Grammatical Relations

OX F O R D S U R V E Y S I N S Y N TA X A N D M O R P H O LO G Y
g en er al e dit o r: Robert D Van Valin, Jr, State University of New York , BuValo
a d v i so r y e d i to r s: Guglielmo Cinque, University of Venice ; Daniel Everett,
University of Manchester; Adele Goldberg, Princeton University; Kees Hengeveld,
University of Amsterdam; Caroline Heycock , University of Edinburgh; David
Pesetsky, MIT ; Ian Roberts, University of Cambridge; Masayoshi Shibatani, Rice
University ; Andrew Spencer, University of Essex; Tom Wasow, Stanford University

published
Grammatical Relations
by Patrick Farrell

i n prepar ation
The Acquisition of Syntax and Morphology
by Shanley Allen and Heike Behrens
The Processing of Syntax and Morphology
by Ina Bornkessel and Matthias Schlesewesky
Phrase Structure
by Andrew Carnie
Information Structure : the Syntax–Discourse Interface
by Nomi Erteschik-Shir
Morphology and the Lexicon
by Daniel Everett
Syntactic Change
by Olga Fischer
The Phonology–Morphology Interface
by Sharon Inkelas
The Syntax–Semantics Interface
by Jean-Pierre Koenig
Complex Sentences
by Toshio Ohori
Extraction Phenomena
by David Pesetsky and Norvin Richards
Syntactic Categories
by Gisa Rauh
Computational Approaches to Syntax and Morphology
by Brian Roark and Richard Sproat
Language Universals and Universal Grammar
by Anna Siewierska
Argument Structure : The Syntax–Lexicon Interface
by Stephen Weschler
Grammatical Relations

PAT R I C K FA R R E L L

1
 3
Great Clarendon Street, Oxford ox2 6dp
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
ß Patrick Farrell 2005
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First published 2005
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloguing in Publication Data
Data available
Typeset by SPI Publisher Services, Pondicherry, India
Printed in Great Britain
on acid-free paper by
Biddles Ltd., King’s Lynn, Norfolk
ISBN 0-19-926401-5 978-0-19-926401-8
ISBN 0-19-926402-3 (Pbk.) 978-0-19-926402-5 (Pbk.)
1 3 5 7 9 10 8 6 4 2
 Contents
Abbreviations viii

1 Introduction 1
1.1 DiVerent kinds of grammatical relations 3
1.2 Case, voice, and grammatical relations 8
1.3 DiVerent theoretical approaches to grammatical relations 14
1.3.1 Subject and direct object 14
1.3.2 Indirect object 16
1.3.3 Oblique and other syntactic functions 28
1.3.4 Semantic roles 38
1.4 Theories of grammar 42
2 Grammatical relations across languages 44
2.1 Grammatical relations and major typological parameters 44
2.1.1 Accusative languages 46
2.1.2 Ergative languages 49
2.1.3 Split-intransitive languages 54
2.2 Voice and grammatical-relation alternations 62
2.2.1 Passive voice 65
2.2.2 Antipassive voice 71
2.2.3 Inverse voice 74
2.2.4 Applicative constructions 83
2.2.5 The Philippine voice system 91
2.3 Function splitting and quasi-subjects/objects 96
2.3.1 Quasi-subjects and oblique subjects 98
2.3.2 Quasi-objects and oblique objects 104
3 Relational Grammar 112
3.1 Basic design of the theory 112
3.2 Basic language types 116
3.3 Some case studies 118
vi Contents

3.3.1 Dative quasi-subjects 118

3.3.2 Phillipine voice alternations 119
3.3.3 A/O-reversal in Jarawara 121
3.3.4 Unaccusativity and the alignment problem 122
3.3.5 Icelandic dative subjects 127
3.4 Other relational theories 130
3.4.1 Lexical-Functional Grammar 130
3.4.2 Arc Pair Grammar 134
4 Role and Reference Grammar 135
4.1 Basic design of the theory 135
4.1.1 Lexical decomposition and semantic roles 135
4.1.2 Macroroles 139
4.1.3 Clause structure and syntactic functions 143
4.2 Basic language types 148
4.2.1 Voice constructions across languages 148
4.2.2 Ergativity, accusativity, and split intransitivity 152
4.3 Some case studies 153
4.3.1 Preverbal pronominal slots in Jarawara 154
4.3.2 Passive and switch reference in Seri 154
4.3.3 Dative subjects and quasi-subjects 155
4.3.4 Applicative objects vs. quasi-objects in Indonesian 160
4.3.5 Quasi-objects in Brazilian Portuguese 161
4.4 Other functional theories 165
4.4.1 Functional Grammar 166
4.4.2 Cognitive Grammar 168
5 Transformational Grammar 173
5.1 Basic design of theory 173
5.1.1 Classical TG 173
5.1.2 The GB approach 176
5.1.3 The Minimalist approach 178
5.2 Typological variation 181
5.2.1 Ergativity 181
5.2.2 Split intransitivity 185
 Contents vii

5.3 Some case studies 188

5.3.1 Multiple subjects and quasi-subject phenomena 188
5.3.2 Applicative constructions and the indirect object notion 193
5.4 Related theories 197

Notes 199
References 217
Index 231
 Abbreviations
1 1st person
2 2nd person
3 3rd person

A Subject of transitive
Abl Ablative
Abs Absolutive
Acc Accusative
Antipass Antipassive
Appl Applicative
APG Arc Pair Grammar
Arb Arbitrary (or unspeciWed) reference
Asp Aspect
Aux Auxiliary

Ben Benefactive

C Complementizer
CG Cognitive Grammar
Cont Continuative
CP Complementizer phrase (¼ clause)

Dat Dative
Decl Declarative
Det Determiner
DCA Direct core argument
Dist Distal realis mood
DS DiVerent subject

EPP Extended projection principle (¼ clausal subject requirement)

Erg Ergative
Excl Exclusive

Fem Feminine
FG Functional Grammar
Fut Future
 Abbreviations ix

Gen Genitive
GB Government-Binding Theory

I InXection
Imperf Imperfective
Iness Inessive
Ind Indicative
Inf InWnitive
Instr Instrumental
Intr Intransitive
Inv Inverse voice
IP InXection phrase (¼ clause)
Irr Irrealis mood

Lcontr Limited control

LF Logical form
LFG Lexical-Functional Grammar

Neut Neuter
Nom Nominative
Nominal Nominalization
Nonfut Nonfuture
NP Noun phrase

Masc Masculine
M-transitive Macrorole transitive
MR Macrorole
Neg Negative
NSg Non-Singular

O Object
Obj Objective
Obl Oblique
O1 Primary object
O2 Secondary object

P Preposition
Part Partitive
Pass Passive
Perf Perfective
Pl Plural
Pos Possessive
PP Prepositional phrase
x Abbreviations

Pred Predicative
Pres Pesent tense
pro null pronoun
PSA Privileged syntactic argument
Purp Purposive
Pst Past tense

Real Realis mood

Rec Reciprocal
Recip Recipient
ReX ReXexive
Rel Relational aYx
Relat Relative clause
RG Relational Grammar
RRG Role and Reference Grammar

S Subject of intransitive
Sa A-marked S
So O-marked S
Sg Singular
Subj Subject
Sjunct Subjunctive
Stim Stimulus

TG (Generative)-Transformational Grammar
Tns Tense
Tr Transitive

V Verb
VP Verb phrase
 1

Introduction

Language is used to say things about things that happen in the world. We
conceive of what happens in the world as involving people and other entities
doing things, often to each other. In order to say something about things that
happen, it is critical to identify the participants and the precise process in which
they are involved. Noun phrases (NPs) such as the boy, a dog, and the branches
are used to identify event participants, by virtue of the fact that they are built on
nouns, which (typically) name kinds of things. Verbs, such as touch, hit, sneeze,
and shout, are used to identify types of processes. Knowing the syntactic and
semantic conventions for constructing NPs, how the world is generally con-
sidered to operate, and the meanings of the words the, boy, touched, and
branches is not enough to tell me everything I need to know about the meaning
of a sentence such as (1.1) in a hypothetical language like English—but with no
conventions governing the relative order of verbs and noun phrases.
(1.1) Touched the branches the boy.
Hypothetical free word-order English
I would know that contact was made between the participant named by the
branches and the participant named by the boy. However, I wouldn’t know
whether the branches, aVected perhaps by the wind, made contact with the
boy who was stationary or whether, instead, the boy reached out and made
contact with the branches. The participants in the touching event play
diVerent roles: one is conceived of as the entity that does something and
one as the entity to which something is done. Languages use diVerent
mechanisms to indicate which entity plays which role. For example, the
NPs themselves may be marked by a morpheme that indicates something
crucial about the role they play. Thus, the -t and -ick suYxes in the following
version of (1.1) in a hypothetical English might indicate that the boy is the
acting participant and the branches are the target of his action.
(1.2) Touched the branches-ick the boy-t.
Hypothetical free word-order English with dependent marking
2 Introduction

I would simply need to know that by convention the NP marked with -t is to be

interpreted as the toucher and that the NP marked with -ick is to be interpreted
as the touched entity. Languages that mark role relationships in this way are said
to be dependent-marking languages (Nichols 1986), because the morpho-
logical indicators of the roles played by the participants in the process desig-
nated by the verbal elements in the clause (i.e. the verb and/or any auxiliary
verbs) occur on or within the phrases designating the participants themselves,
that is the dependents of the verb or verbal complex. English actually uses
dependent marking—but only in certain circumstances. For example, if a
third person singular masculine pronoun were used in place of the NP the boy,
the form he would be used rather than him. By convention the pronoun he is
necessarily interpreted as identifying the toucher when used as a dependent of
the active verb touched. For more peripheral roles, such as that of an instrument
used in an action, English uses prepositions to mark dependents. In The boy
touched the branches with a stick, the preposition with marks the NP a stick as the
referent of an instrumental participant in the touching event.
Another kind of role-identifying mechanism that languages use is known as
head marking (Nichols 1986). In a head-marking language the verb of a
clause and/or the auxiliary verb(s) bear morphological indications of the
features of one or more of the participants in the designated process. In a
hypothetical head-marking English, the suYx -ish on the verb touch, in a
sentence such as (1.3), might indicate that the touched entity is designated by
a third person plural NP and the suYx -ant might indicate that the toucher is
designated by a 3rd person singular NP.
(1.3) Touch-ish-ed-ant the branches the boy.
Hypothetical free word-order English with head marking
The fact that the branches is plural and the boy is not would provide a
suYcient clue for me to ascertain that the boy is to be interpreted as the
toucher. Actual contemporary English also uses head marking to a limited
extent, since present tense verbs with a 3rd person subject, for example, have a
special morphological form that registers agreement with one of the partici-
pants (The boy touches the branches vs. The branches touch the boy).
A third common kind of role-identifying mechanism—the one that is most
important in actual English—is constituent order. By convention the NP that
precedes the active verb form touch is interpreted as designating the touching
participant and the NP immediately following it is interpreted as designating
the touched participant. Thus, in the following example, the fact that the boy
comes Wrst provides suYcient role information for me to know that the boy
did the touching.
 Introduction 3

(1.4) The boy touched the branches.

Actual English
This book is devoted to the domain of grammar having to do with the ways
in which languages identify these kinds of roles—broadly known as gram-
matical relations—and the function these roles play in the grammars of
languages. Among the issues of concern are what kinds of roles are recognized
in languages, how they should be categorized, how languages diVer in their
use and characterization of such roles, and how diVerent theories of grammar
approach these matters. The remainder of this chapter lays out the key issues
surrounding grammatical relations. Chapter 2 surveys important phenomena
in languages in which grammatical relations play a critical role and identiWes
diVerent kinds of grammatical-relation systems employed in languages. The
remaining chapters examine the varying approaches to grammatical relations
and related phenomena of several diVerent prominent theories.

1.1 DiVerent kinds of grammatical relations

It is possible to imagine a dependent-marking language in which a particular
suYx is devoted to identifying NPs whose referent plays the role of toucher
with the verb touch, another is devoted to identifying NPs whose referent
plays the role of thrower with the verb throw, yet another is devoted to
identifying NPs whose referent plays the role of killer with the verb kill, and
so forth. At one extreme of logical possibilities, for every verb there would be
an idiosyncratic morphological mechanism for identifying the particular roles
involved in the speciWc kind of process that verb designates. This kind of a
system would obviously be very ineYcient; and languages do not employ such
a system. Instead, their marking systems are organized around relatively
general role categories of various kinds. If one takes, for example, the set of
English verbs shown in Figure 1.1, it is clear that, for each of the verbs, the roles
of the dependents are essentially the same and could be given descriptive
labels of some kind, as indicated.
A key question for a theory of grammar is what type of role categories play
an integral role in the grammars of languages. For the set of verbs shown in
Figure 1.1, at least two generalizations could be made: (a) recipients are
indicated by the preposition to and (b) sources are in preverbal position.
However, if more verbs are taken into account it becomes clear that, although
it may be a relevant factor, designating what is commonly thought of as a
recipient—that is someone who is conceived of as getting something—is not
in fact requisite for an NP to co-occur with the preposition to, and designating
4 Introduction

SOURCE RECIPIENT

The woman a book to the teacher

x y z
gave Process = x does something to y
threw and because of this z gets y from x
delivered
brought
sold
handed
etc.

Figure 1.1 Potential role categories for one class of English verbs

what is commonly thought of as a source is not requisite for an NP to occupy

the preverbal slot in sentence structure. One can see this from the four classes
of verbs shown in Figure 1.2.
Sentences with verbs of communication, such as promise and explain, and
sentences with certain verbs of perception, such as seem and look, show that
the preposition to can go with dependents whose roles might better be
described as addressee and perceiver. Now, although it might well be that to
is simply a multifunctional word that is used to indicate diverse kinds of roles,
because these roles are often indicated in the same way in many languages, it
seems plausible that there is a single very general role category of some kind
that encompasses these (and other) highly speciWc roles. However, although
there may be enough similarity between the roles of addressee, perceiver, and
recipient as to allow some kind of an abstract semantic characterization,
perhaps something like ‘participant that actively participates in a process in
some kind of receptive way’, there is no general purpose English word that
names such a concept. One tack that has often been taken is to use a label with
no independent meaning at all, such as ‘indirect object’.
Along similar lines, sentences such as The woman stole/took the book from
the teacher make it clear that the role of the preverbal NP can be described as
that of recipient rather than source.1 In both cases, however, there is an
alternative way of looking at the role of the referent of the preverbal NP,
since it can usually be interpreted as the participant that initiates the action
that the verb designates. The descriptive label ‘agent’ is typically used for this
more schematic and more encompassing role category. The acting participant
in a communicative act can also be characterized as an agent. However, the
preverbal NP in a sentence such as The woman looks strange to me doesn’t seem
to refer to an initiator of an action, at least not in anything like the same way.
The question What did the woman do? can have She promised me something or
 Introduction 5

PERCEPT PERCEIVER

The woman strange to me

x y z
seems Process = z perceives x
SUBJECT looks as being y
SPEAKER ADDRESSEE

INDIRECT
The woman something to the teacher OBJECT
x y z
promised
Process = x says something to z
explained
and because of this […z…y…]
admitted
AGENT etc.
SOURCE RECIPIENT

The woman the book to the teacher

x y z
sold Process = x does something to y
gave and because of this z gets y from x
delivered
RECIPIENT etc. SOURCE

The woman the book from the teacher

x y z
bought
Process = x does something to y
took
and because of this x gets y from z
stole
etc.

Figure 1.2 DiVerent kinds of role categories for four classes of English verbs

She took something from me as an answer, but not She seemed/looked strange to
me. Although there is clearly some kind of correlation between preverbal
position and the role of agent, being an initiator of an action is not criterial.
Even if one were to deWne the agent role in such a way as to have it encompass
what might be characterized as a causing percept role,2 there are other more
problematic preverbal NPs, such as in The books were stolen from the library.
Thus, if it is assumed that there is a single role category for the preverbal NP,
there does not seem to be an obvious descriptive label that elucidates its
conceptual character. The label that is generally used is ‘subject’.
Just considering a few sets of verbs in English, it is clear that there are
potentially various ways of categorizing the roles played by the dependents of
verbs. One can use quite speciWc role categories with descriptive labels
6 Introduction

that emerge from a consideration of the detailed meanings of semantic verb

classes (e.g. source, recipient, addressee, agent) as well as categories that
conXate or encompass various speciWc roles and whose labels make little or
no claim about their conceptual character (e.g. subject and indirect object).
But which of these role categories are useful for describing the grammar of
English and other languages? Considering, for example, the phenomenon of
verb agreement in English (manifested, for example, in the use of the -s suYx on
present tense verbs and the varying forms of the verb be), roles such as recipient,
source, and agent are not directly relevant in any obvious way. The verb agrees
with the preverbal NP (or subject), whatever more speciWc role it happens to
play (The woman steals books from people, The woman seems strange to us, The
books were stolen from the library). Although there do not appear to be any
morphosyntactic phenomena in English whose characterization depends on
such speciWc roles as percept or addressee, roles of this general kind can have
grammatical ramiWcations in English and other languages. For example, there
is a rule of English with the eVect of ensuring that an NP with a source referent
that is not realized as a subject is, at least by default, marked with the prepos-
ition from, as in The woman stole the book from the library or I removed the label
from the box and another rule (Foley and Van Valin 1984: ch. 3) that ensures that
if the source is encoded as the direct object (¼ immediately postverbal NP), the
NP with the role of aVected thing is necessarily marked with the preposition of,
as in the case of They robbed him of his cash or We’ll have to rid the room of insects.
By contrast, if a goal, that is a location to which something is moved, is encoded
as the direct object, the preposition with is used for the aVected thing, as in
We Wlled the glass with water or They loaded the wagon with hay.
The use of such role categories as agent and source in linguistic theorizing
has its ultimate roots in Pānini’s grammar of Sanskrit from the 5th century bc
(Kiparsky and Staal 1969,_ Singh 2001, Kiparsky 2002), which includes a
procedure for specifying the kārakas or roles of the dependents of verbs, on
the basis of verb meanings, and various kinds of grammatical rules that
depend on these. In the Sanskrit sentence corresponding to The man came
from the forest, for example, a grammatical analysis would assign the role of
kartr ‘agent’ to the man, because it refers to the participant that is ‘independ-
ent’,_ and the role of apādpāna ‘source’ to the forest, because it refers to the
participant that is Wxed in an act of separation. It is important to note that
although these roles are recognized as having a bearing on choice of the
inXectional aYxes on nouns (i.e. the intricate system of dependent marking),
they are not conceived of as the same thing at all. For example, even though
odana ‘rice’ has the accusative ending in the active sentence (1.5a) and the
nominative ending in the passive sentence (1.5b), it bears the kāraka of kar-
 Introduction 7

man ‘patient/goal’, because it refers to the participant in the action that is the
agent’s main target.
(1.5) a.Krsna-h paca-ti odana-m
__ _ _
Krishna-Nom cook-3Sg rice-Acc
‘Krishna cooks rice.’
b. Krsn-ena pac-ya-te odana-h
_ _ _
Krishna-Instr cook-Pass-3Sg rice-Nom _
‘Rice is cooked by Krishna.’ (Kiparsky 2002)
Although the -m suYx is a morphological reXex of the karman role, it does not
appear in (1.5b) because the agreement suYx on a passive verb form is analysed
as the morphological reXex of this role, for which reason the default nomina-
tive ending is used instead. In essence, this analysis claims that role identiWca-
tion in Sanskrit crucially involves both head marking and dependent marking.
In combination with information from the passive morpheme or lack thereof,
the verb’s agreement morphology (i.e. head marking) is used to identify either
the agent or the patient. The other roles are identiWed by the inXectional
morphology on the dependents. Although kārakas necessarily conXate what
can easily be conceived of as distinct roles, since only six of them are posited,
they are nevertheless fundamentally semantic categories around which the
rules of grammar revolve. In contemporary theories of grammar, these kinds
of roles are generally either called thematic roles/relations (sometimes
abbreviated as ‘u-roles’), following Gruber (1965), case relations/roles,
following Fillmore (1968), or simply semantic roles/functions.
Concepts such as subject, direct object, and indirect object, which are
thoroughly ingrained in Western linguistic theories, have their roots in the
notion of subject-predicate structure in logic and grammar from Aristotle’s
Categories and traditional grammars of Latin and Greek which, unlike Pānini’s
grammar of Sanskrit, take the dependent-marking morphology in these_ lan-
guages to stand essentially in a one-to-one correspondence with the primary
grammatical-relation categories. In Latin, like in Sanskrit, which dependent is
marked with so-called nominative morphology depends in part on whether
the verb is in its passive form or not, as illustrated by the following sentences.
(1.6) a. Roman-i null-os tyrann-os laudabant
Roman-NomPl no-AccPl tyrant-AccPl praise.Imperf3Pl
‘The Romans used to praise no tyrants.’
b. Null-i tyrann-i ab Roman-is laudabantur
no-NomPl tyrant-NomPl by Roman-AblPl praise.Imperf3PlPass
‘No tyrants used to be praised by the Romans.’
8 Introduction

With an active verb, as in (1.6a), the patient is marked with the accusative
ending, whereas the dependent with the same role is marked with a nominative
ending when the passive form of the verb is used, as in (1.6b). Under the
assumption that the dependent-marking morphology fundamentally an-
nounces the main grammatical relations, ‘the Romans’ (agent) and ‘no tyrants’
(patient) can alternatively play the same role, depending on the form of the
verb. Since this role does not have a clear semantic characterization, one must
either use a very schematic (or vague) label, such as ‘subject’, or a label with no
other meaning at all, such as ‘nominative’. Accommodating languages such as
contemporary English in which the dependent-marking morphology trad-
itionally designated by the word nominative plays a subordinate role to con-
stituent order in indicating the role in question, the most common term for
this role is ‘subject’. In Latin, the subject of the sentence is said to be marked
with nominative morphology and to determine verb agreement, etc. The direct
object is said to be marked with accusative morphology; and the indirect object
with dative morphology. These kinds of grammatical roles, which are less
clearly amenable to a semantic characterization than roles such as agent and
source, are often formally distinguished in theories of grammar and are
alternatively called syntactic relations/functions or grammatical
relations/functions, in contrast with semantic roles.
As is made clear in what follows, there are theories or approaches to gram-
matical analysis, including that of Pānini discussed above, that fail to draw a
_
clear distinction between syntactic functions and semantic roles, and not all
theories that do draw a distinction do it in the same way. Moreover, one of the
main concerns of theories that clearly distinguish syntactic functions and
semantic roles is the nature of the principles governing the relationship between
them, since NPs are conceived of as having both and the syntactic function is
systematically related to the semantic role. For these reasons, in this book the
term grammatical relation is used in a maximally general and inclusive way. The
terms semantic role and syntactic function are used to draw a distinction where
appropriate, with the understanding that diVerent theories take varying stances
about where, if at all, the boundary between the two categories lies and adopt
varying terminological conventions. Nevertheless, syntactic functions such as
subject and direct object, which are taken to be prototypical of the grammat-
ical-relation category, constitute the main focus of attention.

