatial principles, which we can describe using
three-dimensional coordinate geometry and vectors, for apprehending both physical
space and the more abstract relationships such as figure-ground ‘distance’, temporal
‘distance’ and epistemic ‘distance’.
512 LANGUAGE, COGNITION AND SPACE

This chapter began by describing some aspects of the way the brain processes the
spatial environment, proceeded to describe the linguistically mediated conceptualisa-
tion of spatial relations, and then applied geometrical principles to the description of
grammatical constructions. Whether the sequence in which I have presented these three
domains should be taken to suggest a causal or an evolutionary sequence is another
matter. But there are at least grounds for speculating that the egocentric-allocentric
alternation, which appears to be neurally attested, may be of significance beyond
spatial cognition. What I have suggested in this paper is the following. First, neurally
embodied spatial cognition can be described geometrically and in terms of egocentric
and allocentric relativisation. Second, spatial prepositions can at least in large part be
described in the same terms. Thirdly, at least some key grammatical constructions can
also be insightfully described in the same terms, given the abstract discourse space
model. Finally, these parallelisms, one may speculate, indicate an underlying motivation
for grammar in the structure of spatial cognition.

Notes
1 This is not to say that ‘functional’ concepts are not also involved, as argued by several
contributors to the present volume.
2 Levinson (2003: 44–45) describes something similar. It is important to note that the
axis system located on John is allocentric and also embedded in that of a speaker S. I
have not attempted to represent S’s axes explicitly in the diagrams, since S can of course
occupy an infinite set of spatial positions relative to John.
3 This is no problem in algebraic format; high dimensional vector spaces are the basis of
connectionist modelling and the design of textual search engines (cf. Widdows 2004).
4 Sentences such as The message was seen by John need a slightly different analysis.
5 How far this extends cross-linguistically requires investigation, but note that some
languages need to use a translation vector: the event ‘proceeds from’ the agent (German
von, Latin ab). In French, later Latin and other Romance languages (per, par), conceptu-
alisation also involves spatial movement albeit motion through a medium (= the agent).
It should be noted that the analysis seems to apply also to the antipassive construction
in such languages as Dyirbal.
6 This linguistically attested ability corresponds to what psychologists call ‘theory of
mind’ (Baron-Cohen 2001).
7 It is inferred that Sarah is in the same location as the party in this particular example.
The DSM gives the same coordinate point for both party and Sarah and it is important
to note that the d-axis does not represent physical space. Obviously, this is not the
case in all conditional sentences: e.g. If John had gone to the party, Sarah would have
gone to Manchester. Spatial relations, including those represented by prepositions and
discussed in section 2 above are not part of DST, though there is an important analogy
between them. DST does not model physical spatial relationships. DST describes certain
fundamental aspects of discourse processing. I assume other systems handle representa-
tions arising from the many other aspects of discourse processing, including relations
 FROM MIND TO GRAMMAR: COORDINATE SYSTEMS, PREPOSITIONS, CONSTRUCTIONS 513

in physical space. So far as DST is concerned, Sarah and party in (6) and (7) occupy the
same ‘place’ in terms of discourse distance (or figure-ground separation), as explained
in section 3 above. It is also important to note also that the DSM can be regarded as
a composite of two successive DSMs, one for each clause, in which coordinates are
assigned to referents as the sentence is sequentially processed – whence the marking of
the two referents as (i) and (ii) in figures 10 and 11.