1.2 Case, voice, and grammatical relations

Two morphological categories that are intricately involved in the identiWca-
tion of grammatical relations in many languages are case, that is morpho-
 Introduction 9

logical marking on or within NPs that provides information concerning

grammatical relations (see Blake 1994), and voice (Klaiman 1990), that is
inXectional morphology on a verb or other morphosyntactic marking that
functions to indicate which of two or more alternative interpretations of the
grammatical relations of its dependents is intended. Both of these categories
were illustrated in the Sanskrit and Latin examples discussed above. Consider
again the following Sanskrit sentences and Pānini’s analysis (Kiparsky 2002).
_
AGENT PATIENT

(1.7) a. Krsna-h
.. . . paca -ti odana -m
Krishna-Nom cook-3Sg rice-Acc
‘Krishna cooks rice.’

AGENT PATIENT

b. . . . -ena
Krsn pac-ya -te odana-h.

Krishna-Instr cook-Pass-3Sg rice-Nom

‘Rice is cooked by Krishna.’
If the verb is in its active-voice form (indicated by absence of a passive
morpheme), it agrees with the dependent bearing the agent role. The patient
is identiWed by the accusative case marking. If the verb includes the passive-
voice suYx, however, it agrees with the dependent bearing the patient role.
The instrumental case suYx on the other dependent indicates that it bears the
agent role. In both sentences, the dependent that agrees with the verb takes
the nominative case suYx, for which reason nominative case itself does not
specify a particular semantic role. An alternative analysis using the subject and
direct object grammatical relations would go as follows. The nominative case
suYx indicates the subject in both cases. The accusative case suYx indicates
the direct object. If the verb is in its active-voice form, the dependent that
bears the agent role is also the subject and the dependent that bears the patient
role is the direct object. The passive-voice suYx on the verb indicates that the
patient is the subject, for which reason the agent is neither the subject nor the
direct object, but what is generally called an oblique, which happens to be
marked with instrumental case. Whichever analysis one adopts, information
from the case marking on the dependents and information from the voice
marking on the verb function jointly to identify the grammatical relations
(semantic and/or syntactic) of the dependents.
10 Introduction

Because the morphological phenomenon known as case-marking functions

primarily to indicate grammatical relations, the term case is sometimes used in
one way or another in place of grammatical relation. In one trend, initiated in
Fillmore (1968) and continued and developed in diVerent ways in, for example,
Localist Case Grammar (Anderson 1971, 1977) and Lexicase Grammar (Starosta
1988), case relation/role (or, sometimes, deep case) is used essentially as a
translation of Pānini’s kāraka. That is to say, case relations are considered to
_
be very general semantic roles with grammatical import, which may or may not
have an invariant or direct relationship with the morphological case categories
used in the dependent-marking system of a language. For Pānini, for example,
agent and instrumental are both among the recognized kārakas. _ The nominal
suYxes of Sanskrit, however, do not correspond exactly, as the agent role is
analysed as being indicated either by the agreement morphology on an active
verb or the instrumental suYx (when the verb agrees with some other depend-
ent, as in passive clauses). The instrumental suYx is used to either indicate the
instrumental role or, sometimes, the agent role.
In Lexicase Grammar there is a universal set of semantically characterized
case relations (agent, patient, locus, means, and correspondent) that form
part of the underlying basis for the use of case forms (nominative, accusative,
dative, etc.). Abstracting away from numerous representational idiosyncrasies
of the theory, the Lexicase analysis of the Sanskrit active and passive sentences
discussed above would go roughly as follows, where case relations are in
uppercase letters and case forms are not.
AGENT & PATIENT &
Nominative Accusative

(1.8) a. . . . -h.
Krsna paca-ti odana -m
Krishna-Nom cook-3Sg rice-Acc
‘Krishna cooks rice.’
MEANS & PATIENT &
Instrumental Nominative

b. . . . -ena
Krsn pac-ya-te odana -h.
Krishna-Instr cook-Pass-3Sg rice-Nom
‘Rice is cooked by Krishna.’
The idea is that the active verb in (1.8a) has a valence (Hockett 1958, Tesnière
1959) or a set of speciWcations concerning what it can combine with, that
 Introduction 11

includes a requirement for a dependent with the case relation agent (the
perceived external instigator of an action or state) and the case form nomina-
tive, which the NP krs nah satisWes. The nominative suYx is essentially a
_ _ _ on
morphological realization _ the NP of the nominative case form speciWed in
the verb’s valence. The NP odanam satisWes the verb’s valence requirement for a
dependent with the patient case relation (the perceived perceptual centre of an
action or state) and an accusative case form. The passive verb, shown in (1.8b)
has a diVerent valence, which speciWes that the patient dependent is linked with
the nominative case form and the other dependent is interpreted as a means
(the entity perceived as necessarily involved in an action or state aVecting the
patient) with the instrumental case form. Although case relations are essen-
tially semantic roles,3 case forms are not exactly the same thing as morpho-
logical case markers. The case form nominative, for example, is assumed to be
universal and is therefore a feature in the valence of verbs in languages that do
not have any kind of case-marking aYxes on or within NPs. Thus, case forms
are more abstract than case markers. They are, in essence, another form of
grammatical relation, which in some languages have a direct morphological
manifestation in the form of morphological case markers and in others may be
manifested by other morphosyntactic means, such as constituent order.
In another trend, initiated in the Government-Binding Theory incarnation
of Transformational Grammar (Chomsky 1981), the term case (often capital-
ized or called ‘abstract case’ to distinguish it from morphological case) is used
in reference to a role-diVerentiation and dependent-licensing mechanism that
works in terms of grammatical relations (see Chapter 5 for further details).
The use of case to designate both morphological case categories in languages
and grammatical relations is potentially confusing, inasmuch as the names of
the individual cases do not diVer for the two diVerent uses. In essence,
dependents bearing the subject role are said to be licensed by a nominative
case relation that holds between them and an inXectional/functional element
or position in a clause. Similarly, direct objects are licensed by an accusative
case relation that holds between them and a verb (or, in some versions of the
theory, an inXectional/functional element or position in a clause). In this
theory, what are called nominative and accusative case features constitute one
instantiation of the traditional subject and direct object grammatical rela-
tions. Although these relations occur across languages, the morphological
reXexes of the abstract case features vary considerably. A language may or may
not indicate them with either or both head marking (overt agreement) or
dependent marking (overt case marking). Although case, in this grammatical-
relation sense, is used for syntactic functions, in the form of nominative and
12 Introduction

accusative cases, also known as ‘structural cases’, it is also used for semantic
roles, insofar as ‘inherent’ or ‘semantic’ cases such as instrumental, locative,
partitive, and ablative are assumed to be licensed directly by semantic roles.
In view of the inXuence of analyses of case-marking distinctions in par-
ticular languages on the development of theories of grammatical relations as
well as the various ways in which the term case has been used within diVerent
theories, it is important to point out that there are good reasons for clearly
distinguishing grammatical relations and morphological case-marking cat-
egories, as suggested by such terminological contrasts as Case vs. case (Trans-
formational Grammar), case relation and case form vs. case marking (Lexicase
Grammar), and deep case vs. surface case (Fillmore 1968).
One particularly striking example of why grammatical relations and case-
marking categories need to be distinguished comes from Icelandic—a lan-
guage that is renowned for its nominative, accusative, dative, and genitive
subjects (Zaenen et al. 1985, Van Valin 1991, Barðdal 2001, Sigurðsson 2002).
Although Icelandic uses nominative case morphology for prototypical agent-
subjects in active-voice clauses and accusative case morphology for prototyp-
ical patient-direct objects, there are various constructions which, for diVerent
reasons, show alternative case-marking patterns. For example, there is a rather
large class of verbs having to do with psychological or emotive states that
allows either the NP designating the experiencer of the state or the NP
designating its stimulus (or target or cause) to function as the clausal subject.
However, in both of these instances the case marking is determined by the
experiencer vs. stimulus distinction, rather than by the subject vs. object
distinction. The following examples show that the stimulus of the verb
henta ‘like, please’ is in the nominative case form whether it precedes the
verb of its clause or follows it, just as the experiencer is in the dative case form
in either of these positions.
(1.9) a. Ég veit að þetta mun henta mér.
I know that this.Nom will please 1SgDat
‘I know that this will be pleasing to me.’
b. Ég veit að mér mun henta þetta.
I know that 1SgDat will like this.Nom
‘I know that I’ll be pleased with this.’ (Barðdal 2001)
SigniWcantly, there is a general convention in Icelandic that restricts the
preverbal position in clauses of the kind containing henta in these examples
to the subject. That is to say, there is a constraint on preverbal position in
clauses that is generally sensitive to a particular dependent of the verb—like
various other constraints in the language, including conjunction reduction,
 Introduction 13

subject–verb inversion, subject-to-subject raising, and reXexive anaphor ante-

cedence. With typical verbs designating physical actions, only the nominative
agent satisWes these constraints. Thus, the constraints are sensitive to some
kind of role category like subject (or the most prominent or most privileged
dependent of a clause, whatever this is called). Since with verbs like henta the
nominative stimulus can be in preverbal position and also manifests the other
subject properties, it follows that this dependent can be the subject of its
clause. However, since the dative experiencer can also be in preverbal position
and it too manifests the other subject properties, it follows that this depen-
dent can be the subject of its clause as well. Needless to say, in Icelandic
there is a role—ostensibly a subject role—that is identiWed by constituent
order and other correlative syntactic phenomena rather than nominative case
marking.
There are numerous similar phenomena in languages that make it diYcult
to maintain the view that case-marking categories reliably indicate syntactic
functions such as subject and direct object. In Finnish, for example, accusative
case can be used on certain constituents that are not direct objects and fails to
be used on certain constituents that are direct objects. More speciWcally,
accusative case is used for adjunct adverbial phrases, which are not direct
objects, and is replaced by partitive case on direct objects in the context of
imperfective sentential aspect, as illustrated by the following examples (from
Megerdoomian 2000).
(1.10) a. Matti luk-i kirja-t tunni-ssa
Matti.SgNom read-3SgPst book-PlAcc hour-Iness
‘Matti read the books in an hour.’ perfective interpretation
b. Matti luk-i kirjo-j-a tunni-n
Matti.SgNom read-3SgPst book-Pl-Part hour-Acc
‘Matti read books for an hour.’ imperfective interpretation
As noted in Kiparsky (2001), accusative adverbial phrases, such as tunni-n in
(1.10b) do not have the deWning syntactic properties of direct objects, such as
the ability to be expressed as the subject of a passive sentence or to control
raising to object or secondary predication. Moreover, although the partitive
vs. accusative distinction expresses something about the semantics of clauses,
it apparently does not express a grammatical-relation distinction, since the
partitive object has the deWning syntactic properties of direct objects
(Kiparsky 1998, Ackerman and Moore 1999). Thus, accusative case identiWes
neither all nor only direct objects.
The issue of mismatches between grammatical relations and morphological
marking is considered further in Section 1.3 and in subsequent chapters.
14 Introduction

SuYce it to say, for now, that one of the fundamental challenges for a theory
of grammar is that, although there is a systematic relationship between
grammatical relations and case-marking categories, the correspondence is
not always straightforward.

1.3 DiVerent theoretical approaches to grammatical relations

Theories of grammar diVer not only with respect to how they categorize
grammatical relations, but also with respect to how many and which relations
are recognized, how individual relations are labelled, and, of course, how,
precisely, the roles of dependents are assigned or determined.

1.3.1 Subject and direct object

If a theory recognizes the grammatical relations subject and direct object,
there is a fair amount of consensus concerning what is meant by these terms.
The basic idea is that the grammar of a language revolves primarily around
two core roles. At least in the most common basic language type (known as an
accusative language—see Section 2.1), of which English and most other Indo-
European languages are examples, the most syntactically privileged of these
two is the subject. Although dependents with various semantic roles can be
subjects (depending on the voice of verbs among other things), the proto-
typical subject is an agent (or the most active participant in an event) and the
agent is the subject in a basic active-voice clause. Similarly, although depend-
ents with various semantic roles can be direct objects, the prototypical direct
object is a patient (or the primary target of an agent’s action) and the patient
is a direct object in a basic active-voice clause.
However, not all theories recognize subject and direct object grammatical
relations. As noted above, these concepts played no role in Pānini’s grammar
of Sanskrit. In Lexicase Grammar, these labels are also not used, _ although the
case form nominative is close to the same notion as subject. One key diVer-
ence is that the grammatical relation subject, in theories that employ it, is not
necessarily constrained to being associated with the NP that bears nominative
case marking (or else there could be no such thing as dative subjects, for
example), whereas the nominative case form in Lexicase Grammar is. The
dependent that expresses the verb’s nominative case form need not manifest a
case feature morphologically; but if it does, this case feature cannot conXict
with the case form. In fact the notion subject can be characterized in alter-
native ways in Lexicase Grammar. Consider, for example how dative subjects
in Icelandic might be handled:
 Introduction 15

PATIENT & ACTOR & LOCUS &

Nominative Dative
9
(1.11) a. Ég veit a Þetta mun henta mér
I know that this.Nom will please 1SgDat
‘I know that this will be pleasing to me.’
PATIENT & ACTOR & MEANS &
Dative Nominative
9
b. Ég veit a mér mun henta Þetta
I know that 1SgDat will like this.Nom
‘I know that I’ll be pleased with this.’

Crucially, in this theory, case relations (semantic roles) need not be the same
for corresponding NPs across paraphrases. How a speaker conceives of the
situation described by a sentence is what counts—not what happens in
the situation from an objective point of view. Thus, either the experiencer
or the stimulus of the psychological state designated by henta can be conceived
of as the perceptual centre of the state, that is the patient. If the stimulus is
conceived of as the patient, as in (1.11a), the experiencer is naturally conceived
of as the locus (i.e. the perceived abstract or concrete location of the action or
state), for which reason the dative case form is appropriate, since it is one of
the common choices for the locus role. In addition to the case relations and
case forms, there is also a semantic macrorole called ‘actor’ (the entity
perceived as instigating or carrying out an action or bearing the attributes
of a state), which was omitted from the Lexicase analysis of Sanskrit active
and passive pairs presented in Section 1.2, for the sake of simplicity.4 Since
there is no accusative case form in the verb’s valence, the dependent bearing
the patient role is, by general rule, assigned the actor macrorole, which always
goes to one and only one of the dependents. The same analysis of case
relations, case forms, and macrorole assignment would be appropriate for
the English sentence This is pleasing to me. In (1.11b), the experiencer of the
psychological state designated by henta is conceived of as the patient. The
stimulus of the psychological state, which is an entity perceived to be neces-
sarily involved in the state or action aVecting the patient, plays the role of
means. Again, since there is no dependent with an accusative case form, and
this is therefore an intransitive clause, the patient has the actor macrorole.
The case relations and macrorole assignment are the same as in I am pleased
with this. What makes Icelandic diVerent from English is that the case form
16 Introduction

speciWcations remain dative vs. nominative in both of the valence frames of

henta, and the unmarked alignment of actor and nominative is abandoned in
the dative subject construction.5 Why then does the dative subject have the
usual grammatical properties of subjects (case marking aside)? A plausible
answer is that these phenomena are sensitive to the actor macrorole rather
than the nominative case form. That is, the rule for constituent order, for
example, speciWes that the dependent bearing the actor macrorole occupies
preverbal position in clauses.
In short, by appealing to a nontraditional set of grammatical relations
without the usual subject and direct object categories, Lexicase Grammar
makes possible alternative characterizations of what is generally thought of as
the subject of a clause. One kind of ‘subject’ is the dependent linked to the actor
macrorole; another kind is the dependent linked to the nominative case form.
Although the actor macrorole and nominative case generally coincide, they
need not. The traditional subject and object relations are also not employed in
Role and Reference Grammar (Foley and Van Valin 1984, Van Valin and La Polla
1997, and see Chapter 4), which, like Lexicase Grammar, provides alternative
ways of describing grammatical phenomena for which the traditional subject
role is usually invoked, including the possibility of appealing to an actor
semantic macrorole (a notion that in fact originates in this theory).6

1.3.2 Indirect object

Use of the indirect object grammatical relation and its characterization
are also subject to considerable cross-theoretical variation. When such a
category can be identiWed—which is not always the case—the prototypical
role appears to be that of recipient in an act of giving, with other semantic
roles such as beneWciary, addressee, perceiver, etc. being included in ways that
vary across languages (Dixon and Aikhenvald 2000). One kind of evidentiary
basis for an indirect object relation is that typological research on certain
syntactic phenomena suggests a grammatical-relation hierarchy with a third
core relation (in addition to subject and direct object). For example,
languages are known to diVer in terms of which types of clausal dependents
can be targeted by their relative-clause formation strategies (Keenan and
Comrie 1977). The key constraints can be formulated in terms of a hierarchy
such as the following.

(1.12) Subject > direct object > indirect object > oblique > possessor >
object of comparison
NP accessibility hierarchy (Keenan and Comrie 1977)
 Introduction 17

All languages appear to allow relativization of subjects.7 Some languages,

however, draw the line at subject; others at subject and direct object;
others at subject, direct object, and indirect object; and so forth. The top
end of the same hierarchy has been claimed to govern agreement systems
in languages (Moravcsik 1974, Whaley 1997: Ch. 3), which come in four main
types: no agreement (e.g. Lahu, Mandarin), agreement with the subject only
(e.g. Russian, Turkish), agreement with the subject and direct object (e.g.
Quiché, Ahmaric), or agreement with the subject, direct object, and indirect
object (e.g. Abkhaz, Manam).8 Implicit in these claims about the role of
a grammatical-relation hierarchy in syntactic phenomena is the idea that
the grammars of some languages revolve around three primary relations
rather than one or two. However, the empirical research underlying
such typological generalizations is based on the assumption that all languages
have indirect objects, which can be identiWed on an a priori basis (for
example, the recipient in a process of giving is an indirect object). In fact,
there is considerable variation across languages in terms of the way depend-
ents with a recipient role function in the grammar. Moreover, when a
special grammatical relation with recipient as prototype can be identiWed in
particular languages, diVerent theoretical characterizations of this relation are
possible.
The Romance language family provides good examples of a system
that appears to have three primary grammatical relations. In Brazilian
Portuguese there is a special pronominal form for direct objects. As the
following examples show, a 3rd person pronominal direct object can either
be placed in canonical direct object position following the verb, in the same
unmarked form in which it appears when it is a subject or the object of a
preposition (1.13a–b), or it can be realized as a verb-initial pronoun
in an accusative form (1.13c), that is a form restricted to direct objects
(technically a clitic, which the standard orthography doesn’t indicate as
such systematically).9
(1.13) a. Ela mudou ele.
3SgFem changed 3SgMasc
‘She changed him.’
b. Ele mora com ela.
3SgMasc lives with 3SgFem
‘He lives with her.’
c. Ela o mudou.
3SgFem 3SgMascAcc changed
‘She changed him/it.’
18 Introduction

The primary evidence for an indirect object category is that there is another
3rd person pronominal object clitic form that is not used for direct objects, but
for dependents of a verb that are conceived of as typically active, receptive
participants in the event described by the verb (i.e. recipients, beneWciaries,
addressees, or experiencers of cognition, perception or emotion), which is to
say, more or less the typical class of participants that trigger dative case marking
or indirect object agreement cross-linguistically. The following examples illus-
trate the use of dative lhe, which alternates with a phrase that is invariably
marked with a preposition (usually para ‘for/to’ or a ‘to’), as shown by (1.14a).10
(1.14) a. Ela lhe deu um presente / deu um presente para ele.
3SgFem 3SgDat gave a present gave a present to him
‘She gave him/her a present/gave a present to him.’
b. Isso lhe parece doido.
that 3SgDat seems crazy
‘That seems crazy to him/her.’
c. Ela lhe falou disso.
3SgFem 3SgDat spoke of.this
‘She spoke to him/her about this.’
d. *Ela lhe mudou.
3SgFem 3sgDat changed
‘She changed him/her.’
e. *Ela lhe mora.
3SgFem 3sgDat lives
‘She lives with him/her.’
There are various ways in which the class of dependents that can be expressed
as a dative pronominal clitic diVers from the class of dependents that can be
expressed as an accusative clitic (direct objects). For example, only the latter
can be the subject of a passive clause or the pivot in the so-called tough
movement construction and a certain kind of contrastive focus construction
known as the pseudocleft construction (see Section 2.3.2). However, the dative
class is distinguished from the class of obliques not only by the special
pronominal form but also by the fact that, like direct objects and unlike
obliques, its members can be the recipricolization pivot (or omitted element)
in the reXexive/reciprocal se construction:
(1.15) a. Eles se deram as mãos. (indirect object pivot)
3MascPl 3Rec gave the hands
‘They held hands.’ (Literally, ‘They gave each other the hands.’)
 Introduction 19

b. Eles nunca mais se falaram. (indirect object pivot)

3MascPl never more 3Rec spoke
‘They never again spoke to each other.’
c. Eles se amam. (direct object pivot)
3MascPl 3Rec love
‘They love each other.’
d. *Eles se moram. (*oblique pivot)
3MascPl 3Rec live
‘They live with each other.’
Thus, there are at least three main categories of non-subject dependents:
obliques, direct objects, and a kind of dependent that is like a direct object
in some ways and like an oblique in others.
As detailed in Chapter 3, in Relational Grammar (Perlmutter 1983, Perl-
mutter and Rosen 1984, Blake 1990), as in traditional analyses of English and
Romance languages, there are three primary grammatical relations around
which the grammars of languages are assumed to largely revolve: subject,
direct object, and indirect object. Other dependents of verbs fall into the
general category of obliques. Schematically, the analysis of Brazilian Portu-
guese non-subject dependents would be as follows.
direct object indirect object
(1.16) a. Ela deu [um presente] para [a mãe dela]
3SgFem gave a present to.the the mother of.3SgFem
‘She gave a present to her mother.’
oblique
b. Ela mora com [a mãe dela]
3SgFem lives with the mother of.3SgFem
‘She lives with her mother.’
Accusative pronominal clitics are reserved for direct objects; dative pronom-
inal clitics for indirect objects. The pivot of the se reciprocal/reXexive con-
struction can be either a direct object or an indirect object—but not an
oblique. It is, of course, unclear whether this is the only or best way of
categorizing and labelling the non-subject dependents in Brazilian Portu-
guese.
There are languages in which there appears to be no evidence for an
indirect object category. For example, in Kamaiurá, an indigenous language
of Brazil of the Tupı́-Gurananı́ family (Seki 2000), the grammar is organized
around two core grammatical relations, corresponding roughly to subject and
direct object in Brazilian Portuguese (see Section 2.1.3 on some precise
20 Introduction

diVerences in the conception of these). NPs with the semantic roles that are
characteristic of what are often considered to be indirect objects in other
languages, including recipient, beneWciary, and addressee, are marked with a
postposition, like obliques and unlike subjects and objects, as illustrated by
the following examples.

(1.17) a. ywyrapara o-me’e
kara’iwa upe

bow 3-give foreigner to
‘He gave the bow to the foreigner.’
b. tata atsã e-monyk je¼upe
Wre Dim 2Sg-light.Imper 1Sg¼to
‘Please light the Wre for me.’
c. tarekaja’a upi’a 0-mo’at ’ytsi
a ’arim
turtle egg 3-lay sand on/in
‘The turtle laid eggs in the sand.’
Although various constituent orders are possible, if there are three verbal
dependents, the subject and object, if either or both are expressed as NPs,
normally occur in preverbal position without a postposition and other NP
dependents of verbs normally occur in postpositional phrases in postverbal
position, as in all three of the examples in (1.17). Grammatical relations may
be indicated by head marking in the form of either preWxes or pronominal
clitics (the latter being indicated by the ‘¼’ separator, following Seki’s nota-
tion), which index (or agree with) either the subject or the object, the choice
being determined Wrst by which is higher on a hierarchy of person (1st person
> 2nd person > 3rd person) and, when they are of the same rank by person, by
a grammatical relation hierarchy on which subject outranks object. Postpos-
itions can have the same person-sensitive morphemes on them, which index
the NP that they go with. Thus, upe ‘to’ (or what Seki glosses as ‘dative’) in
(1.17b) has the pronominal clitic je on it, indicating that its object, the
recipient dependent of the verb, is 1st person singular. If the subject or object
dependent of a verb or the object of a postposition is indicated with head-
marking morphology and is pronominal, it need not be—and usually isn’t—
expressed as an independent pronoun. SigniWcantly, upe-marked NPs, which
might be analysed as indirect objects by virtue of the semantic roles that upe
typically indicates, do not appear to have any distinguishing syntactic prop-
erties that make them more object-like than obliques.
Consider, for example, the phenomenon of recipricolization in Kamaiurá.
As shown by (1.18a) the reciprocal preWx jo- appears on the verb when the
reciprocal pivot (the dependent that is not syntactically expressed) is the
direct object.
 Introduction 21

(1.18) a.a’e-wana o-jo-u’u

3Masc-Pl 3-Rec-bite
‘They bit each other.’
b. karamemã o-me’e
¼awa o-jo-upe
present 3-give¼Pl 3-Rec-to
‘They gave presents to each other.’
c. o-jepe’a¼awa o-jo-wi
3-separate¼Pl 3-Rec-from
‘They separated from each other.’
The same preWx is used when the reciprocal pivot is a recipient. However, as
shown by (1.18b), jo- appears on the postposition rather than the verb, just as
it does when the reciprocal pivot is any oblique dependent, as in (1.18c). There
are various further syntactic phenomena that distinguish direct objects from
obliques, including passivization, reXexivization (for which a preWx diVerent
from jo- is used), and relative clause formation. In every case the upe-marked
dependent behaves like an oblique. It is, of course, possible to maintain that
upe-marked NPs in Kamaiurá are indirect objects, as would be the case in a
standard Relational Grammar analysis. However, it appears that nothing in
the grammar of the language suggests that this syntactic function is operative
in a way that distinguishes it from a more general oblique function.
There are also languages in which there are alternative constructions for
expressing NPs with recipient-type semantic roles and diVerent potential
analyses of the grammatical relations in such constructions. In English, for
example, there are various potential ways of deWning an indirect object
category and a lack of consensus among theoreticians about how, if at all,
to deWne it. For example, one might claim that indirect objects are identiWed
by the preposition to, which is often used with NPs playing the same kinds of
semantic roles as Brazilian Portuguese indirect objects. Under this analysis,
the italicized NPs in the following examples would be indirect objects.
(1.19) a. I gave a present to my sister.
b. They transferred the title to the new owners.
c. The candidate promised very little to the voters.
d. It seems to me that there is a serious problem.
e. I’m going to drive Bill to the supermarket.
The problem with such a claim is that the indirect object relation would
play no role in the grammar other than to account (vacuously) for the
occurrence of the preposition to. One could just as easily say that the gram-
matical relation of indirect object is identiWed by the preposition from or for.
22 Introduction

The overall class of NPs marked with the preposition to (or occurring in a PP
with to) has no syntactic privileges or restrictions that do not extend to NPs
marked with from or for. If there is an indirect object category in English, it
appears not to be overtly marked. The prepositional system does not zero in
on such a category. Moreover, it is marked neither by pronominal case (the he
vs. him contrast only yields a distinction between nominative and non-
nominative), nor agreement (only subjects trigger very limited agreement
on verbs), nor even constituent order, as shown below. Thus, unlike in
Brazilian Portuguese, the usual markers of the primary syntactic functions
fail to announce an indirect object relation.
Another possibility is that indirect object names the subset of to-marked
NPs (or P-marked NPs), that can appear alternatively as the Wrst object (O1)
in the so-called double-object construction,11 illustrated by the following
examples.
(1.20) a. I gave [O1 my sister] [O2 a present].
b. The candidate promised [O1 the voters] [O2 very little].
c. *I’m going to drive [O1 the supermarket] [O2 Bill].
d. *They transferred [O1 the new owners] [O2 the title]
Under this alternative analysis, my sister and the voters in (1.19a) and (1.19c),
respectively, would be indirect objects, unlike the other italicized NPs in
(1.19). The new owners, for example, would not be an indirect object in
(1.19b), because it cannot be an O1 in the double-object construction, as
shown by the ungrammaticality of (1.20d). The problem with this analysis is
that indirect objects would only be identiWed by a single criterion: ability to be
alternatively expressed as an O1 in the double-object construction. There
appear to be no other syntactic phenomena for which a constraint needs to
be formulated in terms of this precise class of dependents. Moreover, there are
semantic and lexical constraints on the double-object construction (e.g.
Green 1974, Pinker 1989, Gropen et al. 1989, Goldberg 1995) which obviate
the need for an account in terms of an indirect object syntactic function. For
example, (1.20c) is bad because the participant expressed as the O1 has to be
conceived of as receiving something and supermarkets don’t receive people
that are taken to them in cars. On the other hand, although the potential O1 in
(1.20d) is readily conceived of as receiving something, transfer belongs to a
lexical class of Latinate verbs, which happens to dislike the double-object
construction, for whatever reason.
A third possible approach is that indirect object names a class of dependents of
verbs that can be deWned as including both O1s in the double-object construc-
tion and all to-marked (or P-marked) NPs that have the same kind of semantic
 Introduction 23

role as P-marked NPs that can be alternatively expressed as an O1.12 Under this
approach, all of the italicized NPs in the grammatical examples in (1.20) would
be indirect objects, as would the italicized NPs in (1.19a–d). Although attempts
at explicit deWnitions of the indirect object category in English are relatively rare,
textbook introductions to grammar and linguistic analysis alike often assume a
notion of indirect object something like this (e.g. Matthews 1997: 175, Barry 1998:
ch. 3, Carnie 2002: ch. 3). There is, however, no preposition, case aYx, or
agreement aYx that marks this putative class of NPs. There also appears to be
no syntactic phenomenon in which this class of NPs demonstrably plays a role.
Consider, for example, the passive construction. An O1 can be the subject of a
passive clause (My sister was given a present, The voters were promised very little);
but none of the to-marked NPs in (1.19) can (*My sister was given a present to, *I
was seemed to that there was a problem).
Based on distinctions that matter in the grammar of English, there seem to
be two main categories of objects (i.e. non-subject dependents that are not
marked by a preposition). A super-category of objects is identiWed by a
constituent-order constraint and by a constraint on the passive construction.
The constituent-order constraint says, in eVect, that the (primary) object
must be right-adjacent to the verb, without an intervening adverb, for ex-
ample,13 as illustrated by the following examples.

(1.21) a. I left [OBJECT my keys] [OBLIQUE on the table] [ADV just now].
b. *I left [ADV just now] [OBJECT my keys] [OBLIQUE on the table].
c. *I left [OBLIQUE on the table] [OBJECT my keys] [ADV just now].
(1.22) a. I sent [O1 my mother] [O2 some Xowers] [ADV just now].
b. *I sent [ADV just now] [O1 my mother] [O2 some Xowers].
c. *I sent [O2 some Xowers] [O1 my mother] [ADV just now].
The O1 in the double-object construction and the sole object in any other
construction are the dependents that are subject to the verb-adjacency con-
straint. These are also the dependents that can be alternatively expressed as
the subject in the passive construction (at least in most dialects of English):14

(1.23) a. My keys were left on the table.

b. *The table was left my keys on.
c. My mother was sent some Xowers.
d. *Some Xowers were sent my mother.
Given a choice between two non-subject dependents of a verb, the one that
can be expressed as the subject in the passive construction is the same one
that must appear right-adjacent to the verb otherwise.15
24 Introduction

Based on these facts, it would appear that there is simply one category of
primary object, which might simply be labelled ‘direct object’ and whose
members are the O1 in the double-object construction and the sole object in
any other construction. However, these two kinds of object are distinguished
by other grammatical phenomena. For example, the object in a single object
construction can be, and preferably is, placed at the rightmost edge of the
clause (overriding the verb-adjacency constraint) just in case it has a complex
internal structure, with an embedded relative clause, for example. This alter-
native constituent ordering, commonly known as heavy NP shift, is not
possible for the O1 in the double-object construction (e.g. Stowell 1981,
Larson 1988):
(1.24) a. I should have left [OBLIQUE at home] [HEAVY OBJECT all those things
that you’re always telling me to get rid of].
b. *I should have sent [O2 some Xowers] [HEAVY O1 all the people that
deserve to be remembered for what they’ve done].
Similarly, the sole object of a verb can alternatively be expressed as a subject in
the tough-movement construction, whereas an O1 cannot (Larson 1988, Sie-
wierska 1991: 97), as shown by the following examples, in which the (b)-
sentences in each pair illustrate tough movement.
(1.25) a. It will be easy to leave [OBJECT these things] at home.
b. These things will be easy to leave at home.
(1.26) a. It will be easy to send [O1 those people] [O2 some Xowers].
b. *Those people will be easy to send some Xowers.
Passivization and the verb-adjacency constraint operate in terms of what
seems to be a super-category of objects, including O1s and the sole object in
a clause with only one object. O1s are distinguished by certain other gram-
matical phenomena, in which they fail to behave like other objects and/or
other non-subject dependents.16
The question is how to draw the necessary distinctions. It is possible to use a
system of syntactic functions with three types of non-subjects: direct object,
indirect object, and oblique, as in Relational Grammar. The key idea of the
standard Relational Grammar analysis is that the double-object construction
results from what can be viewed as a revaluation of grammatical relations: the
indirect object is promoted to direct object, such that it is both an indirect
object (at an underlying or initial level of representation) and a direct object
(at a superWcial or Wnal level of representation). Ignoring various representa-
tional details, the analysis is as follows, where arrows represent the direction of
 Introduction 25

the revaluation and chômeur is the Wnal relation borne by a dependent whose
initial grammatical relation has been ‘taken over’ by another dependent.
direct object indirect object
(1.27) a. My son gavethe Xowers to his mother
indirect object ) direct object )
direct object chômeur
b. My son gave his mother the Xowers
The direct object properties of the O1 (right-adjacency to the verb, for
example) are accounted for by its status as a Wnal direct object. The ways in
which the O1 diVers from prototypical direct objects (inability to be the pivot
in the tough-movement construction, for example) are accounted for by its
initial indirect object status.
Another possibility, however, would be to say that there is simply a single
syntactic category, (direct) object, which encompasses both O1s and the sole
object in the single-object construction (as in Dryer 1986 and Siewierska 1991:
95). The special behaviour of O1s might be attributed to a semantic role
diVerence between them and other direct objects. For example, suppose that
the potential variety of semantic role distinctions is somehow collapsed into a
small set of super-categories, along the following lines, as in Lexicase Gram-
mar, discussed above, as well as various other theories.
(1.28) Agent, Recipient, Patient, Instrument, Locative
Generalized semantic roles
The idea would be that locative, for example, might include the more speciWc
roles of source, as with the from-marked NP in I removed this from the box ;
goal or destination, as with the to-marked NP in I drove my brother to the mall;
and location, as with the on-marked NP in My keys are sitting on the table.
Similarly, recipient might include, among others, not only the prototypical
role of recipient which gives the category its name, as with the O1 in I gave my
mother some Xowers; but also perceiver, as with the O1 in I showed my mother
the Xowers; and addressee, as with the O1 in I told my mother a lie. Under this
kind of analysis, two kinds of English direct objects might be distinguished:
those with a generalized recipient semantic role, that is O1s, and those with a
generalized patient semantic role, that is the sole object in the single object
construction. The grammatical relations of the non-subject dependents in the
sentences in (1.27) might be as follows, where uppercase letters are used for
semantic roles and 2nd object is the syntactic function of a patient that is
neither an object nor an oblique.
26 Introduction

patient & recipient &

object oblique
(1.29) a. My son gave the Xowers to his mother
recipient & patient &
object 2nd object
b. My son gave his mother the Xowers
The verb-adjacency constraint and the main constraint on passive subjects
could be sensitive to the object syntactic function. Non-patient or recipient
direct objects would be excluded from certain phenomena, such as tough
movement and heavy NP shift.
Lexical-Functional Grammar (Bresnan 1982a, 2000, Falk 2001 and see
Section 3.4.1) adopts a theory of grammatical relations in which the syntactic
function indirect object, as such, plays no role.17 The analysis of English
grammatical relations would be essentially as indicated in (1.29), various
representational details aside.18 In this kind of theory, all language-speciWc
and typological evidence for a category such as indirect object must be
accounted for in terms of (generalized) semantic roles or other semantic or
grammatical distinctions among oblique dependents. For example, the gram-
matical relation hierarchies implicated by typological restrictions on relative
clause formation strategies and verb agreement systems (see (1.12)) might be
alternatively formulated to include the composite grammatical relation
OblREC (i.e. the syntactic function oblique and the semantic role recipient),
in place of indirect object. By the same token, the grammatical relations of the
Brazilian Portuguese examples in (1.16) might be as follows.
patient & recipient &
object oblique
(1.30) a. Ela deu um presente para a mãe dela
3SgFem gave a present to the mother of.3SgFem
‘She gave a present to her mother.’
locative
& oblique
b. Ela mora com a mãe dela
3SgFem lives with the mother of.3SgFem
‘She lives with her mother.’
Under this analysis, accusative pronominal forms are used for the object
relation and dative forms for oblique-recipients. Given a grammatical relation
hierarchy such as Subject > Object > OblREC > OblINSTR > etc., one might
say that the pivot of the se reXexive/reciprocal construction cannot be lower
 Introduction 27

than OblREC on the hierarchy. OblREC dependents fail to behave like direct
objects in most respects because they simply have a syntactic function other
than object. Of course, the Lexical-Functional Grammar approach to indirect
objects extends straightforwardly to languages such as Kamaiurá, in which
recipients behave like obliques, since they are, quite simply, categorized as a
kind of oblique.
Returning to the question of indirect objects in English, there is at least one
other kind of analysis in which the indirect object relation can be engaged. In
recognition of the fact that the double-object construction evolved from an
Old English construction with a dative-marked NP preceding an accusative
object, one might simply analyse the O1 in the double-object construction as
the only kind of indirect object (e.g. Jespersen 1927, Herriman 1995).19 Like
other PPs, to-phrases would simply be obliques. The direct object category
would encompass the usual object in the single-object construction as well as
the second object (O2) in the double-object construction. Thus, assuming the
same generalized semantic roles, the grammatical relations of the non-subject
dependents of the sentences in (1.27) and (1.29) might be characterized as
follows.
patient & recipient &
direct object oblique
(1.31) a. My son gave the Xowers to his mother
recipient & patient &
indirect object direct object
b. My son gave his mother the Xowers
Under this analysis, a question arises as to why the indirect object rather than
the direct object in (1.31b) patterns like the direct object in (1.31a) with respect
to the verb-adjacency constraint and the constraint on the subject of passive
clauses. The answer is that direct and indirect objects are two diVerent kinds
of object, as in Relational Grammar, for example. In essence, they are mem-
bers of the class of dependents that are neither subjects nor obliques. Assum-
ing also a ranking of semantic roles, as in Lexical-Functional Grammar and
certain other theories (e.g. JackendoV 1972, Grimshaw 1990), according to
which recipient outranks patient, the constraints in question can be said to
hold for the object with the highest-ranking semantic role, that is the object
that is semantically most prominent.20
It should be clear that there are diVerent ways of conceptualizing and
labelling grammatical relations. If a theory utilizes the category of indirect
object, it may characterize its relationship to the dependents of a verb in
diVerent ways. If a theory does not utilize a syntactic category of indirect
28 Introduction

object, it may utilize a generalized semantic role in its place. Although these
diVerent characterizations of the linguistic facts have potential ramiWcations,
the extent to which the diVerences are ultimately just terminological remains
unclear. Still, the contrasting approaches do seem to entail some diVerent
claims and predictions. Implicit in the Relational Grammar approach, for
example, is the claim that there is no viable semantic characterization of the
(Wnal) indirect object relation. In principal, any dependent of a verb ought to
be able to ‘change’ its initial syntactic function to indirect object. Implicit
in the oblique-recipient approach, on the other hand, is the claim that
although the oblique-recipient category may diVer somewhat from language
to language, given that semantic roles can have diVerent precise conceptions,
it nevertheless ought to be restricted to a narrow range of precise semantic
roles, with some kind of similarity or shared feature, in a way that the subject
and object categories are not.

1.3.3 Oblique and other syntactic functions

Dependents of verbs that do not have one of the primary syntactic functions
are generally considered to be obliques,21 as in the discussion above. For
marking obliques, languages generally use adpositions or case markers that
can be related to semantic roles in a wide variety of ways, although other
marking devices are sometimes used, such as the relational nouns of Tzotzil
and other Mayan languages (Aissen 1987). At one end of the spectrum are
languages with a generalized oblique marker, such as Jarawara (Dixon 2000)
and Halkomelem (Gerdts 1988a).22 In Halkomelem, for example, the single
adposition ?@ works for all obliques and, thus, fails to distinguish semantic
roles in any way, as illustrated by the following examples.
(1.32) a. ni ném? ?@ L@ sLéni? kwu@ sw@
y?qe?
Aux go Obl Det woman Det man
‘The man went to the woman.’
b. ni ?ı́m@š L@ sLéni? ?@ kwu@ scš éšt
Aux walk Det woman Obl Det stick
‘The woman walked with a stick.’
c. ni c @n qš wál ?@ kwu@ n@-tél@
Aux 1Subj speak Obl Det 1Pos-money
‘I spoke about my money.’
At the other end of the spectrum are languages like English with numerous
prepositions for marking obliques or Hungarian with its sixteen or so oblique
case-marking categories. Although semantic roles are reXected in the marking
 Introduction 29

choices in such languages, the situation is complex. English, for example, uses
from for NPs that have the source role (I stole that from my brother, They came
from Illinois), to for NPs with the goal/destination semantic role (I ran to the
store, We drove the car to Chicago), and with for NPs with the instrument role
(He sliced the bread with that knife, I made this with my own hands). Hungar-
ian has ablative, allative, and instrumental morphological case-marking cat-
egories to make roughly the same semantic distinctions (Rounds 2001). As
with more generalized types of markers, such as nominative and accusative
adpositions or case aYxes, which can often be used for dependents with
virtually any semantic role, there is generally some kind of role conXation
with oblique markers as well, although the range tends to be more restricted.
The range for English to is relatively large, as it is used not only for goal/
destination but also for a variety of other roles (recipient, experiencer, per-
ceiver, addressee, etc.) which for some analysts constitute an indirect object
syntactic function rather than an oblique function (see Section 1.3.2). With has
various uses in addition to its instrumental use, including a comitative use
(I went to the movies with my sister), as well as what might be considered a
comitative-addressee use (e.g. Bill spoke with Tom about the problem desig-
nates an event with Bill speaking to Tom and Tom responding) and a
comitative-locative use (e.g. Sue lives with her mother designates a state of
aVairs with Sue and her mother living together at her mother’s place). In
some languages, such as the Greenlandic Eskimo-Aleut language Inuit
(Woodbury 1977, Bittner 1987), the instrumental case marker is used not
only for instruments but also agents in passive clauses and patients in anti-
passive clauses (see Section 4.2.1). Needless to say, the so-called instrumental
marker does not have a uniWed meaning across or within languages.
Not only do oblique morphological cases and adpositions conXate seman-
tic roles but they can also make very Wne-grained semantic distinctions
concerning details of an entity with a general semantic role. For example,
various subtle diVerences in the conception of the referent of an NP with
the goal role can be indicated by choice of preposition in a sentence such
as I loaded the hay onto/on/into the wagon. Similarly, in Hungarian, in
addition to a less speciWc ablative case category for the source role (corre-
sponding to English from), elative case morphology can be used for a source
conceived of as containing the moving participant (corresponding to English
out of ), and delative case morphology can be used for a source conceived of as
supporting the moving participant (corresponding to English oV of ). Al-
though there are regularities of various kinds, the relationship between
semantic roles and oblique markers within and across languages can be
complex.
30 Introduction