References
Anderson, J. (1971) The Grammar of Case. Towards a Localist Theory. Cambridge:
Cambridge University Press.
Baron-Cohen, S. (2001) Mindblindness. An Essay on Autism and Theory of Mind.
Cambridge, MA: MIT Press.
Burgess, N. (2002) The hippocampus, space, and viewpoints in episodic memory.
Journal of Experimental Psychology 55A(4): 1057–1080.
Chilton, P. A. (2005) Vectors, viewpoint and viewpoint shift: Toward a discourse
space theory. Annual Review of Cognitive Linguistics 3: 78–116.
Dancygier, B. and Sweetser, E. (2005) Mental Spaces in Grammar. Cambridge:
Cambridge University Press.
Fauconnier, G. (1994) Mental Spaces. Cambridge: Cambridge University Press.
Gallistel, C.R. (1999) Coordinate transformations in the genesis of directed action.
In B. O. M. Bly and D. E. Rummelhart (eds) Cognitive Science 1–42. New York:
Academic Press.
Givón, T. (1995) Functionalism and Grammar. Amsterdam: John Benjamins.
Goodale, M. A. and Milner, A. D. (2005) Sight Unseen. Oxford: Oxford University
Press.
Hartley, T. and Burgess, N. (2003) Spatial cognition, models of. In L. Nadel (ed.)
Encyclopedia of Cognitive Science Vol. 4 111–119. London: Nature Publishing
Group.
Hurford, J. R. (2003) The neural basis of predicate-argument structure. Behavioural
and Brain Sciences 23(6): 261–283.
Klatzky, R. L. (1998) Allocentric and egocentric spatial representations: Definitions,
distinctions, and interconnections. In C. Freksa, C. Habel and K. F. Wender (eds)
Spatial Cognition: An Interdisciplinary Approach to Representation and Processing
of Spatial Knowledge 1–17. Berlin: Springer-Verlag.
Levinson, S. (2003) Space in Language and Cognition. Cambridge: Cambridge
University Press.
Lyons, J. (1977) Semantics. (2 vols.) Cambridge: Cambridge University Press.
Milner, A. D. and Goodale, M. A. (1995) The Visual Brain in Action. Oxford: Oxford
University Press.
O’Keefe, J. (1996) The spatial prepositions in English, vector grammar, and the cogni-
tive map theory. In P. Bloom et al. (eds) Language and Space. Cambridge, MA:
MIT Press.
O’Keefe, J. and Nadel, L. (1978) The Hippocampus as a Cognitive Map. Oxford:
Oxford University Press.
514 LANGUAGE, COGNITION AND SPACE

Rolls, E. T. (1999) Spatial view cells and the representation of place in the primate
hippocampus. Hippocampus 9: 467–480.
Widdows, D. (2004) Geometry and Meaning. Stanford, CA: CSLI Publications.
Zwarts, J. (1997) Vectors as relative positions: A compositional semantics of modified
PPs. Journal of Semantics 14: 57–86.
Index
A Broca’s area 86
A-series time 481–482, 484–485, 490,
C
492–493, 496
categorisation 419–421, 437–440, 449
absolute frame of reference 67, 79, 147,
causality 14, 396, 421, 425, 429, 434–436,
154, 158–160, 162, 295, 297, 483, 487,
440, 442–444
489, 491–494
Chinese 84, 86, 92, 101, 459, 480, 498
accusative case 269–270, 274, 277, 282,
classifier subsystem 335, 337–340, 345–346
284–290
closed-class 9–10, 12, 79, 163, 228–229,
adpositions 8, 79, 190, 247, 251, 262, 293,
238, 319–324, 328–333, 335–337, 340,
501
342–343, 345–346
agency 421, 427, 429, 433–434, 436, 438,
cognitive linguistics ii, iv, 4, 11, 17, 47–48,
444, 446, 449
90, 112–114, 163–164, 191–192, 247–
agentivity 11, 14, 205, 259, 261–262
248, 290–291, 312–314, 349, 351–353,
allocentric 2, 39–41, 56–57, 139, 142, 153,
380–385, 417–418, 419–420, 448–449,
294, 499–503, 512–513
476–477, 498, 504, 513
American Sign Language (ASL) 10, 12, 87,
cognitive model 215, 228, 290
91, 150, 164, 334, 336–337, 340–341,
colour terms 8, 192, 480, 498
343, 348–349, 357, 377, 380, 382,
computational models of language 87
384–385
conception 2, 16, 21, 27, 114, 171, 176, 191,
analogy 40–41, 420, 482, 484, 487–488,
225, 231, 325, 475
490–492, 494, 512
conceptual metaphor 10, 80, 216–217, 351,
angular gyrus 8, 149–150
379, 383, 457–458, 476, 479
animacy 14, 102, 105, 107–109, 111, 284,
conceptual primitives 48, 76, 79, 82
422, 427, 429, 432–434, 436, 440, 444,
conceptual relativism 420
446, 449
conceptual representation 43
articulator(s) 355–356, 362, 364, 373, 377
conceptual spaces 7, 153, 210, 212, 272
asymmetry 65, 130, 295, 377–378,
conceptual system 22, 43, 45, 79, 186–187,
408–409, 413–414, 458, 465, 483
376
atelic motion 422, 424–425, 444
conceptual typology vi, 419–421, 430, 436,
Aymara 85–86, 92, 384, 480, 492–493, 497
438, 445, 447
axis of motion 81
conditional sentences 509, 512
B containment 7, 9, 44–45, 47, 79, 90,
B-series time 481–482, 485–487, 491–493, 97–98, 108, 114, 142–144, 153, 157–158,
495 175–176, 179, 181–182, 185–189, 192,
Basque vi, 10–11, 162, 251–259, 262–265 194–195, 197, 208–211, 221, 229–232,
body parts 29, 42, 145, 151, 154, 165, 241, 258, 262, 361–362, 364, 374
297, 339, 357, 391, 430 contiguity 142, 153, 176, 195, 197, 206,
bottom-up processing 23 211, 374
bridging context 218

515
516 LANGUAGE, COGNITION AND SPACE

coordinates 6, 56, 66, 154, 258–259, extended network 273

356–357, 394–395, 483–484, 490, 493,
F
495, 500–503, 505, 510, 513
fictive motion 4, 18, 83, 91, 447
coordinate system 258–259, 295–297, 394,
figure-ground 23, 31–32, 36, 103, 486–487,
483–487, 489, 491, 493, 501–504, 508
504, 511, 513
D figurative expressions 83
deictic orientation 280 force dynamics 9, 14, 114, 212–213,
deictic relations 7, 140–141, 148, 152 267–268, 274–275, 327, 384, 419, 422,
determinism 186–187, 384, 448 427–428, 431, 436, 446, 478
directionality 14, 252, 422–424, 434–436, force vectors 193, 198–200, 208–209, 505
446, 474 frames of reference vi, 5–6, 9, 11, 15–16,
distributed meaning 268, 274–275 56–57, 59, 61, 67, 74–75, 79, 136–137,
Dutch 67–69, 149, 158–160, 162, 179, 139–140, 144, 148, 153, 160, 162–163,
183, 189, 194, 196–197, 204, 206, 208, 166, 176, 258, 293–300, 302–303,
210–212, 327 309–314, 357, 382, 395, 478, 479–480,
Dyirbal 512 483–484, 486, 488, 490–491, 494–497,
dynamicity 11, 259–260, 262 511
French 14, 101, 114, 137, 171, 177–180,
E
182–183, 190, 192, 213, 228, 248, 255,
ego-RP 488–490, 494
265, 315, 380, 390, 392–393, 399–401,
egocentric 2, 39–41, 56–57, 67–69, 89, 139,
403, 405–418, 419, 421, 438–440,
141, 159, 294, 413, 499–503, 512–513
447–449, 468, 512
English vi, 8–12, 14, 16–17, 48, 55–57,
front/back vi, 56, 77, 86, 121, 145–146, 293,
62–64, 68–69, 79, 84–88, 90–91, 96,
295–304, 310–311, 313, 483, 487, 504
98–100, 104, 109–114, 137, 139–151,
functional category 232, 236, 238–239,
154–161, 165, 168, 174–177, 179–191,
242–243
194–196, 205–206, 208, 210–212,
functional relations 112, 136
215–216, 226–228, 242–243, 246–248,
255, 264–265, 267–273, 280, 282, 285, G
289, 291, 296–297, 299, 302, 313, 315, Gestalt psychology 23, 33, 48
320–329, 331, 333–337, 339–343, geometry 1–2, 6, 9, 13, 17, 55, 64, 70,
348–349, 354, 370, 380, 384, 390–392, 95–100, 103–106, 110, 112–113, 116,
397–398, 400–401, 404, 414, 417, 419, 140, 145, 163, 194, 197, 210–212, 325,
421, 423–424, 426, 438–440, 447, 449, 331–332, 336, 352, 358, 364, 377–379,
453, 459, 462, 465–476, 479–482, 504, 506, 511, 514
484–490, 492–496, 507, 513 geons 23, 30
embodiment 2, 16, 42, 47, 375, 381, 476 German 33, 79, 88, 101, 268, 402, 493–494,
emotional states 90 496, 512
evolution 2, 4, 17, 36, 51, 74, 88, 173, Gerstmann syndrome 151, 164, 167–168
177, 179–180, 348–349, 453, 455, 460, gesture vi, 12–13, 85, 92, 160, 333, 351–
474–475, 496 363, 365–367, 369–384, 487, 491–493,
Ewe 162, 177, 256 495, 497
exaptation 15, 456–457, 477
 INDEX 517