In any case, there are essentially two theoretical stances to the notion of an
oblique grammatical relation. In some theories, there is a formally recognized
distinct oblique syntactic function, with a variety of semantically determined
kinds that can be marked in diVerent ways across languages, as in Lexical-
Functional Grammar, for example. In other theories, oblique NPs simply lack
the deWning features of the core grammatical relations (e.g. by not occupying
one of the privileged positions in syntactic structure) and are marked in a way
that may or may not be semantically determined, as in Transformational
Grammar (see Chapter 5) and Role and Reference Grammar (see Chapter 4).
Under either approach, (at least most) adpositionally-marked dependents in
both languages like English and languages like Halkomelem are treated as
belonging to a diVerent category than subjects and objects. The marking rule
for Halkomelem is straightforward: (in a theory that utilizes the oblique
relation) any kind of oblique is marked with the preposition ?@ ; or (in a
theory without an oblique relation) any kind of NP that is not a subject or
object is marked with the preposition ?@. For English, obliques are marked,
following a complex schema, according to the kind of oblique relation borne;
or any kind of NP that is not a subject or object is marked according to its
semantic role, following a complex schema.
There is a bigger theoretical divide concerning the question of whether
Wner distinctions than oblique vs. core (subject/object) functions should be
drawn, and, if so, how. One issue has to do with a distinction between oblique
arguments and adjuncts, that is verb phrase modiWers such as in the kitchen in
I ate breakfast in the kitchen, which speciWes a setting for the event designated
by eat rather than expressing one of the deWning participants in the type of
event speciWed by the verb. Another has to do with the analysis of construc-
tions in which the default subject or object has been displaced by some other
dependent of the verb or ‘demoted’. For example, if the O1 in a double-object
construction of the kind exempliWed by give [O1 the boy][O2 a bike] is the
direct object, what is the relation of the O2 (or demotee from direct object)?
Similarly, what is the relation of the agent (or demotee from subject) in the
passive construction exempliWed by The dog was fed by the boy ?
Generally, some kind of distinction is drawn between adjuncts and what are
often called ‘arguments’ of verbs, that is NPs or other phrases that express
participant roles speciWed in the meanings of verbs. Relational Grammar,
however, does not formally distinguish these, since all co-constituents of a
clause are assumed to bear one or more relations to the clause rather than to
verbs per se. Although the conceptual distinction between argument and
adjunct is relatively clear, the empirical basis for it is problematic, even with
respect to well-studied languages such as English. As noted in Whaley (1993),
 Introduction 31

applying the available criteria to English yields more of a continuum than a

clear-cut distinction between oblique arguments and adjuncts. From a cross-
linguistic perspective, this murkiness is exacerbated by the fact that the same
morphological marking is often used across the full range of constituent
types, as evidenced, for example, by the Finnish use of accusative and inessive
cases for both core and oblique arguments and what are presumably temporal
adjuncts (see (1.10)) and the similar multiple use in English of such preposi-
tions as in, which can mark a goal participant in an event designated by a verb
(put the toys in the box), a locational setting for an event (eat breakfast in
the kitchen), or even a temporal setting (go to the store in the morning).
Consequently, not only is the issue of where and if to draw a line between
adjuncts and oblique arguments a diYcult one, but so is the question of how
to categorize such demotees as the by-marked agent in the English passive
construction and its analogues in other languages, since the murkiness of the
adjunct vs. argument distinction can extend to them as well. Indeed, theories
take various stances concerning by-marked agents: from an adjunct designa-
tion in Lexical-Functional Grammar and Role and Reference Grammar, for
example, to the Relational Grammar analysis, in which demotees of all kinds
belong to an altogether distinct category, called ‘chômeur’. Thus, modulo
certain cross-linguistic diVerences in the primary or core grammatical rela-
tions (considered in Section 2.1) and certain theory-speciWc implementation
details (concerning the indirect object category, for example), the distinction
between core and oblique relations is relatively uncontroversial. However,
there is more uncertainty surrounding such potentially distinct categories as
adjunct and demotee.
As an example of the adjunct vs. oblique argument problem, consider the
following diVerent kinds of with-marked NPs in English.
(1.33) a. Sue lives with her mother.
b. Jake spoke with us about that.
c. I’m going to Wll the glasses with wine.
d. You should slice the bread with the serrated knife.
e. I painted the house with my sister.
If the basic idea is that arguments express deWning elements of the process or
state designated by a verb, then it seems clear that the bread in You should slice
the bread in the kitchen is an argument and in the kitchen is an adjunct. The
meaning of slice is presumably something like ‘x does something to y and
because of this y becomes thin Xat pieces.’ Because the bread in slice the bread
speciWes the entity indicated by the variable y in the hypothesized meaning of
slice, it is an expression of an argument of the verb. Since adding ‘in someplace
32 Introduction

(z)’ to the hypothesized deWnition would not serve to clarify the meaning of
the verb in any way or to distinguish slice from other verbs, in the kitchen is
presumably an adjunct. But what about the with phrases in (1.33c–d)? The
diYculty is that the meanings of words are not available for direct inspection.
Does slice really mean ‘x uses something sharp (z) to do something to y and
because of this y becomes thin Xat pieces’?
One kind of rationale for claiming that wine in (1.33c) is an argument goes as
follows. First, this phrase has a theme or ‘moving participant’ meaning (see
Section 1.3.4) as opposed to a standard instrument meaning. This is revealed
by the possibility of adding an instrumental with phrase to (1.33d) (I’m going to Wll
the glasses with wine with a funnel ). Ordinarily, two with phrases with a simple
instrumental meaning cannot be added to the same clause (*I’m going to Wll the
glasses with a funnel with a pitcher), althoughwith phrases with diVerent meanings
can (I’m going to Wll the glasses with wine with my brother). Hence, the with phrase
in (1.33c) expresses the semantic role of theme (the entity conceived of as under-
going motion) or, perhaps, instrumental theme, since wine is both used to Wll and
undergoes movement to a speciWed goal. Now, with phrases can have this theme
meaning only in the context of verbs in the same semantic class as Wll, that is verbs
of goal-oriented caused motion (e.g. load, pack, spray, etc.). As the theme inter-
pretation does not occur in other contexts, the with phrase is expressing a
semantic argument of the verb. It is, therefore, not an adjunct.
The same kind of reasoning applies to the locative and addressee with phrases
in (1.33a–b). For example, a sentence such as Tom spoke with the children about
that with his brother can mean ‘Tom and his brother spoke to the children about
that’ but not ‘Tom and the children spoke to his brother about that’. The
addressee role, which goes to the Wrst with phrase, seems to come not from
with per se but from the verb. An addressee interpretation occurs only with a
with phrase that accompanies a verb of communication (speak with, talk with,
argue with, chat with, etc.). It is less clear whether the same reasoning can be
extended to cases like (1.33d). One could say that the instrumental meaning of
with is contingent upon co-occurrence with a verb whose meaning entails the
use of an instrument. Slice, for example, designates a process that ordinarily
requires the use of a sharp implement. However, instrumental with phrases can
be added to clauses with a wide array of verb types, not all of which designate
inherently instrumental processes (e.g. I Wlled the glasses with wine with a funnel,
She spoke to the crowd with a bullhorn, He reads with a magnifying glass).
Potential omissibility is another factor that can have a bearing on the
adjunct issue. For example, the fact that the bread is not omissible in (1.33d)
(*You should slice in the kitchen) provides a clue that it is an argument.
Prototypical adjuncts are rarely, if ever, required (e.g. You should slice the
 Introduction 33

bread (in the kitchen)). Although not being able to be omitted provides
evidence for argument status, omissibility does not provide evidence against
argument status, since patient/direct object dependents of a verb that are
clearly arguments based on other criteria are often omissible, as in the case of
I ate (a donut) in the kitchen, We drove (the car) to Chicago, or I can see (things)
with these glasses. Moreover, prototypical PP arguments are also often omis-
sible (That looks funny (to me), I stole some money (from him), He lied (to
me)), although not always (e.g. in the refrigerator is not omissible in He put
the beer *(in the refrigerator)). As for the diVerent types of with-marked
dependents, the omissibility criterion is only applicable to the comitative-
locative with phrase, which shows evidence of being an argument:
(1.34) a. Sue lives *(with her mother).
b. Jake spoke (with us) about that.
c. I’m going to Wll these glasses (with wine).
d. You should slice the bread (with the serrated knife).
e. I painted the house (with my sister).
Another kind of syntactic evidence come from the pseudocleft construction
(Vestergaard 1977, Radford 1988: Ch. 5, Whaley 1993: Ch. 3), illustrated by the
following examples.
(1.35) a. What I’m going to do in the kitchen is slice the bread.
b. What I’m going to do in the kitchen is dance.
c. *What I’m going to do in the refrigerator is put the beer.
d. *What I’m going to do the bread (in the kitchen) is slice.
The generalization is that the focused verb phrase in the pseudocleft con-
struction (i.e. the phrase appearing sentence-Wnally following the copula)
must include the verb and any arguments that it may have. In (1.35a–b),
slice the bread and dance can be in the sentence-Wnal focus position because
neither phrase is missing any of the arguments of its verb. Examples (1.35c–d),
on the other hand, are ungrammatical because put the beer is focused without
its goal argument and slice is focused without its direct object argument. The
pseudocleft test conWrms the argument status of the comitative-locative with
phrase and diVerentiates the remaining types as follows.
(1.36) a. *What I’m going to do with my mother is live.
b. *What Jake is going to do with us (about that) is speak.
c. What I’m going to do with the wine is Wll these glasses.
d. What I’m going to do with the serrated knife is slice the bread.
e. What I’m going to do with my sister is paint the house.
34 Introduction

Since, an ordinary comitative with phrase need not accompany a verb and any
of its arguments in the pseudocleft focus position, as in (1.36e) as well as in the
case of verbs without any non-subject arguments (What I’m going to do with
my sister is work/leave/play), such phrases appear to be adjuncts. Also falling
on the adjunct side of the divide by this criterion are the theme-instrumental
with phrase accompanying transitive Wll and the ordinary instrumental with
phrase accompanying transitive slice, as shown by the grammaticality of
(1.36c–d). The comitative-addressee with phrase accompanying speak, on
the other hand, behaves like an argument (1.36b).
The placement of emphatic reXexives provides another possible means of
distinguishing adjuncts and arguments (Radford 1988: ch. 5, Whaley 1993: ch.
3). An emphatic reXexive is an intonationally focused reXexive Xself form that
does not itself express an argument of a verb and is necessarily construed with
the subject, as in the following examples.
(1.37) a. Tom’s gonna slice the bread in the kitchen himself.
b. Tom’s gonna slice the bread himself in the kitchen.
c. Tom’s gonna put the beer in the refrigerator himself.
d. * Tom’s gonna put the beer himself in the refrigerator.
As the contrast between (1.37b) and (1.37d) shows, it is possible to place an
emphatic reXexive in a postverbal position before a prototypical PP adjunct
but not before a prototypical PP argument. There appears to be a constraint
to the eVect that an emphatic reXexive cannot immediately follow a verb,
independently of the adjunct/complement status of the following phrase (e.g.
*Tom reads/studies/works/sleeps himself in the library). For this reason, the
emphatic reXexive criterion appears to be inapplicable to the live with X
construction. It works as follows with the other with phrases under consid-
eration here:
(1.38) a. *Jack’s gonna speak about that himself with us.
b. ?Jack’s gonna Wll the glasses himself with wine.
c. Jack’s gonna slice the bread himself with the serrated knife.
d. Jack’s gonna paint the house himself with a friend.
The key diVerence between the pseudocleft and emphatic reXexive criteria is
that instrumental themes (with wine in Wll the glass with wine) are only
marginally adjunct-like with respect to the latter.
In short, these diVerent kinds of with-marked NPs show varying degrees of
argument/adjunct behaviour. In the case of live with my mother, all the
available evidence suggests that with my mother is an argument. In the case
of paint the house with my sister, with my sister displays relatively clear adjunct
 Introduction 35

behaviour. With wine in Wll the glasses with wine falls somewhere in the
middle. It clearly satisWes only one adjunct criterion (omissibility in the VP
pseudocleft construction). Phrases such as with a serrated knife in slice the
bread with a serrated knife are often considered to be arguments, presumably
because of the theoretical prominence of a semantic role category labelled
‘instrument’ (owing, for example, to Fillmore 1968), the relative importance
of an instrumental case-marking category in languages with robust depend-
ent-marking morphology, and the fact that instrumentally-marked depend-
ents and/or subject and object dependents with an instrumental semantic role
can be arguments sometimes.23 However, at least in English, it seems hard to
justify an argument classiWcation for common instrumental with phrases.
Tagalog has what has been called an adjunct-fronting construction
(Schachter and Otanes 1972, Kroeger 1993), which seems to announce which
constituents of a clause are adjuncts. Tagalog is a verb-initial language in
which, outside of special constructions, constituents follow the verb. In one
such special construction adverbial phrases and other typical adjuncts are
placed in sentence-initial position, followed by any pronominal dependents of
the verb, as illustrated by the following examples, adapted from Kroeger (1993:
ch. 2).
(1.39) a. [dahil sa¼iyo] ako nahuli
because Loc¼you 1SgNom late
‘Because of you, I was late.’
b. *[ang¼libro¼ng ito] ko binili para kay¼Pedro
Nom¼book¼Lnk this 1SgGen bought for Loc¼Pedro
‘This book, was bought for Pedro by me.’
c. *[ng¼nanay] siya pinalo
Gen¼mother 3SgNom spanked
‘By mother, he was spanked.’
d. [para kay¼Pedro] ko binili ang¼libro¼ng ito
for Loc¼Pedro 1SgGen bought Nom¼book¼Lnk this
‘For Pedro, this book was bought by me.’
e. [sa¼pamamagatin ng¼sandok] siya kumuha ng¼sabaw
Loc¼use Gen¼ladle 3SgNom take Gen¼soup
‘With this ladle, I took some soup.’
f. *[ng¼papel na iyon] niya binalutan ang¼libro
Gen¼paper Lnk that 3SgGen wrapped Nom¼book
‘With that paper, the book was wrapped by her.’
A prototypical adjunct, such as a ‘because’ phrase, can undergo adjunct
fronting (1.39a), whereas a subject (marked with nominative case) for
36 Introduction

example, cannot (1.39b).24 Indeed, although the agent and patient can either
be the subject (if marked nominative) or not (if marked genitive), they can
never undergo adjunct fronting (1.39c). The distinction between NPs marked
with a case such as genitive and nominative and those contained in prepos-
itional phrases seems to correlate with the possibility of adjunct fronting. A
beneWciary dependent marked with the preposition para, for example, can
undergo adjunct fronting (1.39d), as can an instrumental phrase headed by
the complex preposition sa¼pamamagatin (1.39e). A genitive-marked instru-
ment(-theme), however, cannot (1.39f). Although both the use of a prepos-
ition and the possibility of adjunct fronting appear to be plausible adjunct
criteria, the situation is complicated by the fact that recipients, which are
marked with the all-purpose locative/dative case marker sa, can undergo
adjunct fronting:
(1.40) [sa¼akin] nila ibinigay ang¼premyo
Loc¼me 3PlGen given Nom¼prize
‘To me, the prize was given by them.’
Although Tagalog may indeed draw a line between arguments and adjuncts
with this fronting construction, the line is drawn quite diVerently than it
would be in English, in which the recipient dependent of a verb such as give
would quite clearly be an argument by all criteria. Another possibility is that
genitive and nominative are the cases for NPs with a core syntactic function
(object and subject respectively), whereas sa and prepositions mark oblique
dependents (independently of an adjunct/argument distinction), in which
case adjunct fronting is really oblique fronting. An interesting implication
of this analysis is that the functional equivalent of the English passive con-
struction, as illustrated by (1.39c), has the non-subject agent as a core
dependent, rather than an oblique or adjunct. This issue is considered further
in Section 2.2.1.
Returning to the question of ‘demotees’, it seems pretty clear that the O2 in
the double-object construction in English is an argument, since it cannot be
omitted (I gave my mother *(some Xowers)), must accompany the verb and the
O1 in the pseudocleft construction (What I did was give my mother some
Xowers vs. *What I did some Xowers was give my mother), cannot be preceded
by an emphatic reXexive (*Tom sent Sue himself the Xowers vs. Tom sent Sue
the Xowers himself), and only occurs with a limited class of verbs designating
events of transferring (of varying degrees of abstractness) in which its referent
plays the role of thing transferred. Still, it remains unclear whether it should
be categorized as a direct object, as in some traditional analyses (see Section
1.3.2), an oblique (in eVect), as in certain Transformational Grammar analyses
 Introduction 37

in which it receives an ‘inherent’ rather than a ‘structural’ case (Larson 1988,

Baker 1988b), or a non-oblique of a special category, such as the 2nd object of
Lexical-Functional Grammar or the chômeur of Relational Grammar.
It is clear that in many languages with similar constructions, the analogue
of the O2 is overtly marked like an oblique. Halkomelem, for example, uses
the generalized oblique preposition ?@ in the corresponding construction in
which the recipient has the direct object function:
(1.41) ni ?ám-@s-t-@s kwu@ sqw@m@
y? ?@ kwu@ suš ám?
Aux give-Recip-Tr-3 Det dog Obl Det bone
‘He gave the dog the bone.’
However, given the complexity of the relationship between morphological
categories and grammatical relations within and across languages, there is no
reason to assume that ?@ in Halkomelem, for example, is a reliable indicator of
an oblique syntactic function. It could be said to work for all constituents in a
clause that are not in the general subject/object category. Indeed, Gerdts (1988a)
shows that although this kind of constituent generally behaves like an oblique
with respect to syntactic phenomena, it is distinguished from other obliques in
that its verb appears with a distinct nominalizing preWx when it is fronted in a
question or focus construction. She suggests an analysis according to which it
may either be a chômeur or an indirect object. The indirect object analysis is
made possible by the fact that dependents with a generalized recipient semantic
role necessarily bear one of the two primary grammatical relations, as in (1.41).
Whether the ‘demoted’ agent of a passive clause in English is an oblique
argument, an adjunct, or a member of some special category such as chômeur
is also a diYcult question. Like an argument, it expresses what is clearly a
semantic role of the verb it goes with. The omissibility criterion fails to
establish such by phrases as arguments, however, since they are generally
omissible. The pseudocleft test turns out not to be illuminating since VP
pseudoclefting doesn’t work well with passive clauses at all (*What the car was
was sold by Max/*What the car was by Max was sold ).25 An emphatic reXexive
seems only marginally possible precedeing a passive by phrase (?Max was sold
a car himself by Tom). However, this may simply be due to the marginality of
a postverbal emphatic reXexive with a patient subject (?Max was sold a car
himself). Again, it is unclear whether the choice between possible classiWca-
tions can be established on solid empirical grounds. From a cross-linguistic
perspective, the question is even more complex, since (as discussed in Section
2.2.1) languages mark the grammatical relation of passive agents in a variety of
ways and they behave syntactically like subject arguments to varying degrees.
Even in English, the passive agent behaves like an agent subject in that it can
38 Introduction

control a rationale clause (e.g. Roeper 1987, Roberts 1987: ch. 3). That is to say,
just as the interpretation of the implicit subject of the bracketed rationale
clause is controlled by the subject agent in (1.42a), it is controlled by the
passive agent, whether this is overtly expressed or not, in (1.42b).
(1.42) a. The owner torched this house [Ø to collect on the insurance].
b. This house was torched (by the owner) [Ø to collect on the
insurance].

1.3.4 Semantic roles

As should be clear even from the few contrasting analyses of basic linguistic
phenomena considered above, theories vary even more dramatically in their
labelling and conceptions of semantic roles than in their treatment of syntac-
tic functions. At one extreme of the spectrum of approaches, semantic roles
are verb-speciWc and play no role in the grammar other than to relate the
syntactic function of NPs to the correct semantic interpretation of a sentence
relative to individual verbs, as in Head-Driven Phrase Structure Grammar
(Pollard and Sag 1994). That is to say, in this theory the verb love has a ‘lover’
semantic role, a ‘lovee’ semantic role, and a lexical valence specifying that the
lover role is that of an NP with the subject function; the verb hate has a ‘hater’
semantic role, a ‘hatee’ semantic role, and a lexical valence specifying that the
hater role is that of an NP with the subject function; and so forth.
At the other extreme of the spectrum is Dowty’s (1991) theory of semantic
roles, according to which only two (generalized) roles are recognized: proto-
agent and proto-patient. Two roles suYce because they only function to
determine which dependent of a transitive verb is the subject and which is
the object. No other function is attributed to semantic roles as such. Each of
these two generalized roles is characterized by a list of several properties, such
as volitional involvement in the event, sentience, independent existence, etc.
for the proto-agent. A dependent of a verb can be more or less agent-like and
more or less patient-like, according to which properties it has from each set.
Among two dependents, the one with the most proto-agent properties is
realized as the subject.
How semantic roles are deWned also varies widely across theories. Dowty’s
approach in terms of semantic entailments stands somewhere in between
theories such as Lexicase Grammar or Fillmore’s (1968) Case Grammar, in
which semantic roles are primitive elements that are only roughly character-
ized with prose deWnitions, and theories for which terms such as agent and
patient are merely mnemonic labels for elements in a grammar of verb
meanings, or a lexical-conceptual semantics, as in Localist Case Grammar
 Introduction 39

(Anderson 1971, Delancey 1991). The key idea of the so-called localist ap-
proach to semantic roles, which can be traced back to Gruber (1965) and is
reWned and developed in various ways in much subsequent work (especially
Talmy 2000), is that the most basic types of events and states involve location
and/or motion and all other types of events and states are metaphorical
extensions of these. For Delancey the grammar of the semantics of states
and events, in terms of which the core semantic roles of agent, theme, and
location are deWned, is characterized as follows:26
(1.43) a. theme at location (state)
b. theme go-to location (change of state event)
c. agent cause theme go-to location (caused change of state event)
Death has the role of location in They put the murderer to death, just as the
mall does in I drove my brother to the mall, by virtue of the very general
metaphorical conception of states as locations, manifested not only in the
routine use of locative prepositions for both states and locations, as in I’m in
shock vs. I’m in my room, but also in such expressions as fall asleep and go
crazy. The conceptual schema in (1.43c) is equally applicable to cognitive,
emotional, and perceptual processes. For example, the verb bug, as in The
music is bugging me, designates an event in which the subject (agent) is
conceived of as causing the direct object (theme) to ‘go’ to an emotional
state of displeasure. By general rule, if the agent is expressed it is encoded as
subject and the theme is encoded as direct object. As in Lexicase Grammar
(see Section 1.3.1), the construal of events is what dictates the roles of the
participants designated by NPs. Thus, even though one might say that the
wagon is the location in both (1.44a) and (1.44b), since both sentences
describe events that from an objective perspective only have hay doing any
moving, the generalization that themes are encoded as direct objects remains
valid because (1.44b) encodes a construal of the event in which the wagon
(metaphorically) moves into the state of being Wlled.
(1.44) a. They loaded hay onto the wagon.
b. They loaded the wagon (with hay).
In (1.44b) with hay is analysed as an adjunct phrase (see Section 1.3.3) that
expresses only a peripheral aspect of the event, rather than one of the core
roles.
DiVerent approaches to lexical semantic structure and diVerent degrees of
commitment to an elaborated theory of verb meaning result in considerable
theoretical variation in terms of number and kind of semantic roles and
terminological conventions. Generally, roles such as agent and patient (or
40 Introduction

theme) are recognized in addition to at least several others (experiencer, goal,

source, instrument, beneWciary, etc.). It is generally taken to be obvious that
there is some kind of default or natural direct object semantic role. In
Fillmore (1968) this role was labelled ‘objective’ and deWned as the role
assigned to an NP that does not have one of the other roles. The label is
indicative of the preferred alignment of this role with direct object/accusative
case. However, since roles for Fillmore remain constant across paraphrases
and, more generally, across the alternative syntactic frames that a verb can
appear in, the ball has the objective role in He rolled the ball down the hill, The
ball was rolled down the hill by him, and The ball rolled down the hill.
Although what is meant by the term patient varies considerably, it is often
used in eVect as a synonym of objective, as in various analyses discussed above.
Although the term goal is used most often for some kind of locative role (e.g.
that which something moves to, perhaps encompassing the recipient role), it
is sometimes also used with essentially the same meaning as patient or
objective (e.g. Dik 1997), presumably owing to a conception of the role as
the ‘target’ of the agent’s action, for which reason target is also sometimes
used (e.g. Dixon 1994). Probably the most popular term for this role, however,
is theme. Transformational grammarians, in particular, have generally been
drawn to the terminological conventions introduced in Gruber (1965) and
reWned and developed in various ways in the work of JackendoV (1972, 1987,
1990). However, since the underlying theories of lexical-conceptual semantics
that explicate and justify the terminology are not a standard component of
Transformational Grammar, there is little terminological uniformity.
JackendoV (1987, 1990) proposes a grammar of conceptual semantics in
which roles of theme and patient are diVerentiated and can sometimes be
associated with diVerent NPs in a clause, although they generally go together.
The basic idea is that there are two main aspects of the meaning of a verb that
can be separated. One aspect encodes an elaborate version of the kind of
localist grammar of events and states expressed in (1.43); the other encodes
information about the main action, that is who or what is conceived of as
acting and who or what is acted on. Avoiding JackendoV’s complex notational
system, Delancey’s simpler system can be augmented as follows to show the
key idea, for one kind of caused-motion schema.
(1.45) actorx act-on patienty
agentx cause themey go-to goal
caused change of location event, with patient ¼ theme
In a sentence such as They loaded hay onto the wagon, for which (1.45) is an
appropriate conceptual structure, hay would be both the patient, by virtue of
 Introduction 41

being the principal target of the action, and the theme, by virtue of being the
moving entity in the event. In the case of They loaded the wagon with hay, the
wagon still has the role of goal; but it also has the role of patient, as load has an
alternative conceptual structure that diVers from (1.45) in that goal and
patient, rather than theme and patient, are co-indexed. The generalization
about the relationship between semantic roles and syntactic functions is that
the patient of a verb in the active voice is realized as the direct object.
Owing in part to an inXuential principle known as the Theta Criterion
(Chomsky 1981), which says in eVect that each NP bears one and only one
semantic role (or ‘theta-role’), and in part to the minimal burden of semantic
roles in the theory, practitioners of Transformational Grammar generally
don’t embrace JackendoV’s semantic distinction between patient and theme,
however much sense it may make. They operate with a small set of semantic
roles, which are often not explicitly deWned, and use the term theme (or
patient or theme/patient) in much the same way that Fillmore uses objective
or Delancey uses theme. Chomsky makes little attempt in his own work to
explicate semantic role terms or to articulate or endorse a grammar of verb
meaning to which semantic roles might be related. Textbook introductions to
the theory generally just give lists of semantic roles of varying length with
prose deWnitions, the content of which varies somewhat from author to
author. Radford (1997: 326), for example, deWnes theme/patient as the role
of an ‘entity undergoing the eVect of some action’, thus designating something
more like JackendoV’s patient role, while leaving the deWnition vague enough
as to cover his theme role as well—reXective of common practice.
The same kind of terminological and deWnitional issues arise with most
other semantic roles. However, the main diVerences between theories of
grammar that make use of semantic roles have to do with whether or not
they are precisely characterized in terms of an overall theory of verb meaning
and whether or not they are considered to have a direct impact on grammat-
ical phenomena. In theories with a well-articulated characterization of verb
meanings, semantic roles are more likely to be used, at least to some extent, in
place of syntactic functions. Theories that place most of the explanatory
burden for grammatical phenomena on primitive syntactic functions or
positions in phrase structure are more likely to work with an undeWned or
imprecise list of semantic roles.
In this book, the terminological conventions of theories or analyses under
consideration are used and explicated as necessary. For general descriptive
purposes where precision is not critical, the terms agent and patient are used
to designate generalized semantic roles, more or less in the sense of Dowty’s
proto-agent and proto-patient or Delancey’s agent and theme.
42 Introduction