gesture space vi, 13, 351–353, 355–357, L

359, 365–367, 369, 376–378 Latin 29, 182, 512
Greek 1, 28–29, 55, 62–63, 85–86, 91, 401, LCCM Theory 9, 216–217, 225, 228,
424, 467–472, 475–476 237–238, 246
Guugu Yimithirr 79, 147, 151, 154, lexical concept v, 9–10, 45, 215–220,
159–160, 381–382 222–247
lexical formation 171–175, 180–181,
H
184–185, 188–189
hand motion 13, 351, 367
lexical profile 226, 229–230, 237, 246
hand shape 13, 355, 358, 360, 367
lexicon 52–53, 74, 88, 162, 171–172, 184,
hippocampus 2, 40, 48, 76, 166, 500,
191, 246–247, 254–255, 319, 331–332,
513–514
345, 466, 469, 478
horizontality 258, 262, 322
linguacentrism 14, 420–421
human language faculty 88
linguistic relativity 17–18, 77, 114, 168,
I 191–192, 348, 384, 389–390, 400–402,
iconicity 12–13, 338, 340, 346–347, 361, 413, 417–418, 421, 448, 472, 478
372, 381, 384, 491 linguistic typology 91, 98, 151, 153–155,
identification and recognition 21–22 160–161, 163, 165, 264, 394, 415, 418,
image schema 41–47, 88, 191, 362, 364, 448
374, 380, 472 localism 4
implicational scale 8, 171–175, 184–185 localization 80, 86, 171, 179–180, 183, 188,
inferencing 267, 283, 299, 378–379 190
inferior parietal lobe 149 located object 111, 115, 118, 120, 128,
instrumental case 11, 274, 278, 282, 131–132
284–285, 287–289
M
intentionality 284, 421–422, 433, 444, 447,
Mandarin 85, 90, 380, 459, 476, 488,
449
495–496
interference effect 82–83, 464
manner verbs 406, 411, 416, 449
intrinsic orientation 280–281, 295, 501
maps 21, 39–41, 88, 101, 144, 154, 258,
intrinsic frame of reference 145, 151,
297, 381, 423, 497
153, 158, 258, 294–295, 300, 483–487,
mental imagery 80–81, 377
490–492, 494
mental metaphor 15, 457–458, 464, 472,
Indonesian 85–86, 91, 113, 175, 256,
474
475–476
mental rotation 89, 92
J mental simulation 80
Japanese vi, 10–11, 67–69, 88, 101, 111, metaphor 5, 15, 48, 84–85, 91–92, 216, 351,
114, 148, 160, 268, 293, 296–304, 307, 353–354, 365, 376–377, 380–385, 453,
310–315, 320, 401–402, 480 455, 457, 460, 464, 471–472, 476, 488,
497–498
K
metalanguage 421, 436, 447
Korean 99, 101, 111, 155–158, 176,
Mixtec 88, 137, 328–329, 349
179–180, 184–185, 187–191, 321, 329
Modern English 114, 180, 183, 190
518 LANGUAGE, COGNITION AND SPACE