1.4 Theories of grammar

A main goal of this book is to provide an overview of the treatments of
grammatical relations in diVerent modern theories of grammar and to bring
out similarities and diVerences and strengths and weaknesses by showing how
they have dealt with or might deal with a range of the interesting and
challenging phenomena involving grammatical relations in diVerent lan-
guages. Although numerous theories are commented on along the way, the
focus is necessarily limited to a subset of the available theories. The theories
chosen for consideration reXect primarily a concern for covering a wide
spectrum of approaches with representative examples of the main types of
theories. Each theory has been suYciently articulated as to have a clear stance
on the typological issues concerning grammatical relations and has had
suYcient appeal to theoreticians and descriptive linguists as to have yielded
analyses of a broad range of the phenomena across the languages of the world
that bear on the key matters surrounding grammatical relations.
Role and Reference Grammar (Foley and Van Valin 1984, Van Valin and La
Polla 1997) broadly represents what has come to be known as the functional
approach to grammatical analysis, which is characterized by a concern for
meaning and the functions that language performs and how these inXuence
and are manifested in grammatical organization. It deWnes grammatical
relations primarily in terms of a theory of verb meanings, that is a conceptual
semantics of verbs. Semantic roles are therefore brought front and centre in its
characterizations of grammatical phenomena. Although it can deWne depend-
ent categories corresponding more or less to the primary traditional syntactic
functions (subject and object), these are not primitive notions of the theory
and are not held to play a role in all languages. An in-depth look at Role and
Reference Grammar and how it analyses grammatical-relation phenomena in
diVerent kinds of languages is supplemented by shorter overviews of two
other theories with a functional orientation: Functional Grammar (Dik 1978,
1980, 1997), which articulates a semantic-role based account of grammatical-
relation phenomena in a somewhat diVerent way than Role and Reference
Grammar, and Cognitive Grammar (Langacker 1991, 1999), which subsumes
grammatical structure in an all-encompassing theory of conceptual structure,
yielding a cognitive characterization of syntactic functions and, therefore, no
signiWcant distinction between syntactic functions and semantic roles.
The other theories to be examined are probably best portrayed as falling
somewhere on the other side of a presumed divide between the functional
approach to grammar and the formal approach, which emphasizes the role
 Introduction 43

of linguistic form and syntactic structure in the organization of language and

makes minimal appeal to meaning and the functions of language as a source
of explanation for grammatical phenomena. Although these theories take
varying approaches to the labelling, identiWcation, and deWnition of semantic
roles, none includes as an integral component a fully articulated conceptual
semantics of verbs. They focus, instead, on syntactic functions (in some
form), which are seen as playing a central role in the analysis of sentence
structure in all languages. How these syntactic functions are characterized
constitutes the key diVerence. In Transformational Grammar (Chomsky 1965,
1981, 1995) the constituents of a sentence are portrayed as being systematically
arranged in a hierarchical structure expressible as a tree, which encodes lexical
class membership (noun, verb, adjective, etc.), constituency (i.e. information
about which words are parts of which phrases), and linear precedence.
Syntactic functions are deWned, in complex ways, in terms of speciWc posi-
tions in trees or relationships with certain positions in trees. NPs can change
their positions in trees (or undergo transformations), which can give rise to
changes in syntactic function. Relational Grammar (Perlmutter 1983, Perl-
mutter and Rosen 1984, Blake 1990) rejects the tree formalism for sentence
structure, in favour of a representation that basically only calls attention to
constituency and grammatical relations. However, the Transformational
Grammar idea that NPs can bear more than one syntactic function (albeit
due to ‘promotion’ or ‘demotion’ rather than ‘movement’) is an important
component of the theory. Relational Grammar makes the most elaborate
attempt of any theory to account for grammatical phenomena and linguistic
typology primarily in terms of the traditional notions of subject, object, and
indirect object. The chapter on Relational Grammar is supplemented by a
shorter overview of Lexical-Functional Grammar (Bresnan 1982a, 2000, Falk
2001), which is the other main theory in which syntactic functions such as
subject and object are conceived of as primitive, central, and universal
elements in the grammatical organization of languages.
 2

Grammatical relations
across languages

There are essentially three broad classes of fact about languages for which a
theory of grammatical relations needs to be responsible. First, it appears that,
at least in terms of their head-marking and dependent-marking systems for
indicating grammatical relations, most languages draw a fundamental dis-
tinction between two basic roles, corresponding roughly to a category with
agent as prototype and a category with patient as prototype. However, the
diVerent ways in which these categories are deWned constitute one of the main
syntactic typologies of languages. Secondly, in addition to active vs. passive
voice alternations (see Section 1.2) and alternations between an object þ
oblique construction and a corresponding double-object construction (see
Section 1.3.2), there are various kinds of ‘relation-changing’ and voice con-
structions in languages that have important implications for an understand-
ing of grammatical-relation systems. Thirdly, languages frequently have verb
classes or constructions for which the dependents manifest some but not all of
the properties of canonical subjects or objects (as, for example, with the
dative-subject construction in Icelandic discussed in Section 1.2).

2.1 Grammatical relations and major typological parameters

For typological purposes, it has become customary to recognize three main
kinds of syntactic functions, in terms of a distinction between transitive and
intransitive clauses. In a transitive clause in English, for example, there is a
more or less agent-like subject and a more or less patient-like direct object, as
in The boy touched the branches. In an intransitive clause there is a subject and
either no non-subject dependent of the verb (The boy slept) or an oblique (or
at least a non-direct object) dependent of the verb (The boy ran to the store).
Given this distinction, the three categories are subject of a transitive clause, or
A (for agent-like dependent), subject of an intransitive clause, or S, and direct
object, or O (for object; sometimes labelled P for patient-like dependent). In
terms of their head-marking and dependent-marking systems for indicating
 Grammatical relations across languages 45

grammatical relations, rarely do languages either systematically distinguish all

three of these categories or distinguish none.1 Rather, these three categories
are generally made to Wt into a two-category marking system.
As should be clear from the discussion in Chapter 1 of basic grammatical-
relation distinctions in English, Latin, Brazilian Portuguese, and Sanskrit,
these languages treat the O category (direct object or accusative) as one and
collapse A and S into the other main category (subject or nominative).
Languages of this type are generally known as accusative languages. erga-
tive languages (Dixon 1979, 1994), exempliWed by Mayan languages, indigen-
ous Australian languages of the Pama-Nyungan group, Eskimo-Aleut
languages, Caucasian languages, and Basque, among others, treat the A
category (commonly labelled ‘ergative’) as one and collapse the S and O
categories into the other main category (commonly labelled ‘absolutive’). In
split-intransitive or active languages (Sapir 1917, Merlan 1985, Van Valin
1990, Mithun 1991), which include primarily certain indigenous languages of
the Americas but also such languages as Acehnese (Durie 1985, 1988) and
Georgian (Harris 1982, 1990), A and O are the prototypes of the two main
categories, with the sole main dependent of intransitive verbs either belonging
to the A category or the O category, depending on how A-like or O-like it is
on semantic grounds which vary somewhat from language to language. Split-
intransitive languages are in some sense a compromise between the accusative
and ergative choices. Indeed, they can be portrayed as either basically accusa-
tive languages with the wrinkle that some Ss are marked like members of the
direct object category or as basically ergative languages with the wrinkle that
some Ss are marked like members of the ergative category.
In short, the types of languages whose primary grammatical-relation
marking mechanisms identify two main grammatical relations are as shown
in Figure 2.1. Of course, this typology can be extended to account for variations
among languages concerning a third primary relation, or indirect object (see
Section 1.3.2), by adding presence or absence of an IO category as an inde-
pendent parameter for each language type. Brazilian Portuguese, for example,
is an accusative language in which the indirect object category plays a role,
hence its primary grammatical relations are S/A, O, and IO. Languages such as
Kamaiurá, in which the class of NPs marked with a ‘dative’ adposition behave
systematically like other obliques, could be said to simply lack the IO category.
Similarly, languages can vary with respect to the constitution of the O category,
independently of whether they have indirect objects or not and independently
of questions of ergativity. In a primary-object language (Dryer 1986),
illustrated by Halkomelem, as discussed in Section 2.2.4, recipient and
46 Grammatical relations across languages

ACCUSATIVE SPLIT-INTRANSITIVE ERGATIVE

S S S
A O A O A O
Subject or Object or Ergative Absolutive
nominative accusative

Figure 2.1. Types of languages with two main grammatical relations

beneWciary dependents are included in the O category, whereas in either an IO

language such as Brazilian Portuguese or a non-IO language such as Kamaiurá
they are necessarily excluded.

2.1.1 Accusative languages

Accusative languages are by far the most common (Dixon 1994: ch. 1).
Moreover, languages of the ergative and split-intransitive types often show
signiWcant accusative tendencies of various kinds. Like most Indo-European
languages, Brazilian Portuguese (see Section 1.3.2) manifests many of the
key properties of accusative languages. The head-marking morphology
identiWes an A vs. O distinction. As shown by the following examples, a
proclitic pronominal morpheme is used only for the O of a transitive clause,
whereas a verbal agreement suYx indexes the A or the S, that is the subject.
(2.1) a. Eles o mudaram.
3MascPl 3MascSgO change.Pst3PlSubj
‘They changed it/him.’
b. Ele os mudou.
3MascSg 3MascPlO change.Pst3SgSubj
‘He changed them.’
c. Eles caı́ram / gritaram
3MascPl fall.Pst3PlSubj shout.Pst3PlSubj
‘They fell/shouted.’
An accusative classiWcation of the language is justiWed by the fact that the main
dependent of every type of intransitive clause is indicated by the indexing
system in the same way as the A of a transitive clause. Moreover, to the extent
that syntactic phenomena distinguish among the A, S, and O dependents of
verbs, the lines are usually drawn in terms of an S/A vs. O distinction. For
example, there is a constraint on the controller of the reXexive/reciprocal se
construction, illustrated by the following examples.
 Grammatical relations across languages 47

(2.2) a. Eles se amam. (A controller; O pivot)

3MascPl 3ReX/Rec love.Pres3Pl
‘They love each other/themselves.’
b. Eles nunca mais se falaram. (S controller; IO pivot)
3MascPl never more 3Rec speak.Pst3Pl
‘They never again spoke to each other.’
c. Ele não se mostrou aos hóspedes.
3MascPl Neg 3ReX show.Pst3Sg to.the guests
‘He didn’t show himself to the guests.’ (A controller; O pivot)
*‘He didn’t show the guests each other.’ (*IO controller; O pivot)
As noted in Section 1.3.2, the pivot of this construction, that is the omitted
dependent whose interpretation is keyed to another dependent, must be the
direct or indirect object. In (2.2a), the pivot is the O, which is not expressed
other than by the presence of the reXexive/reciprocal clitic se. Its interpret-
ation is controlled by the A of the sentence.2 Example (2.2b) shows that the
controller can also be an S (the clause is intransitive because the addressee
functions as IO). Example (2.2c) shows that an indirect object, for example,
cannot be the controller. The generalization is that the controller must be a
member of the S/A category.
Another phenomenon whose understanding depends on the same
grammatical-relation category is what is generally known as control of
inWnitival complements. There is a class of verbs with prototype querer
‘want’ that can have a clausal complement, as illustrated by (2.3a). If it
designates the same participant as the subject of querer, the subject of the
embedded verb, which is necessarily in inWnitival form, is not expressed, as in
(2.3b).

(2.3) a. Quero [que ele compre um carro].

want.Pres1Sg that 3MascSg buy.PressSjunct3SgSubj a car
‘I want him to buy a car.’

b. Quero [∅ comprar um carro]. (controllee = A of embedded V)

want.Pres1Sg buy.Inf a car
‘I want to buy a car.’
The constraint of interest in this context has to do with which dependent of
the embedded verb can be the controllee, that is the missing element. Example
(2.3b) shows that it can be the A of the embedded verb. It can also be an S, but
not an O (or any other non-subject):
48 Grammatical relations across languages

(2.4) a. Quero [∅ morrer / ir embora] (controllee = S)

want.Pres1Sg die.Inf go.Inf away
‘I want to die/go away.’

b. *Quero [você amar ∅] (*controllee = O)

want.Pres1Sg you love.Inf
‘I want you to love (me).’
The generalization is that the controllee must be the S/A dependent of the
embedded verb.
It is important to bear in mind that languages are often not all or
nothing with respect to the typology in Figure 2.1. Some grammatical phe-
nomena in a language of one type may work in a way expected for another type.
Although the overall orientation of Brazilian Portuguese is that of an accusative
language, there are some grammatical phenomena that suggest a split-intransi-
tive orientation. For example, the main constraints on constituent order are
sensitive to diVerent classes of intransitive verbs. In general, and ignoring
constructions with emphatic focus in particular, subjects (S/A) must precede
the verb, whereas other dependents follow the verb, with a strong preference
for direct object Wrst, as illustrated by the following examples.
(2.5) a. O vizinho comprou um carro para mim. (A V O oblique)
‘The neighbour bought a car for me.’
b. *Comprou o vizinho um carro para mim. (*V A O oblique)
‘Bought the neighbour a car for me.’
c. O vizinho foi para o supermercado. (S V oblique)
‘The neighbour went to the supermarket.’
d. *Foi o vizinho para o supermercado. (*V S oblique)
‘Went the neighbour to the supermarket.’
e. O vizinho não trabalha. (S V )
‘The neighbour doesn’t work.’
f. *Não trabalha o vizinho. (*V S)
‘Doesn’t work the neighbour.’
The problem is that the S/A category is not monolithic with respect to
constituent order. The postverbal position is often preferred for an S that is
conceived of primarily as a participant—particularly an inanimate one—that
undergoes a change of state or location and, thus, is O-like:3
(2.6)a. De repente aparece o vizinho na porta. (V S oblique)
‘Suddenly shows up the neighbour at the door.’
 Grammatical relations across languages 49

b. Acabou o fósforo. (V S)
‘Ran out the matches.’
c. Não saiu a mancha. (V S)
‘Didn’t come out the stain.’
The postverbal position is restricted to a category of verb dependents that
includes all Os and a certain kind of S, whereas the preverbal position is
required for a category of dependents that includes all As and other kinds of Ss.
At least with respect to basic constituent order, Brazilian Portuguese ap-
pears to take a split-intransitive rather than a strictly accusative approach to
grammatical relations. This is not uncommon for accusative languages, as
various kinds of phenomena implying a split-intransitive organization occur
in such otherwise accusative languages as Japanese (Kishimoto 1996), French
(Legendre 1989b), and Italian (Perlmutter 1989, Van Valin 1990), among
others. The broad implication is that which combinations of the A, S, and
O categories a language uses in its overall grammar is not necessarily deter-
mined by which combinations deWne its main system for marking grammat-
ical relations morphologically (with case and agreement).

2.1.2 Ergative languages

In Ergative languages one or more of the main grammatical-relation marking
mechanisms works in terms of two categories that are deWned diVerently than
in the accusative scenario. The O and S categories are collapsed into one, for
which the case-marking category is generally called ‘absolutive.’4 The other
category (ergative) is restricted to the A dependent. Unlike with most accusa-
tive languages, the A vs. S/O distinction is sometimes only used in the head-
marking or dependent-marking morphological system (or both) and does not
Wgure in the grammar otherwise. Such languages are said to be morphologic-
ally ergative. Even syntactically ergative languages will also often have mix-
tures of accusative-type marking, using the ergative system for dependent
marking, for example, and the accusative system for head marking, or show-
ing ergative marking in the past tense or main clauses and accusative marking
in non-past tenses or subordinate clauses. This phenomenon is generally
known as split ergativity.
Basque is an example of a morphologically ergative language whose overall
syntactic organization appears not to be ergative (Ortiz de Urbina 1989). The
following examples show that the head-marking morphology on the tense
auxiliary agrees with both the A and the O in transitive clauses and with the S
in intransitive clauses and that the dependent-marking morphology treats O
and S in the same way and A diVerently—with the ergative suYx -k.
50 Grammatical relations across languages

(2.7) a. zu-k hura hil zenuen

2Sg-Erg 3Sg.Abs die/kill 2SgA3SgOAux
‘You killed him.’
b. zu hil zen
2Sg.Abs die/kill 2SgSAux
‘You died.’
Outside of the morphology, neither the absolutive nor the ergative grammat-
ical relations appear to play a role in syntactic phenomena. Conjunction
reduction is the phenomenon whereby one of two co-referential NPs is
omitted in one of the conjuncts of two or more conjoined clasues (e.g. [[Susiei
went to the store] and [Øi bought something]]). In Basque, the S/A category is
what matters, as in accusative languages such as English:

(2.8) [∅ seme-a eskolan utzi] eta [klasera joan zen]

A son-Abs at.school leave and to.class go 3SgSAux
‘X left his/her son at school and X went to class.’
That is to say, the interpretation of the omitted A of the Wrst clause is
controlled by the agreement-identiWed S of the second clause. The omitted
NP (or pivot) in such a conjunction-reduction construction would be in the
S/O category in a syntactically ergative language. Other languages reputed to
be strictly or mainly morphologically ergative include Warlpiri (Bittner and
Hale 1996b), Khinalug (Comrie 1978), Abkhazian (Anderson 1976), and Wal-
matjari (Dixon 1979), among many others.
The Australian language Dyirbal (Dixon 1972, Dixon 1994) is probably the
best known example of a syntactically ergative language. As the following
examples show, the case-marking system on full NPs draws a distinction
between A (ergative -Ngu suYx) and S/O (absolutive, with no suYx). (Here
and below, Dyirbal examples are from Dixon 1994: ch. 6).
(2.9) a. Numa miyanda-nyu (Absolutive S)
father.Abs laugh-Nonfut
‘Father laughed.’
b. Numa yabu-Ngu bura-n (Absolutive O; Ergative A)
father.Abs mother-Erg see-Nonfut
‘Mother saw father’.
Although this A vs. S/O dependent-marking system works for all intransitive
verbs and across all tenses and in both main and subordinate clauses, Dyirbal
 Grammatical relations across languages 51

does show a split for the distinction between pronominal and full NPs, the
former being marked on an accusative basis:
(2.10) a. Nana miyanda-nyu (Nominative S)
we all.Nom laugh-Nonfut
‘We all laughed.’
b. Nana nyurra-na bura-n (Accusative O; Nominative A)
we all.Nom you all-Acc see-Nonfut
‘We saw you all.’
c. nyurra Nana-na bura-n (Accusative O; Nominative A)
you all.Nom we all-Acc see-Nonfut
‘You all saw us.’
Now, in spite of this morphological ergative/accusative split, the syntax of
Dyirbal is fundamentally ergative. This can be seen, for example, in conjunc-
tion reduction. As the following example shows the interpretation of the
omitted NP in the second conjunct is controlled by the O of the Wrst clause
and not the A.

(2.11) [ŋuma yabu-ŋgu bura-n] [ ∅ banaga-nyu]

father.Abs mother-Erg see-Nonfut S return-Nonfut
‘Mother saw father and he/*she returned.’
The S/O (absolutive) category is operative in a similar way in control of
purposive clauses. In the purposive control construction, a suYx (-ygu or -li)
is added to the verb of a subordinate clause and a dependent of the subor-
dinate verb is omitted and interpreted as being the same as (or is controlled
by) the S/O of the main clause:

(2.12) a. ŋuma banaga-nyu [ ∅ yabu-ŋgu bura-li]

father.Abs return-Nonfut O mother-Erg see-Purp
‘Father returned in order for mother to see him.’

b. yabu ŋuma-ŋgu giga-n [ ∅ gubi-ŋgu mawa-li]

mother.Abs father-Erg tell-Nonfut O doctor-Erg examine-Purp
‘Father told mother to be examined by the doctor.’

c. yabu ŋuma-ŋgu giga-n [ ∅ banaga-ygu]

mother.Abs father-Erg tell-Nonfut S return-Purp
‘Father told mother to return.’
52 Grammatical relations across languages

The controller can be either an O, as in (2.12b–c) or an S, as in (2.12a); but

cannot be an A. At least with these main clause verbs, the same situation
would obtain in English. What is signiWcantly diVerent about Dyirbal is
that the pivot or controllee must also be an S or O and cannot be an A.
The absolutive category also plays a crucial role in relative clause formation,
as the relativized position is restricted to S/O. Other languages
with both morphological and robust syntactic ergativity include Hurrian
(Anderson 1976), Inuit (Bittner and Hale 1996b), Sama (Foley and Van
Valin 1984).
Many languages have a complex blend of ergative and non-ergative prop-
erties, both morphologically and syntactically, as in the case of Chukchee
(Comrie 1978), Tzotzil (Aissen 1987), Tongan (Dixon 1994), and various Salish
languages, as documented in some detail for Coast Tsimshian (Mulder 1994),
Coeur d’Alene (Doak 1998), and Halkomelem (Gerdts 1988a), for example. As
for dependent marking, Halkomelem only indicates A/S/O vs. oblique (or
non-core), as noted in Section 1.3.3. The preposition ?@ is used for obliques;
other NP dependents of verbs are unmarked. The head-marking morphology
is much more complex. In essence, 1st and 2nd person S/A pronominals are
realized in preverbal position, whereas (nonemphatic) 3rd person A and 1st
and 2nd person O pronominals are realized as verbal suYxes and 3rd person
S/O pronominals are not overtly expressed, as shown by the following
examples.5
(2.13) a. ni ?im@š (3rd person S ¼ Ø)
Aux walk
‘He/she/it walked.’
b. ni c@n ?im@š (1st person S ¼ S/A pronoun)
Aux 1Subj walk
‘I walked.’
c. ni c@n q’wáqw-@t (1st person A ¼ S/A pronoun;
Aux 1Subj club-Tr 3rd person O ¼ Ø)
‘I clubbed him/her/it.’
d. ni q’wáqw-@t-@s (3rd person A ¼ ergative suYx;
Aux club-Tr-3A 3rd person O ¼ Ø)
‘He/she clubbed him/her/it.’
e. ni q’wáqw-@u-ám?š-@s (3rd person A ¼ ergative suYx;
Aux club-Tr-1O-3A 1st person O ¼ O suYx)
‘He/she clubbed me.’
The system is accusative with respect to 1st and 2nd person, as evidenced by
the use of preverbal c@n ‘I’ for both S (2.13b) and A (2.13c) as opposed to a
 Grammatical relations across languages 53

verbal suYx for O (2.13e). Ergativity comes into play only for 3rd person,
which is Ø for both S (2.13a) and O (2.13c–d) but a verbal suYx (-@s) only for
A (2.13d–e). The system is split-ergative not only in terms of person but also
in that even the 3rd person category works on an accusative basis in subor-
dinate clauses.
Certain syntactic phenomena are only sensitive to the A/S/O vs. oblique
distinction. For example, clefting, question, and relative clause constructions
use a simple fronting strategy if the pivot is any kind of A/S/O dependent.
When the pivot in these constructions is an oblique, on the other hand, the verb
of its clause must be nominalized. There are also phenomena that suggest a
split-intransitive orientation, such as a constraint (essentially) limiting syn-
thetic causativization to intransitive verbs whose subject is in control of the
action designated by the verb. The pivot in the raising-to-object construction,
on the other hand, is restricted essentially to the S/A category, including the
‘logical A’ in the passive construction but excluding O (see Section 2.2.1).
Among the phenomena suggesting syntactic ergativitiy are quantiWer Xoat
and possessor extraction. In the quantiWer Xoat construction, instead of
being expressed within the NP that it modiWes, the quantiWer ‘all’ can be placed
in clause-initial position as illustrated by the following examples.
(2.14) a. ni x̌welenčén@m m@
k’w kwu@ sl’@l?ı́q@L


Aux run.Pl all Det children

‘All the children ran.’
b. m@
k’w niw x̌welenčén@m kwu@ sl’@l?ı́q@L


all Aux.Lnk run.Pl Det children

‘The children all ran.’
In (2.14b) the quantiWer is fronted or ‘Xoated’ from an S. If a quantiWer is
Xoated in a transitive clause it can be associated with the O but not the A, as
shown by (2.15). Thus, the generalization is that only an absolutive NP can be
associated with a Xoated quantiWer.
(2.15) m@
k’w niw q’w@
l-@t-@s tu@ sl’@l?ı́q@L kwu@ s@plı́l


all Aux.Lnk bake-Tr-3A Det children Det bread

‘The children baked all the bread/*All the children baked the bread.’
Possessor extraction in the form of clefting, which puts a dependent in
contrastive focus, is shown for the possessor of an S dependent in (2.16a).
Although it is possible to cleft the possessor of an O, as well (2.16b), it is not
possible to cleft the A of a transitive clause (2.16c). The constraint is, therefore,
that only the possessor of an absolutive dependent can be clefted (or, more
generally, fronted in an extraction construction).
54 Grammatical relations across languages

(2.16) a. ?@
n?u@ ni

t
s q’áy kwu@ n@-sqw@m@
y?
1Emph Aux die Det 1Pos-dog
‘It was me whose dog died.’
b. n@
w@ ni ?á -t-?é n?
t
s
t
s u@ @n-stá?l@s
2Emph Aux call-Tr-1SgSubj Det 2Pos-spouse
‘It was you whose wife I called.’
c. *n@
w@ ni ?á -u-ám?š-@s u@
t
s @n-stá?l@s
2Emph Aux call-Tr-1O-3A Det 2Pos-spouse
‘It was you whose wife called me.’
The key facts about ergativity that any theory of grammatical relations
needs to be able to account for are the following.
. One or more of the main morphological grammatical-relation marking
devices in a language can work in terms of a distinction between A and S/O
categories, that is on an ergative basis.
. Languages with ergative morphological marking devices generally have at
least some morphological marking devices that work on an accusative (or
non-ergative) basis and may or may not have any syntactic phenomena
whose constraints operate on an ergative basis.
. In syntactically ergative languages the most prominent grammatical rela-
tion is generally absolutive, that is the S/O category.

2.1.3 Split-intransitive languages

In split-intransitive languages the S, A, and O categories are collapsed, at least
for one or more of the main grammatical-relation marking phenomena, into
two primary categories: A with some A-like Ss (henceforth Sa ) and O with
some O-like Ss (henceforth So ). As with languages generally classiWed as being
ergative, split-intransitive languages tend to vary quite a bit in terms of how
pervasive their deWning two-category system is in the grammar. In some
languages head marking and/or dependent marking operates on a split-
intransitive basis, although other grammatical phenomena are by and large
insensitive to the A=Sa vs. O=So distinction. In other languages, the A=Sa vs.
O=So distinction is the only one that matters, not only for the head and
dependent marking but also for other grammatical phenomena that depend
on distinctions among A, S, and O dependents.
Kamaiurá (Seki 2000), discusssed in Section 1.3.2, has a split-intransitive
system that appears to be limited to the marking of grammatical relations on
verbs. Two main grammatical relations are indicated by head-marking (agree-
ment) morphology realized on the verb, as well as by a preferred preverbal
ordering and no postposition marking.6 If a clause has both A and O
 Grammatical relations across languages 55

constituents, the preferred order is A O V (oblique). If a clause has an S

constituent, the preferred order is S V (oblique). Thus, in terms of constituent
order, there is an A vs. O distinction; but there is no way of knowing whether S
is considered to be in the A category (as in an accusative language) or the O
category (as in ergative languages). The verb morphology, however, draws the
grammatical-relation lines in a split-intransitive way. In transitive clauses,
preWxes or clitics on the verb generally index either the A or O, the choice
being contingent Wrst on which is higher on a person hierarchy (1st > 2nd >
3rd) and, otherwise, on which is higher on a grammatical-relation hierarchy
(A > O).7 Thus, in (2.17a) the o-preWx indicates that the A is 3rd person, whereas
in (2.17b) the je clitic indicates that the O is 1st person singular. If the A is 1st
person, the preWx a- appears on the verb instead of je as shown by (2.17c).
(2.17) a. ywyrapara o-me’e
kara’iwa upe
bow 3-give foreigner to
‘He gave the bow to the foreigner.’
b. kunu’uma je¼retsak
child 1Sg¼see
‘The child saw me.’
c. kunu’uma a-retsak
child 1Sg-see
‘I saw the child.’
The crucial fact with respect to the core grammatical-relation typology is
that with some intransitive verbs the A-indexing kind of morphology (a- ¼
1st person singular, for example) is used for the S dependent, whereas with
other verbs, the O-indexing kind of morphology is used (je ¼ 1st person
singular, for example), as shown by the following examples.
(2.18) a. a-ja’eo (S marked like A)
1Sg-cry
‘I cried.’
b. je¼’amot (S marked like O)
1Sg-be homesick
‘I’m homesick.’
(2.19) a. kunu’uma o-wawak (S marked like A)
child 3-wake up
‘The child woke up.’
b. je¼yar i-katu (S marked like O)
1Sg¼canoe 3-be good
‘My canoe is good.’
56 Grammatical relations across languages

In general, the A-marking (or nonstative) verbs designate actions or events

(walk, eat, sing, die, fear, speak, break, fall, roll, etc.); the O-marking (or
stative) verbs designate states (long, short, tall, happy, hard, soft, old, black,
etc.). Although the stative verbs correspond typically to what would be
adjectives in languages like English, they behave like verbs in Kamaiurá, as
they cannot occur, for example, with a copula unless they are transformed
into a noun and they can occur with verbal imperative morphology and with
the same nominalizing aYxes that are otherwise only used with transitive and
nonstative intransitive verbs.
Split-intransitive languages cross-linguistically show a similar pattern se-
mantically, with the A-marking vs. O-marking distinction correlating gener-
ally either with a nonstative vs. stative distinction or some kind of controlled
action/state vs. uncontrolled action/state distinction (Mithun 1991).8 In the
North-American Muskogean language Choctaw (Davies 1986), for example,
an S determines A head marking if the action or state the verb designates is
conceived of as having a participant that is in control of it, as shown in Table
2.1.9 Unlike in Choctaw, verbs such as ‘sweat’ and ‘break’ take A marking in
Kamaiurá, because stativity, rather than participant control, is criterial.
Some theoreticians (e.g. Foley and Van Valin 1984: ch. 3, Dixon 1994: ch. 4)
systematically distinguish a sub-type of split-intransitive case-marking
system in which intransitive verbs alternate between A marking and O
marking, depending on a situational interpretation. Choctaw, for example,
displays the ‘Xuid-S’ property, insofar as at least some intransitive verbs, such
as ttola-‘fall’ and habishko-‘sneeze,’ take A marking to designate intentional
actions and O marking otherwise. It is unclear, however, from Davies’ descrip-
tion how Xuid Choctaw is. Although intransitive verbs are not Xuid in this sense
in Kamaiurá, they become Xuid when nominalized for the purpose of forming
relative clauses, for example. More speciWcally, although nominalized stative
verbs show only O-indexing morphology, nominalized nonstative verbs can
show either A-indexing or O-indexing morphology, depending on whether the
S is interpreted as being in control or not, as in the following examples.
(2.20) a. o-je’e
-uma’e (S marked like A; nonstative verb root)
3-speak-Nominal.Neg
‘one who (deliberately) doesn’t speak.’
b. i-je’e
-uma’e (S marked like O; nonstative verb root)
3-speak-Nominal.Neg
‘one who doesn’t speak.’
c. i-pituw-uma’e (S marked like O; stative verb root)
3-lazy-Nominal.Neg
‘one who isn’t lazy.’
 Grammatical relations across languages 57

Table 2.1. Choctaw A vs. O marking and intransitive

verb classes
A and Sa marked with -li O and So marked with sa-

chi-bashli-li-tok is-sa-bashli-tok
2SgO-cut-1SgA-past 2SgA-1SgO-cut-past
‘I cut you.’ ‘You cut me.’
hilha-li-tok sa-hohchafoh
dance-1SgA-past 1SgO-be hungry
‘I danced.’ ‘I’m hungry.’
A-marking intransitive Vs O-marking intransitive Vs
eat break
play hurt
run suVer
go be lost
arrive be cold
work sweat
jump be ashamed
walk be tall

This kind of Xuidity extends to large classes of intransitive verbs in

such languages as Acehnese (Durie 1985), discussed below, and Tsova-Tush
(Holisky 1987).
Looking beyond what the head-marking morphology indicates about
grammatical relations, syntactic phenomena in Kamaiurá appear to be sensi-
tive to various groupings of the A, O, and S dependents. For example, there is
a phenomenon, generally known as ‘possessor raising’, wherein the S or O of a
verb is interpreted as the possessor of the referent of a noun that is incorp-
orated into the verb. More speciWcally, there are alternative ways of treating
the possessor of S and O dependents. In the (a)-examples in (2.21)–(2.23) the
possessor is a dependent of the NP, indicated by indexing verbal morphology
on the head N itself. The (b)-examples, which illustrate the possessor-raising
phenomenon, have the head N incorporated into the verb and the NP that is
interpreted as its possessor functioning as the O-marked dependent of the
verb, as indicated by the pronominal preWx on the verb.
(2.21) a. ne¼atua a-perek
2Sg¼nape 1Sg-hit
‘I hit your nape.’
b. oro-atua-perek
1Sg2Sg-nape-hit
‘I hit you on the nape (of the neck).’ (Literally, ‘I nape-hit you.’)10
58 Grammatical relations across languages

(2.22) a. ne¼’ajura n¼i-huku-ite

2Sg¼neck Neg¼3-be long-Neg
‘Your neck isn’t long.’
b. na¼ne¼’aju-wuku-ite
Neg¼2Sg¼neck-be long-Neg
‘Your neck isn’t long.’ (Literally, ‘You’re not long-necked.’)
(2.23) a. ne¼rea 0-jektosı̃
2Sg¼eye 3-roll
‘You rolled your eyes.’ (Literally, ‘Your eyes rolled.’)
b. ne¼rea-jektosı̃
2Sg¼eye-roll
‘You rolled your eyes.’ (Literally, ‘You eye-rolled.’)
SigniWcantly, possessor raising (with associated incorporation) is only pos-
sible from an O or S dependent and it can occur both with stative verbs, as in
(2.22b), and nonstative verbs, as in (2.23b), although if the possessor is ‘raised’
from the S of a nonstative verb it is indexed with O marking on the verb
rather than the usual A marking. Possessor raising is thus restricted to the S/O
category, which is to say the absolutive dependent.
Although the way possessor raising works implies an ergative syntax, other
phenomena operate in terms of other relational categories, which do not
imply ergativity. For example, there are diVerent precise strategies for forming
relative clauses depending on the grammatical relation of the relativized
position. With respect to relativization, the following categories of verb
dependents are distinguished: A, S, O, and oblique. The Ss of both stative
and nonstative intransitive verbs are relativized in the same way (with a
distinct relativizing aYx on the verb and retention of the pronominal preWx
or clitic that indexes the relativized consitutent), whereas the A and O of
transitive verbs are relativized in diVerent ways (each having its own relativ-
izing aYx on the verb and neither being indexed by a pronominal preWx or
clitic). On the other hand, with respect to reciprocalization and clause-
internal reXexivization, there are constraints on the controller that put the
A and S into the same category, that is the subject or nominative category.
Choctaw, which shows an A=Sa vs. O=So distinction with respect to head-
marking morphology on the verb, also fails to work primarily in terms of an
A=Sa vs. O=So split. For example, according to the analysis presented in
Davies (1986), the controller of reXexivization has to be either an A or any
kind of S. Similarly, the nominative case-marker that occurs on NP depend-
ents puts the A and S together into the same category, as indicated by the
following examples.
 Grammatical relations across languages 59

(2.24) a. oW-yat towa-yã ilhioli-tok

dog-Nom ball-Obl chase-Pst
‘The dog chased the ball.’
b. issoba-yat ı̃pa-tok
horse-Nom eat-Pst
‘The horse ate.’
c. chim-alla-t cha ha-h
t
s

2Pos-child-Nom tall-Pred
‘Your child is tall.’
Head marking doesn’t occur in these examples because the A, S, and O
dependents are all third person and therefore the verbs show no overt
agreement. Nominative case marking obligatorily occurs on S/A dependents;
oblique case marking occurs optionally on all other dependents, including Os.
As can be seen from (2.24a), the A takes the nominative case marking, whereas
the O takes the oblique suYx. The S of verbs in the ‘eat’ class, which show A
agreement, and the S of verbs in the ‘tall’ class, which show O agreement, both
take the nominative suYx, as shown by (2.24b–c) (the -t vs. -yat diVerence
reXects phonologically-conditioned allomorphy). Although the constraint on
the controller of reXexivization and the subject case-marking imply an ac-
cusative syntax, there are some phenomena that operate in other terms.
Unlike in Kamaiurá, which allows possessor raising from both O-marked
and A-marked Ss, Choctaw restricts possessor raising to O and So , thus
showing more of a tendency toward syntactic split-intransitivity.
The Austronesian language Acehnese, spoken in Indonesia (Durie 1985,
1988, Van Valin and La Polla 1997: ch. 6), also draws a distinction between two
categories: A=Sa and O=So , as can be seen from (2.25). (The analysis and
examples here and below are adapted from Van Valin and La Polla 1997.11)
(2.25) a. lôn lôn¼mat¼geuh
1Sg 1Sg¼hold¼3
‘I hold him/her.’
b. gopnyan geu¼mat¼lôn
3Sg 3¼hold¼1Sg
‘He/she holds me.’
c. geu¼jak gopnyan
3¼go 3Sg
‘He/she goes.’
d. lôn rhët¼lôn
1Sg fall¼1Sg
‘I fall.’
60 Grammatical relations across languages

e. *lôn lôn¼rhët
1Sg fall¼1Sg
‘I fall.’
The 3rd person and 1st person singular free-standing pronouns (gopnyan and
lôn), whose morphological form is insensitive to the A vs. O distinction, may
be expressed in addition to pronominal clitics (which are sometimes hom-
ophonous with the free-standing pronouns) as in the examples above, al-
though often only one or the other is expressed. The morphological form of
the pronominal clitics is also insensitive to the A vs. O distinction (geuh, in
(2.25a) being a phonologically-conditioned allomorph of geu). What is sig-
niWcant with respect to grammatical relations is the placement of the pronom-
inal clitics. The generalization is that a proclitic (i.e. verb-initial clitic) involves
A-marking (2.25a–b) or Sa-marking, as with an intransitive verb such as jak
(2.25c) whose main dependent is typically conceived of as being in control of
the action. An enclitic involves O-marking (2.25a–b) or So-marking, as with an
intransitive verb such as rhët (2.25d) whose main dependent is not typically
conceived of as being in control. Example (2.25e) is ungrammatical because
the 1st person clitic is expressed as a proclitic, inappropriately marking the 1st
person dependent like an A.
What makes Acehnese diVerent from languages such as Kamaiurá and Choc-
taw is that there appear to be no grammatical phenomena for which syntactic
functions need to be deWned in terms of a three-way distinction between A, S,
and O, an accusative-type distinction between O and subject (S/A), or an
ergative-type distinction between A and absolutive (S/O). The two categories
A=Sa and O=So appear to suYce for any phenomena requiring a distinction
among the A, S, and O categories. Consider, for example, the ‘possessor raising’
construction. Like in Kamaiurá, a possessor of an O or S can either be expressed
as a constituent of the NP, as in the (a)-examples in (2.26)–(2.27) or can function
as a dependent of the verb, in which case the head (possessed) N is incorporated
or compounded with the V, as in the (b)-examples.
(2.26) a. seunang [até lôn]
happy liver 1Sg
‘I’m happy.’ (Literally, ‘My liver (is) happy.’)
b. lôn seunang-até
1Sg happy-liver
‘I’m happy.’ (Literally, ‘I’m liver-happy.’)
(2.27) a. ka lôn-tët [rumoh gopnyan]
Asp 1SgA-burn house 3Sg
‘I burned his/her house.’
 Grammatical relations across languages 61

b. gopnyan ka lôn¼tët-rumoh
3Sg Asp 1SgA¼burn-house
‘I burned his/her house.’ (Literally, ‘Him/her, I house-burned.’)
(2.28) * gopnyan ka aneuk-woe
3Sg Asp child-return
‘His/her child returned.’
The key fact is that a possessor can only ‘raise’ from an O (2.27b) or So (2.26b);
not from an Sa , as shown by the ungrammaticality of (2.28). Thus, the
phenomenon is keyed to neither the O category (accusative), nor an S/O
category (absolutive), but to an O=So category.
As an example of a phenomenon keyed to the A=Sa category, consider control
of a verb phrase embedded under the verb ‘want’. In English, as in Brazilian
Portuguese (see Section 2.1.1), an inWnitival verb phrase can be embedded under
verbs such as want, in which case the subject dependent of the inWnitival phrase
is not expressed if it is the same as the subject dependent of the main verb (e.g. I
want [to die/fall down/go home/see a movie]). Similarly, in Acehnese the A of the
verb taguen ‘cook’ is not overtly expressed if the phrase taguen bu ‘cook rice’ is
embedded under tém ‘want’ and the experiencer of tém is understood to be the
same participant as the A of taguen:

(2.29) gopnyan geu-tém [∅ taguen bu] (controllee = A of embedded V )

3Sg 3=want cook rice
‘He/she wants to cook rice.’
Thus, the interpretation of the unexpressed dependent of the embedded
phrase is ‘controlled’ by the experiencer of tém. The question is which
dependent of an embedded verb can be controlled in this construction. As
the following examples show, unlike in English and typical accusative lan-
guages, the controllee can be an Sa dependent of an intransitive verb, but not
an So dependent.

(controllee = Sa of embeded V )
(2.30) a. gopnyan geu-tém [∅ jak]
3Sg 3=want go
‘He/she wants to go.’

b. *gopnyan geu-tém [∅ rhët] (*controllee = S0 of embeded V )

3Sg 3=want fall
‘He/she wants to fall.’
62 Grammatical relations across languages

Other syntactic phenomena generally do not systematically distinguish

among A, S, and O dependents. The lack of any phenomena sensitive spe-
ciWcally to O or S/A categories poses a potential challenge to theories built
around the traditional subject and direct object categories of accusative
languages.
The main features of split-intransitive languages can be summarized as
follows.
. At least with respect to a principal role-marking mechanism—typically
head-marking morphology—a distinction is drawn between the A depend-
ent and the O dependent and the S category is split into A-patterning and
O-patterning Ss.
. There is generally a fairly clear underlying semantic motivation for the S-
category split, typically based on a distinction between event-designating
verbs (with Sa ) and stative verbs (with So ) or between participant in control
(Sa ) and participant not in control (So ).
. Although in some languages the only distinction among A, S, and O
categories that any morphosyntactic phenomena are keyed to is the A=Sa
vs. O=So distinction, in other languages this distinction is mainly or only
manifested morphologically (in the form of head or dependent marking)
and other grammatical phenomena are sensitive to grammatical-relation
distinctions of either the accusative kind or the ergative kind or both.