Modern Greek 176 Perky effect 81–82

motion act 426, 434 positron emission tomography (PET) 7, 86,
motion activity 392, 422 149–150, 153
motion event 14, 54, 59, 61–62, 64–65, 75, polysemy 10–11, 191–192, 215, 223–225,
89, 252, 265, 338, 341, 367, 389–393, 228–231, 237, 246–248, 264, 267–268,
398, 400–401, 415–418, 422–424, 270, 274–275, 280, 283–284, 289, 291,
434–436, 443–444, 447–449, 482 319, 331, 480
motion figure 421–423, 430, 433–434, 464 posture verb 255, 257, 259, 262
motion goal 422 pozadi 272–273
motion linguistics 421 pragmatic strengthening 217, 223–224
motion verb 10, 55, 63–64, 200, 252–255, pragmatics 92, 113, 166, 265, 379, 383–384
257, 260, 262, 264, 399, 407, 424, 481 presupposition 508
motion type 422, 429, 445–446 primitive features 88
motor circuitry 90 Principled Polysemy vi, 9, 11, 140, 215,
motor resonance 80, 92 217, 222–224, 228, 246, 267–268, 270,
multimodality 383 275, 289
musical pitch 15, 453, 473–474, 476 projective relations 7–9, 140, 144, 147–149,
153–154, 158–159
N
proto-scene 118, 223–224, 228, 267,
navigation 37, 39, 154, 164–165, 168, 500
270–273, 275, 280–283, 286, 289, 302
neural model 12, 346
psycholinguistics vii, 104, 109, 111, 140,
O 251, 255, 294, 403, 415–416, 438
Old English 175, 177, 180, 183, 188 purpose sense 285–286, 288–289
open-class 9, 12, 228–230, 320, 329, 333,
Q
337, 346
qualitative physics 6, 97–100, 103–105,
optic flow 37–38
107–108, 111
P
R
parameter 10, 110, 116–117, 134, 141–142,
reference frames 7, 11, 16, 39, 57, 66–69,
158, 160, 198–199, 201, 210–211, 218,
116, 134–135, 297, 325
224, 227–236, 238–246, 354–356, 358,
reference object 7, 54–55, 111, 115–120,
394–398, 420, 469
123–124, 126, 128, 130–135, 243, 296,
parietal cortex 2, 39, 74, 86, 167, 500
310, 402
passive construction 506
reference systems 53–57, 66, 69, 482
path 15, 26, 45, 55, 62–65, 83, 159, 164,
relative frame of reference 67, 79, 145–146,
183, 193, 205, 262, 286–287, 289,
149, 154, 158–159, 259, 295–297, 300,
324–325, 327, 330–336, 338–345, 348,
310, 483–486, 488, 490–491, 493–494
352, 359, 365, 374, 390–393, 395–396,
reorientation 57–59, 70, 75
399–416, 418, 420, 422–427, 429–430,
robots 89
434–449, 463, 465, 496
routes 40–41, 212
path verbs 63, 393, 406
Russian vi, 10–11, 88, 101, 112, 267–270,
perceptual meaning analysis 23, 47
272, 274, 284–285, 289–290, 391,
perceptual simulation 80
417–418
 INDEX 519

S time estimation 91, 453, 463–465, 469–470,

satellite-framed pattern 423 472–473, 475–476
scene analysis 23, 36, 47 telicity 14, 409, 422, 424, 429, 435–436,
scene parsing 319, 334, 338 441, 443–444, 446, 449
secondary reference object 324–325, 328, top-down processing 23
335 topological relational marker 252, 256
selectional tendencies 218, 226, 229–230, topological relations vi, 7, 140, 142–144,
237, 246 148–149, 152, 157–158, 174, 251–252,
semantic extension 267–268 261–262, 264
semantic maps 17, 100, 113, 210, 212 Tzeltal 8, 57, 67, 100, 112, 139, 142–143,
semantic primitives 79 145, 147–148, 151, 153–154, 158–161,
semantic typology 77, 100, 104, 114, 163, 166, 173, 175, 179, 191, 391, 417,
166–168, 264, 480 501
semiotics 375, 380, 383
V
sensation 21–22
vantage point 222, 231, 273, 280–283, 286
sensory systems 22–23, 29, 39
vectors 6, 9, 16–18, 193, 197, 199, 210–211,
sign language 12, 150, 166, 333, 336, 343,
213, 503–506, 508, 510–511, 513–514
349
verticality 258, 262, 299, 502
signed language vi, 12, 319–320, 333–340,
visual cortex 24–25, 86, 88, 499
342, 345–349, 380, 385
visual imagery 90
similarity judgement 420
simulation 83, 91 W
simulation semantics 80, 90 Wernicke’s area 86
Spanish 55, 85–86, 91, 141, 144, 174, 176, Whorfian hypothesis 52, 67, 73, 157, 415,
179–182, 184–189, 210, 255, 265, 340, 466
400–401, 417, 447, 467–468, 475–476,
Z
488
za vi, 11, 267–284, 286–290
Spatial case 252, 260
spatial noun 253–259, 261–263, 265
spatial perception 16, 21, 29, 86–87
spatial schemas 12, 75, 84, 319–320,
322–325, 328–331, 345, 349, 459, 474,
477
spatial templates 7, 116–118, 120
stativity 259–260
supramarginal gyrus (SMG) 8, 86–87,
149–150, 153–154, 156
T
temporal frames of reference vi, 291,
479–480, 482, 484, 488, 493
textons 23, 29–30, 48
time-RP 488–490, 494