2.2 Voice and grammatical-relation alternations

As should be clear from the analyses of grammatical-relation marking
parameters sketched in the previous section, categories such as A, S, and O
and the grammatical relations deWned in terms of these (subject, object,
absolutive, etc.) play an important role in the grammatical phenomena
of many languages. Languages generally provide their users with ways of
doing such things as explicitly contrasting one of the participants in a state
of aVairs being talked about with some other. For example, one person says she
thinks Bill likes Melissa; another wants to make clear that Ralph rather than Bill
is the one who likes Melissa. Generally, one of the NPs or other phrases can be
placed in a special position, at the beginning of the sentence for example, in
order to signal contrastive focus of the desired kind. Perhaps a focus-suYx,
particle or auxiliary verb will be added as well and the main clause may be put
in a special subordinate form. Thus, one might get something such as:

(2.31) It was Ralph that [∅ kissed Melissa].

 Grammatical relations across languages 63

Since the subordinate clause is in some sense missing one of the dependents of
its verb because it has been placed in the special position, in order to get the
right interpretation of a sentence such as (2.31), it is necessary to know that
the displaced NP ‘counts’ as the missing NP. Now, in English the AVO
constituent-order convention and the relevant convention concerning the
relationship between semantic roles and the A and O categories make it
clear that the missing NP in the subordinate clause plays the kisser role.
That is to say, the A of kiss is associated with the kisser role and since there
is a gap between the subordinating complementizer that and the verb kissed it
can be inferred that the A is what is missing.
Suppose, however, that English had, instead, a VAO basic order for its
clauses. It would not be possible in that case to tell whether (2.31) means
‘Melissa kissed RALPH’ or ‘RALPH kissed Melissa’, since a gap preceding or
following Melissa would sound the same:
?
(2.32) It was Ralph that [kissed ∅Melissa ∅].
Hypothetical VAO English
This kind of ambiguity can arise for such a construction in languages with verb-
initial, verb-Wnal, or free constituent order. Of course, the problem could be
solved by having a dependent-marking system or, perhaps, requiring the use of a
pronoun in the position of the gap (It was Ralph that kissed he Melissa, for
example, would work in a VAO language). In any case, one way that grammars
manage to facilitate the correct interpretation of such constructions is by
placing grammatical-relation constraints on them. For example, if by general
convention the missing NP in the subordinate clause in such a construction had to
be the subject, Ralph would automatically be interpreted as the kisser in (2.32).
Although the grammatical conventions that languages adopt can be com-
plex and do not always have such transparent functional motivations, they do
often work essentially in this way for all intents and purposes. One of the
problems for a language such as hypothetical VAO English with a subject-only
constraint on the contrastive focus (or ‘cleft’) construction is that there would
be no way to say the equivalent of It was Melissa that Ralph kissed. It turns out,
however, that languages often have mechanisms for changing the relationship
between semantic roles and syntactic functions, such that the kissee, for
example, can have the syntactic function that by default gets assigned to the
kisser. Thus, by using the passive-voice construction in the subordinate
clause, the interpretation ‘It was Melissa that Ralph kissed’ can be achieved,
without violating the subject constraint on clefting. That is, hypothetical VAO
English could just use It was Melissa that was kissed by Ralph.
64 Grammatical relations across languages

In some languages, voice has a somewhat diVerent eVect. Rather than

primarily maximizing the availability of syntactic privilege to NPs with
diVerent semantic roles, voice marking can provide information about the
intended semantic role interpretation of the participants in the event/state
designated by the verb, while allowing a faithful alignment of the privileged
syntactic function with the cognitively most salient verb dependent. Lan-
guages frequently place constraints on which syntactic function can be asso-
ciated with NPs whose referents are ranked highly in some way on a
prominence hierarchy that operates primarily in terms of person and animacy
(e.g. 1st/2nd person > human > animate > inanimate) (Silverstein 1976,
Delancey 1981, Aissen 1999b). The person part of this kind of hierarchy is
operative in Kamaiurá (as discussed in Section 2.1.3)—but without restricting
syntactic functions per se. The head-marking morphology on a verb only
indexes one of the A/S/O dependents, in most circumstances. For example, if
one is 3rd person and the other is 1st or 2nd person, the 1st or 2nd person
dependent is the only one indexed, irrespective of its syntactic function (A or
O). In some languages, however, the A-marked dependent simply cannot be
outranked by the O-marked dependent in terms of person/animacy or, in
some cases, discourse saliency. Thus, in a hypothetical English with such a
constraint, I kissed her would be possible, because the 1st person A is higher on
this hierarchy than the 3rd person human O. But, there would be no way to
say She kissed me without violating the constraint. A passive-voice construc-
tion (I was kissed by her), for example, makes it possible to express the
meaning ‘She kissed me’, without violating the constraint. The 1st person
dependent is the S, which is interpreted as the kissee because the S in a passive
construction is interpreted as having the patient semantic role.
Voice alternations, such as passive vs. active in English, constitute a crucial
component of the grammatical-relation systems in many languages and,
consequently, are one of the main issues of concern for theories of grammat-
ical relations. Just as there are diVerent varieties of languages according to
which grammatical relations tend to be privileged, that is S/A (in accusative
languages) vs. S/O (in ergative languages), for example, there are correlative
diVerences in kinds of voice constructions and kinds of grammatical-relation
alternations. In languages with an accusative syntax in which S/A is often the
most privileged category, the passive construction takes on a special sig-
niWcance, since, although it may have other motivations, one of its main
eVects is to allow dependents with the patient semantic role (and sometimes
other roles) to have the privileged subject function. In languages such as
Dyirbal, the antipassive voice construction is of special importance, since it
allows the agent to have the absolutive function by demoting the patient to an
 Grammatical relations across languages 65

oblique (or other non-core) function. In many languages a critical line is

drawn between the core syntactic functions (S/A/O) and oblique relations.
Correspondingly, some languages have multiple voice constructions (locative,
instrumental, benefactive, etc.) or what are generally known as applicative
constructions, which have the eVect of increasing the privileges or grammat-
ical salience of dependents with such semantic roles as beneWciary, recipient,
goal, instrument, and location, by putting them in the O (or S/O) category.
Although much rarer, an inverse voice construction can allow the A-marked
and O-marked dependents to have the semantic roles of patient and agent,
respectively, which is the inverse of the alignment found in the direct voice.
Although inverse voice may have other motivations, one eVect can be to
increase the privileges or grammatical salience of the patient (in a basically
accusative language), while keeping the agent in one of the core functions.

2.2.1 Passive voice

As with all of the diVerent kinds of voice categories, there are potentially
diVerent ways of deWning the category designated by passive. The prototypical
transitive active clause has an agent in the A function and a patient in the O
function. As schematized in the contrasting valence structures in (2.33), the
prototypical passive paraphrase is intransitive, with the patient in the S
function, the agent omitted or in an oblique function, and some overt
morphosyntactic marking of the voice diVerence.
(2.33) a. VerbX < AAGENT OPATIENT >
Active Voice
b. VerbX -Pass < SPATIENT (OblAGENT ) >
Passive Voice
It is unclear, however, which, if any of the prototypical properties of the
passive construction should be made criterial for taxonomic purposes, and
whether anything of theoretical signiWcance hinges on the choice. Presumably
one would not want to exclude the English passive construction, simply
because rather than having a verbal aYx dedicated to announcing passive
voice, it announces it with a unique combination of multifunctional markers,
that is the be auxiliary together with a past-participle form of the verb.
Suppose, however, a language had a specially-marked construction, diVering
from an active transitive clause in form but not meaning, in which the agent
were marked like and behaved syntactically like the O of a transitive clause
rather than an oblique. Would that construction be classiWed as passive? That
is to say, consider a hypothetical English with active and non-active para-
phrases of the following kind:
66 Grammatical relations across languages

(2.34) a. This person helped my mother.

b. My mother was helped this person.
This hypothetical English would have a special voice construction in which
the patient is promoted to A rather than S and the agent is demoted to O
rather than oblique. A theory of grammatical relations would have to have a
way of characterizing the situation and accounting for the syntactic properties
of the non-active construction, whether it is called passive or not. One could
distinguish it from the prototypical passive by calling it ‘reversal passive’.
Although this general kind of voice construction in fact appears to exist, it
is classiWed in this book as an instantiation of what is generally known as
inverse voice (see Section 2.2.3). Taken to be criterial for the passive designa-
tion of a construction are the following properties:
(2.35) A clause is said to be in passive voice if it
a. is an intransitive clause of a type that functions as a systematic
alternative to some more basic transitive clause type, and
b. the dependent that would be the A in the basic clause type does not
have any A/S/O function.
There are various constraints on syntactic phenomena in English that
operate in terms of the S/A category. For example, the pivot of conjunction
reduction must be the S/A of the second of two conjoined clauses ([Sam went
there] and [Ø interviewed Chris] vs. *[Sam went there] and [Chris interviewed
Ø]). Another construction, commonly known as the ‘raising to subject’
construction, allows a dependent of an inWnitival phrase embedded under
certain verbs such as seem to be overtly expressed only as the subject of seem.
As illustrated by the following examples, only an A or S dependent of the
embedded verb can be the subject of seem in this construction.

A of embedded V = subject of seem

(2.36) a. This drug seems [∅ to help people]

*O of embedded V = subject of seem

b. *People seem [this drug to help ∅]

S of embedded V = subject of seem
c. People seem [∅ to know about this drug]
In order to express the meaning of (2.36b), it is necessary for the patient of the
embedded verb to have either the A or the S syntactic function. The passive
 Grammatical relations across languages 67

construction makes this possible, since the patient (or what would otherwise
be the O) is expressed as the S and the agent is omitted or expressed as an
oblique (or ‘adjunct’):

S of embedded passive V = subject of seem

(2.37) People seem [∅ to helped by this drug]

Similarly, the meaning and eVect of *[Sam went there] and [Chris interviewed
Ø] can be achieved by passivizing the second conjunct ([Sam went there] and
[Ø was interviewed by Chris]). Thus, one of the eVects of the passive con-
struction in English is to increase the syntactic privileges of the patient
dependent.
The passive construction of the Tanoan native American language South-
ern Tiwa (Allen and Frantz 1983) works like English in that it can increase the
syntactic prominence of the patient, as illustrated by the following examples.
(2.38) a. seuanide hliawra-mu-ban
man lady-see-Pst
‘The man saw the lady.’
b. hliawrade mu-che-ban seuanide-ba
lady see-Pass-Pst man-Instr
‘The lady was seen by the man.’
Southern Tiwa is a typical polysynthetic language (Baker 1996), with noun
incorporation and a robust head-marking system for marking A/S/O syntactic
functions. Dependent marking is used only for oblique constituents, which bear
case suYxes. Example (2.38a) is a transitive active clause, with the O incorpor-
ated in the verb, as it must ordinarily be.12 There is no agreement morphology on
the verb because 3rd singular animate S or A and O trigger null agreement.
Example (2.38b) is the passive version of the same proposition. The perceiver (or
agent) dependent is expressed as an oblique, marked with instrumental case; the
percept (or patient) dependent is expressed as the S. There is a complex set of
agreement (or pronominal) preWxes that index the S, A, and O dependents,
including portmanteau markers for numerous A-O combinations. Although
there are preWxes for combinations of A and O such as 1st person singular A and
3rd person O, there is no preWx that indexes the combination 3rd person A and
1st or 2nd person O. There is a prominence hierarchy (1st/2nd > 3rd) that
prevents such a combination. The A of a clause cannot be lower on the person
prominence hierarchy than the O of the same clause. Thus, there is no way to
express the meaning of English The lady saw me, except by using the passive
construction, as illustrated by the following examples.
68 Grammatical relations across languages

(2.39) a. *hliawrade ?-mu-ban ? ¼ any potential preWx or no preWx

lady 3SgA1SgO-lady-see-Pst
‘The lady saw me.’
b. hliawrade-ba te-mu-che-ban
lady-Instr 1SgS-see-Pass-Pst
‘I was seen by the lady.’
There is considerable variation across languages with respect to how the
default A (or agent) is expressed in the passive construction and the extent to
which it displays the typical syntactic properties of A/S/O dependents. The
following examples illustrate some of the kinds of variation in passive agent
expression, beyond the by/instrumental marking of Southern Tiwa and Eng-
lish, for which there are analogues in many languages.
(2.40) Kamaiurá (Seki 2000)
a. ’a
hoka i-’awyky-pyr-era morerekwara upe
this house 3-make-Nominal-Pass chief to/for/by
‘This house was made by/for the chief.’
Halkomelem (Gerdts 1988a)
b. ni q’w@
l-@t-@m ?@ u@ sLéni? tu@ scé Lt@n
t
s

Aux bake-Tr-Intr Obl Det woman Det salmon

‘The salmon was baked by the woman.’
Seri (Marlett 1984, Farrell et al. 1991)
c. m-yo-a ?-kašni
t
s

2SgSubj-Dist-Pass-bite
‘You were bitten.’
In Kamaiurá the passive-voice form of the verb is nominalized and suYxed
with a passive marker. The agent is expressed as an oblique, marked with the
postposition upe that is also used for recipient and beneWciary dependents.
Since the agent can be omitted, (2.40a) is ambiguous between a reading on
which the upe-marked NP is interpreted as the agent and a reading with an
implicit agent and an upe-marked beneWciary. From Seki’s description, the
upe-marked agent of a passive clause has the syntactic behaviour typical of
any oblique. As shown by (2.40b), the Halkomelem passive voice is marked on
the verb by a general intransitive marker that follows the transitive marker. As
expected for an intransitive clause with a 3rd person S, the verb does not show
any agreement. The S of passive clauses behaves in all respects like an S
dependent (see Section 2.1.2). The agent is marked by the all-purpose oblique
preposition. Seri, a Hokan language of Mexico, diVers from these other
languages in that the passive agent must be omitted, as in (2.40c).
 Grammatical relations across languages 69

Although the passive agent is marked like an oblique in Halkomelem, its

syntactic behaviour is A-like in at least one respect. There is a raising-to-object
construction, illustrated by the following examples, that generally allows only
the S/A of the embedded clause to precede the embedded clause and function
as the object of the main clause verb x̌ec- ‘wonder, Wgure out, check out’.

A of embedded clause = O of xec-

(2.41) a. ?i c n e xéc-t tθ sw y?qe?

e e
Aux 1Subj wonder-Tr Det man
[?u ni-? se ce? ?u c’ew- t-ál?xw- s ∅] e e
Lnk Aux-3SgSubj Fut Lnk help-Tr-1P1O-3Ssubj
‘I’m checking out the man if he will help us.’

S of embedded clause = O of xec-

b. ?i c n e xéc-t tθ xw l nít m
e ee e
Aux 1Subj wonder-Tr Det white men
[?u ni-? s e c lkwsta?m t ∅] e e
Lnk Aux-3SgSubj do
‘I wonder what the white men will do.’

*O embedded clause = O of xec-

c. * ?i c n e xe?xcí-t kwθ ní e
Aux 1Subj wonder.Cont-Tr Det 3Emp
[?u ni;n c ? y we l m-n xw
e e ∅] e e
Lnk Aux1SgSubj fut again see-LcontrTr
‘I am wondering if I will see that one again.’
However, if a passive clause is embedded under this raising verb, either the S
or the demoted agent can be raised:
S of embedded passive clause = O of xec-

(2.42) a. ?i c n
e xe?xcí-t kwθ Bob e
Aux 1Subj wonder.Cont-Tr Det Bob
l ?l m-? t- m? ? - y John ∅]
,
[?u ?i-? se e e e e e
Lnk Aux-3SgSubj look-Tr-Intr Obl-Det John
‘I am wondering if Bob is being watched by John.’
70 Grammatical relations across languages

Agent of embedded passive clause = O of xec-

b. ?i c n
e xe?xcí-t kwθ John e
Aux 1Subj wonder.Cont-Tr Det John
[?u ?i-? s
e l ?l m-? t- m? ∅ kwθ Bob
e e e e e
Lnk Aux-3SgSubj look-Tr-Intr Det Bob
‘I am wondering if Bob is being watched by John.’
Languages vary with respect to how many and what kind of S-/A-like priv-
ileges the agent in a passive clause may have. In French, for example, the oblique-
marked agent of a passive clause behaves like an S/A with respect to control of
certain kinds of adverbial phrases (Ruwet 1972, Legendre 1987, 1990), suggesting
the need to recognize essentially the same S/A/Obla category that is needed for
Halkomelem raising to object. There are certain phenomena involving control
(or antecedence) of reXexive pronouns that work in terms of the same kind of
category in the Indo-Aryan language Marathi (Rosen and Wali 1989).
In Seri, even though it cannot be overtly expressed, the agent of a passive
verb takes precedence over the S for the purposes of switch-reference mark-
ing—a morphological marking system for keeping track of the reference of
‘subjects’ across main and adverbial subordinate clauses. The following ex-
amples (adapted from Farrell et al. 1991) illustrate the possibility of using the
diVerent-subject (DS) marker with both active and passive clauses of diVerent
kinds.13
2Sg A in subordinate clause; 2Sg S in main clause; *DS
(2.43) a. [[mi-naiL kom m-po-ki xk t
s (*ta)-x] ?ata p ko-m-si-a
t
s
t
s

2Pos-skin the 2SgSubj-Irr-wet DS-Aux mucus 3Obl-2SgSubj-

Irr-be
?a¼?a]
Aux¼Decl
‘If you wet your skin, you will be with mucus.’ (i.e. ‘get a cold’)
3Sg A in subordinate clause; 1Sg S in main clause; DS
b. [[?im-t-kašni ma-x] ?p-yo-o ?a]
t
s

1SgO-Real-bite DS-Aux 1SgSubj-Dist-cry

‘Since it bit me, I cried.’
Øagent in main passive clause; 2Sg A in subordinate clause; DS
c. [m-yo-a ?-kašni
t
s [kokašni šo m-t-a?o ma]]
2SgSubj-Dist-Pass-bite snake a 2SgSubj-Real-see DS
‘You were bitten, after you had seen a snake.’
 Grammatical relations across languages 71

Øagent in subordinate passive clause; Øagent in main passive clause;

*DS
d. [[?a t t
s ki? p-a ?-kat
s
t
s (*ta)-x] ?e poL ki? mos
t
s

limberbush the Irr-Pass-seek DS -Aux ratany the also

si-a ?-ka
t
s
t
s?a¼?a]
Irr-Pass-seek Aux¼Decl
‘If limberbush is looked for, white ratany should be looked for also.’
Seri is an accusative AOV head-marking language with agreement/pronominal
markers on verbs for S/A, O, and oblique dependents. Third person S/A and O
agreement is zero. The DS marker (ta in irrealis clauses and ma in realis clauses)
appears in clause-Wnal position of a subordinate clause, possibly aYxed to a
tense auxiliary element, if and only if its ‘subject’ is not the same as the ‘subject’
of the main clause. In (2.43a) the DS marker cannot be used because the A of
the subordinate clause and the S of the main clause both refer to the same 2nd
person. In (2.43b), on the other hand, the DS marker must be used because the
A of the subordinate clause refers to a 3rd person, whereas the S of the main
clause refers to the speaker. Even though the S of the passive main clause refers
to the same singular person as the S of the subordinate clause, a DS marker
must be used in (2.43c) because the implicit agent of the passive clause refers to
a 3rd person. By the same token, a DS marker cannot be used in (2.43d), in spite
of the fact that the Ss of the two passive clauses refer to diVerent 3rd person
entities, because the implicit agents refer to the same person. The privileged
category of dependents for purposes of switch-reference marking is the agent
or the S of an active clause, which can have any semantic role, including that of
patient, as in (2.43a).
Languages such as Seri and Halkomelem, in which the ‘demoted’ agent of a
passive clause has subject properties of various kinds, pose potential challenges
for theories of grammatical relations. Even though languages, by and large,
operate in terms of the same syntactic categories that head-marking and depen-
dent-marking systems recognize, that is A, O, S/A, S/O, O/S0 , and A/Sa , gram-
matical phenomena also appear to work in terms of more complex categories.

2.2.2 Antipassive voice

Passive voice changes the default alignment of semantic roles and syntactic
functions in such a way as to increase the syntactic privileges of the patient, or,
more generally, the dependent that would otherwise be the O. Because the
agent is either not overtly expressed or is realized as an oblique in a passive
clause, the patient has the syntactic function of S rather than O. This increases
its prominence and privileges in a syntactically accusative language, since the
72 Grammatical relations across languages

S/A category is more privileged than the O category. In a syntactically ergative

language, a promotion from O to S has less of an eVect since the privileged
syntactic category (S/O) already encompasses the O. Because the S/O function
typically has more privileges than the A function in ergative languages, there
is more functional motivation for a voice that promotes the default A to S.
The antipassive version of a clause with an agent and patient is intransitive
by virtue of the demotion of the patient to oblique (or, possibly, indirect
object or second object), such that the agent is the S. Schematically, proto-
typical active and antipassive clauses have the following contrasting valence
structures.
(2.44) a. Verbx < Aagent Opatient>
Active Voice
b. Verbx-Antipass <Sagent Oblpatient>
Antipassive Voice
There are various construction types in languages that approximate the
prototypical antipassive construction in one or more ways. In this book, the
properties taken to be criterial are as follows.
(2.45) A clause is said to be in antipassive voice if it
a. is an intransitive clause of a type that functions as a systematic
alternative to some more basic transitive clause type, and
b. the dependent that would be the O in the basic clause type does not
have any A/S/O function.
In Section 2.2.3 it was noted that there are various syntactic phenomena in
Dyirbal that are constrained to occur with S/O dependents. For example, the
controller and pivot in the conjunction-reduction construction must both be
either the S or the O of their respective clauses, as illustrated by (2.11),
repeated here as (2.46).

(2.46) [ŋuma yabu-ŋgu bura-n] [ ∅ banaga-nyu]

father.Abs mother-Erg see-Nonfut S return-Nonfut
‘Mother saw father and he/*she returned.’
In order to say the equivalent of English Mother saw father and returned, yabu
‘mother’ has to be in S function and, therefore, in absolutive case. This can be
accomplished by putting the verb in antipassive voice, in which case
the patient is marked dative and does not function as the O. Since an
antipassive clause is intransitive, the agent is the S and can control conjunc-
tion reduction:
 Grammatical relations across languages 73

(2.47) [yabu bural-ŋa-nyu ŋuma-gu] [ ∅ banaga-nyu]

mother.Abs see-Antipass-Nonfut father-Dat S return-Nonfut
‘Mother saw father and she/*he returned.’
Antipassive voice in Dyirbal systematically allows agents of transitive clauses
to have the S/O syntactic privileges of the language, not only with respect to
conjunction reduction, but also relativization and control of purpose clauses
as discussed in Section 2.2.3.
Languages often have both passive and antipassive voice constructions. This
makes sense primarily because both kinds of construction can have functions
other than increasing the syntactic privileges of the default A or O. As noted
above, the passive construction, for example, also or alternatively functions to
background or lower the grammatical prominence of the agent. The antipassive
construction does the same for the patient.14 Halkomelem is an example of a
language that has both passive and antipassive voices. The following examples
(adapted from Gerdts 1988a) illustrate the contrasting morphological proper-
ties of the alternative ways of saying that the children baked the bread.
(2.48) a. ni q’w@
l-@t-@s tu@ sl’@l?ı́q@L kwu@ s@plı́l


Aux bake-Tr-3A Det children Det bread

‘The children baked the bread.’
b. ni q’w@
l-@m tu@ sl’@l?ı́q@L ?@ kwu@ s@plı́l


Aux bake-Intr Det children Obl Det bread

‘The children baked the bread.’
c. ni q’w@
l-@t-@m ?@ tu@ sl’@l?ı́q@L kwu@ s@plı́l


Aux bake-Tr-Intr Obl Det children Det bread

‘The bread was baked by the children.’
In an active voice clause with agent and patient (2.48a), the verb is marked for
transitivity with the -@t suYx and A agreement with the -@s suYx—if the A is
3rd person. Antipassive voice (2.48b), like passive voice (2.48c), is indicated by
lack of A agreement on the verb and intransitive marking. Passive marking is
distinguished by retention of the transitivity suYx. Furthermore, the agent is
in the oblique function (indicated by the preposition ?@) in the passive clause,
whereas the patient is in the oblique function in the antipassive clause.
As in Dyirbal, one eVect of antipassive is to increase the syntactic privileges
of the agent. As noted in Section 2.2.3, possessor extraction and quantiWer
Xoat are restricted to the S/O dependent. The agent can undergo these
syntactic processes only in antipassive voice (in which case it has the S
function), as illustrated by the following examples for quantiWer Xoat, that
is clause-initial placement of a quantiWer.
74 Grammatical relations across languages

QuantiWer Xoated from O-patient/*A-agent of active clause

(2.49) a. m@
k’w niw q’w@
l-@t-@s tu@ sl’@l?ı́q@L kwu@ s@plı́l


all Aux.Lnk bake-Tr-3A Det children Det bread
‘The children baked all the bread/*All the children baked the bread.’
QuantiWer Xoated from S-agent/*Obl-patient of antipassive clause
b. m@
k’w niw q’w@
l-@m tu@ sl’@l?ı́q@L ?@ kwu@ s@plı́l


all Aux.Lnk bake-Intr Det children Obl Det bread
‘All the children baked the bread/*The children baked all the bread.’
2.2.3 Inverse voice
An inverse-voice system is one in which there are two kinds of transitive
clause, each of which is characterized by a diVerent association of semantic
roles and (at least some of) the morphological markers of the A and O
functions. In the direct voice a given morphosyntactic marker of type M1,
typically a verbal pronominal/agreement aYx, might index, say, the depend-
ent with the agent semantic role, and another marker of type M2 might index
the dependent with the patient role. The inverse voice clause has the align-
ment of roles and marker types reversed, typically with some overt voice
marking, as schematized in (2.50).
(2.50) a. Verbx M1agent M2patient
Direct Voice
b. Verbx-Inverse M1patient M2agent
Inverse Voice
The inverse voice phenomenon can be deWned more generally as follows.
(2.51) A clause is said to be in inverse voice if it
a. is a transitive clause of a type that functions as a systematic
alternative to another more basic transitive clause type,
b. it has the patient marked in the way that the agent is marked in the
more basic clause type, and
c. both the agent and patient have some A/S/O function.
Typically, there are limited paraphrase possibilities in the alternative voices
in an inverse system, since the possibility of inverse vs. direct voice construc-
tions for a given meaning is governed by a person/animacy hierarchy in some
way. The following examples from the Chilean language Mapudungun (from
Arnold 1998: ch. 2) illustrate the basic situation.
(2.52) a. mutrüm-W-n ñi ñuke
call-3O-1SgSubj Pos mother
‘I called my mother.’
 Grammatical relations across languages 75

b. mutrüm-e-n-ew ñi ñuke

call-Inverse-1SgSubj-3O Pos mother
‘My mother called me.’
The dependent with the agent role is coded on the verb with the S/A-indexing
1st person singular suYx in the direct-voice clause in (2.52a); and the
patient is coded with a 3rd person O-indexing suYx. In the inverse-voice
clause in (2.52b) the 1st person patient is marked with the same suYx that is
used for the agent in the direct-voice alternative; and a diVerent O-indexing
suYx is used for the 3rd person agent. The indexing suYx combination
together with the presence of the inverse suYx indicate that the dependent
with the S/A marking in (2.52b) is to be interpreted as having the patient role.
The choice between direct and inverse voice is governed by a prominence
hierarchy of the kind operative in the Kamaiurá verb agreement system in a
somewhat diVerent way (see Section 2.1.3). The hierarchy for Mapudungun is
as follows.
(2.53) 1st > 2nd > 3rd proximate > 3rd obviative
Prominence hierarchy for Mapudungun
As is typically the case in inverse systems, the relative prominence of two
3rd person dependents is determined by obviation (e.g. Aissen 1997), that
is cognitive salience due to discourse factors. In essence, a 3rd person depend-
ent is proximate if its referent is an established topic of the preceding
discourse and otherwise is obviative. The constraint on inverse vs.
direct voice is that the dependent that is S/A-indexed on the verb must be
ranked higher on the prominence hierarchy than the dependent that is
O-marked. Thus, ‘My mother called me’, for example, can only be expressed
in the inverse voice, as in (2.52b), since the 1st person dependent needs to be
S/A-marked.
The Algonquian language family is perhaps the most well known for its
inverse voice system. The main diVerences between Mapudungun and Algon-
quian are that the latter has overt verb suYxes for both direct and inverse
voice, often only has an overt marker for one of the dependents, and draws the
relevant distinction only for a certain class of transitive verbs with animate
dependents. The prominence hierarchy is 1st/2nd > 3rd proximate > 3rd
obviative (Klaiman 1993). The following Plains Cree examples (from Wolfart
and Carroll 1981) illustrate the phenomenon.
(2.54) a. ni-sēki-ā-nān atim
1-scare-Direct-1Pl dog
‘We scared the dog.’
76 Grammatical relations across languages

b. ni-sēki-iko-nān atim
1-scare-Inverse-1Pl dog
‘The dog scared us.’
In (2.54a), the person/number aYxes indicate that this transitive verb has a 1st
person plural dependent. Example (2.54b) has precisely the same person/
number aYxes; but the semantic role they are associated with is diVerent.
The prominence hierarchy dictates that the 1st person dependent gets the
primary, and in this case only, marking on the verb. The direct suYx indicates
that the highest-ranking dependent on the hierarchy (i.e. the 1st person
dependent) has the agent semantic role; the inverse suYx indicates that this
same dependent has the patient role.
This kind of system is characteristic of Algonquian languages in general,
including Algonkin (Henderson 1971), Ojibwa (Rhodes 1976), and Fox
(LeSourd 1976), and has been claimed to exist, in some form, in various
other languages or language families, including Tanoan (Klaiman 1993),
Kinyarwanda (Kimenyi 1980, Ura 2000), and Tzotzil (Aissen 1999a). There
are essentially two analytical approaches to the general phenomenon. On one
approach, adopted, for example by Dahlstrom (1991) for Cree and Klaiman
(1993) for Tanoan and Algonquian, the agent is the A and the patient is the O
in both constructions. The voice morphology simply indicates alternative
interpretations of the indexing morphology. That is, as a function of the
voice, the 1st person plural aYxes in (2.54a), for example, index the A, whereas
in (2.54b), the same aYxes index the O. Since the direct and inverse clauses are
claimed to have the same alignment of syntactic functions and semantic roles
(i.e. agent A and patient O), this kind of analysis involves only inverse
morphology, as schematized in (2.55).

(2.55) Inverse morphology analysis of (2.54)

voice prominence ranking
morphology ni- -nān Ø direct 1Pl A > 3 O (2.54a)
Ø ni- -nān inverse 1Pl O > 3 A (2.54b)
j j
syntax A O
j j
valence sēki <agent, patient>
On the alternative inverse syntax approach, adopted by Rhodes (1976) and
Permlutter and Rhodes (1988) for Ojibwa, Arnold (1997, 1998) for Mapudun-
gun, and (in essence) Ura (2000) for Kinyarwanda, the marking on the verbs
is a direct reXection of (surface) syntactic functions. The A of the Mapudun-
 Grammatical relations across languages 77

gun direct-voice clause in (2.52a), for example, is the agent and the patient is
the O. The alignments of syntactic functions and semantic roles are reversed
in the inverse-voice clause in (2.52b), as schematized in (2.56).
(2.56) Inverse syntax analysis of (2.52)
morphology -n -W
j j prominence ranking (2.52a)
direct voice A O 1Sg agent > 3 patient
j j
valence mutrüm <agent, patient>
j j
inverse voice O A 1Sg patient > 3 agent
j j prominence ranking (2.52b)
morphology -ew -n
In other words, inverse voice is essentially a special kind of passive voice. The
key diVerence between the passive and inverse constructions is that the inverse
agent is ‘demoted’ to O rather than being omitted or realized as an oblique.
This kind of analysis presents a potential challenge for theories of grammar,
which diVer with respect to whether and in what way the A and O functions
are distinguishable from ‘agent’ and ‘patient’. An analysis with A–O reversal
entails that ‘A’ cannot really mean anything like ‘agent’ or ‘most agent-like’. It
has to simply mean ‘highest ranking of the two core syntactic functions of a
transitive clause’.
Although the issue has engendered some controversy, inverse voice has
generally been analysed as a purely morphological phenomenon in Algon-
quian and certain other language families, due to a lack of clear behavioural
diVerences between inverse and direct clauses with respect to S/A-sensitive
syntactic phenomena. In some languages, however, there exists compelling
evidence for a reversal of syntactic functions in inverse clauses. The Amazon-
ian language Jarawara (Dixon 2000), for example, has what can be considered
a kind of inverse construction that illustrates in an interesting way why an
inverse syntax (or reversal) analysis of some kind might be entertained. A/S/O
dependents are distinguished in several ways from obliques. The latter
are expressed more peripherally in the clause and are marked with a
postposition. The former are not postpositionally marked, have special slot-
restricted pronominal forms (except for 3rd singular) that are obligatorily
expressed even if a full NP which they agree with is also expressed, and
determine agreement on verbs and auxiliary verbs in several ways. The
following examples illustrate the diVerences between oblique and A/S/O
dependents.15
78 Grammatical relations across languages

(2.57) a. [otaa nijaa] mee bosa na-meke-hete-ke

1NsgExcl Obl 3Nsg get.up.early Aux-follow-TnsFem-DeclFem
‘Then they got up early on us.’
b. otara mee bosa ka-na-hani
1NsgExclO 3Nsg get.up.early Appl-Aux-TnsFem
‘Then they got-up-early-on us.’
In (2.57a) the oblique maleWciary dependent is marked with a postposition.
Example (2.57b) has a transitive paraphrase derived with the applicative voice
preWx ka- (see Section 2.2.4) that indicates that the maleWciary is O. The S in
(2.57a) and the A in (2.57b) are expressed as an A/S/O pronoun that must
occupy the pronominal position in a transitive clause that is closest to the
verb. In certain cases, depending on person and number, the ‘closest’ pro-
nominal is actually realized as a verbal preWx. The O in (2.57b) is expressed as
an accusative pronoun which must occupy the Wrst pronominal position in a
transitive clause. All pronominal elements are closer to the verb than any
other dependent constituents. Because all of the core dependents in both of
these sentences are pronominal, the gender agreement morphology is the
default feminine.
If both A and O dependents are 3rd person singular, gender agreement,
manifested in the form of the verb and/or a mood suYx is the primary and
often the only morphological manifestation of distinctions among the A and O
dependents, since the pronominal form for 3rd singular subject and object is null.
The following intransitive sentences illustrate S-indexing gender agreement.
(2.58) a. Mioto ki-joma-ke-ka
Mioto(Masc) be.in.motion-through.gap-coming-DeclMasc
‘Mioto (a man) came in.’
b. Watati ki-joma-ke-ke
Watati(Fem) be.in.motion-through.gap-coming-DeclFem
‘Watati (a woman) came in.’
With a transitive verb, there is generally a choice between two constructions
to express the same propositional meaning, provided that both dependents
are not 1st/2nd person:
(2.59) a. Mioto Watati awa-ka
Mioto(Masc) Watati(Fem) see-DeclMasc
‘Mioto saw Watati.’
b. Watati Mioto hi-wa hi-ke
Watati(Fem) Mioto(Masc) Inv-see Inv-DeclFem
‘Mioto saw Watati.’ (or, ‘Watati was seen by Mioto.’)
 Grammatical relations across languages 79

In the construction shown in (2.59a), henceforth the direct voice construction

(¼ Dixon’s ‘A-construction’), gender agreement is with the agent, which
preferably comes Wrst if two A/S/O full NPs are expressed. In the alternative
construction shown in (2.59b), henceforth the inverse voice construction
(Dixon’s ‘O-construction’), gender agreement is with the patient, which
preferably comes Wrst if two A/S/O full NPs are expressed. The choice between
constructions is determined by discourse considerations of the obviation
kind. In essence, the prominence hierarchy for Jarawara is proximate >
obviative. When there is a choice between direct and inverse voices, the
construction which puts the proximate (discourse-salient) dependent of the
verb in the privileged syntactic function is chosen.
Recall that in a language with an accusative syntax such as English
the conjunction-reduction pivot must be the S/A dependent, whereas in a lan-
guage with an ergative syntax, such as Dyirbal, the pivot must be the
S/O dependent (see Section 2.1.2). In Jarawara the dependent that determines
the gender of the mood suYx is also the one that is the pivot in the conjunction-
reduction construction, as illustrated by the following examples.

Direct clause agent controller; Direct clause agent pivot

(2.60) a. [Mioto kijomake-ka] [∅ Watati awa-ka]

Mioto(Masc) come in-DeclMasc Watati(Fem)see-DeclMasc
‘Mioto came in and saw Watati.’

Direct clause agent controller; Inverse clause patient pivot

b. [Watati kijomake-ka] [∅ Mioto hi-wa hi-ke]

Watati(Fem) come in-DeclFem Mioto(Masc) Inv-see Inv-DeclFem
‘Watati came in and Mioto saw (her).’ (or ‘was seen by Mioto.’)
When an inWnitival clause is embedded under the ‘raising’ verb seem in
English (see Section 2.2.1), the S/A of the embedded verb (whether in passive
or active voice) functions as the subject of seem:
Agent of embedded active V = subject of seem

(2.61) a. This drug seems [∅ to help people]

Patient of embedded passive V = subject of seem

b. People seem [∅ to be helped by this drug]

80 Grammatical relations across languages

Similarly, if a clause is embedded under awine/awa ‘seem, in my opinion’

in Jarawara, the latter shows gender agreement with the agent of a direct
clause and the patient of an inverse clause, as illustrated by the following
examples:
Direct verb agent controls gender agreement on ‘seem’

(2.62) a. [Sani boro nofa] awine-ke

Sani(Fem) cake(Masc) like.Fem seem.Fem-DeclFem
‘Sani seems to like cake.’

Inverse verb patient controls gender agreement on ‘seem’

b. [boro Sani hi-nofe] awia-ka

cake(Masc) Sani(Fem) Inv-like.Masc seem.Masc-DeclMasc
‘Sani seems to like cake.’ (or, ‘Cake seems to be liked by Sani.’)

What appears otherwise to be a clear case of syntactic function reversal for

the agent and patient dependents in the alternative voices (or construction
types) is complicated by the fact that the pronominal expressions of syntactic
functions are only partially reversed. For example, a 1st person singular S/A
pronoun is expressed as a preWx on verbs. As the following examples show, the
same pronominal preWx on the verb that is used for the agent/S in an
intransitive or direct transitive clause is used for the agent in an inverse
clause.16
Direct voice; feminine mood suYx because agent is 1st person
(2.63) a. Okomobi o-nofa o-ke
Okomobi(Masc) 1SgSubj-like 1Sg-DeclFem
‘I like Okombi.’
Inverse voice; masculine mood suYx because patient is masculine
b. Okombi o-nofa o-ka
Okomobi(Masc) 1SgSubj-like 1Sg-DeclMasc
‘I like Okombi.’
Inverse voice is indicated in (2.63b) only by the gender agreement on the
mood suYx, since the inverse-voice preWx hi- only occurs when both A and O
are 3rd person. As the following examples show, both the A and O preverbal
pronominal slots are Wlled when neither the agent nor the patient is 3rd
singular.
 Grammatical relations across languages 81

Direct voice; feminine suYxes because both pronominal slots are Wlled
(2.64) a. aba mee(-ra) otaa kaba-haro otaa ama-ke
Wsh(Masc) 3Nsg(-O)1NsgExcl eat-TnsFem 1NsgExcl extent-DeclFem
‘We were eating Wsh.’
Inverse voice; feminine suYxes because both pronominal slots are
Wlled
b. aba mee(*-ra) otaa kaba-haro mee ama-ke
Wsh(Masc) 3Nsg(-O)1NsgExcl eat-TnsFem 3Nsg extent-DeclFem
‘We were eating Wsh.’ (or, ‘Fish was being eaten by us.’)

One generalization concerning pronouns in transitive clauses is that the Wrst

slot is reserved for the patient and the second slot for the agent, independently
of voice. A second is that the pronoun that gets obligatorily copied into the
slot preceding what Dixon calls the ‘secondary’ verb (ama) expresses the agent
in direct voice but the patient in inverse voice. Indeed, the only overt
morphological indication of inverse voice in (2.64b) is the form of the copied
pronoun preceding ama-ke. A third restriction, is that although the Wrst
pronominal slot is occupied by a pronoun expressing the patient in both
inverse and direct voices, if this pronoun is 3rd person it can only take the
morphologically accusative form (i.e. be suYxed with -ra) in the direct clause.
Summarizing, the key morphosyntactic facts about grammatical relations
and voice constructions that an analysis of Jarawara needs to account for are
as shown in Table 2.2. Any analysis has to be able to account for the fact that,
independently of voice, both the patient and agent are associated with the two
core syntactic functions of the language, as evidenced by the fact that in both
voices these determine the content of the two preverbal pronominal slots, are
not marked by postpositions, and occupy the two central NP constituent
positions in the clause. Moreover, in the inverse voice the patient, by and
large, has the syntactic privileges and most of the morphological properties
that the agent has in the direct voice (and that the S has in intransitive
clauses), as indicated by the shaded cells in Table 2.2. At the same time, it is
necessary to account for the fact that the direct and inverse voices treat the
agent and patient in the same way with respect to the preverbal pronominal
slots.
One analysis would take the preverbal pronominal forms as the primary
indicators of syntactic functions. Both voices would be transitive and the
agent would be the A and the patient the O in both. The diVerence between
the voices could be attributed to a dual syntax: accusative with respect to
direct voice and ergative with respect to inverse voice. That is to say, the agent
is syntactically privileged in the direct voice because it is in the S/A (nomina-
82 Grammatical relations across languages

Table 2 .2. Jarawara voice and grammatical-relation phenomena

Direct voice Inverse voice

Grammatical phenomena Patient Agent Patient Agent

Preverbal pronominal slot 1st 2nd/preWx 1st 2nd/preWx

Full NP order preference 2nd 1st 1st 2nd
Conjunction-reduction pivot no yes yes no
Verb/mood gender agreement no yes yes no
Agreement on ‘seem’ no yes yes no
Pronoun preceding secondary V no yes yes no
3NsgAcc pronominal form yes no no no

tive) category; the patient is syntactically privileged in the inverse voice

because it is in the S/O (absolutive) category. The problem with this ap-
proach, which is the one taken by Dixon, is that it necessitates disjunctive
characterizations of most of the key phenomena. That is to say, the rules for
constituent order, conjunction reduction, and verb/mood gender agreement,
for example, would have to be formulated as follows:
(2.65) a. The preferred constituent order is AOV in the direct voice and OAV
in the inverse voice.
b. The conjunction-reduction pivot is the S/A dependent in the direct
voice and the O dependent in the inverse voice.
c. Verb/mood gender agreement is controlled by the S/A dependent in
the direct voice and the O dependent in the inverse voice.
Under this analysis, Jarawara is a typological anomaly. Not only is it characterized
as alternating between two basic kinds of syntactic organization (i.e. ergative and
accusative) with respect to the same phenomena, but it also has two diVerent basic
word orders: AOV (when accusative) and OAV (when ergative).
An alternative, which avoids positing such unusual typological properties,
is to analyse the language as fundamentally accusative with the inverse voice
associated with syntactic-function reversal. Although A/O reversal is itself
rare, it does plausibly occur in other languges, as noted above, and simply
amounts, in some sense, to a variation on the passive construction. It is clear
that the inverse voice is more marked morphologically (as there is a verbal
preWx for it in clauses with 3rd person dependents), is less commonly used
(only 30 per cent of the transitive clauses in Dixon’s textual sample), and
serves one of the key functions of passive clauses in accusative languages
(increasing the prominence and privileges of the patient). On the syntactic
inversion analysis, the agent and patient reverse their alignments with the
 Grammatical relations across languages 83

A and O categories, as illustrated by the following analysis of the semantic

role/syntactic function alignments of the examples in (2.59).
agent patient Direct voice
A O

(2.66) a. Mioto Watati awa-ka

Mioto(Mas) Watati(Fem) see-DeclMasc

‘Mioto saw Watati.’
patient agent Inverse voice
A O

b. Watati Mioto hi-wa hi-ke

Watati(Fem) Mioto(Masc) Inv-see Inv-DeclFem

‘Mioto saw Watati.’ (or, ‘Watati was seen by Mioto.’)
Given this kind of an inverse syntax analysis of the inverse-voice construction,
the generalization about preverbal pronominals is that they are determined
directly by the agent vs. patient distinction. Rather than being formulated in
voice-speciWc ways, the other constraints can also be characterized as working
for the language as a whole:
(2.67) a. The preferred constituent order is AOV.
b. The conjunction-reduction pivot is the S/A dependent.
c. The S/A dependent controls verb/mood gender agreement.
Thus, it is possible to characterize Jarawara as a mostly unremarkable accusa-
tive language whose grammatical phenomena work in terms of the categories
S/A and O. It is necessary, however, to recognize an inverse-voice construction
with some kind of A–O reversal and to be able to formulate morphosyntactic
constraints in terms of either ‘underlying’ A and O categories or the kinds of
generalized semantic roles that are implicitly envisioned here in the use of the
terms agent and patient and which have more precise characterizations in
certain theories (e.g. the actor and undergoer macroroles of Role and Refer-
ence Grammar, discussed in Section 4.1.2, or the proto-agent and proto-
patient categories of Dowty 1991).

2.2.4 Applicative constructions

Passive and inverse voices can function to give S/A syntactic privileges to
dependents whose default syntactic function is O. Antipassive voice can
function to give S/O syntactic privileges to a dependent whose default syn-
84 Grammatical relations across languages

tactic function is A. Applicative constructions, which could be viewed as

instantiating a kind of voice, diVer primarily in that they target the O
syntactic function, giving O or S/O syntactic privileges to dependents that
are neither the default O nor the default A. Schematically, the valence
structures of a prototypical applicative and a corresponding basic construc-
tion are as in (2.68); the applicative phenomenon can be deWned as in (2.69).17
(2.68) a. Verbx <S/Aagent . . . >
Basic construction
b. Verbx-Appl <Aagent Onon-patient . . . >
Applicative construction
(2.69) A clause is said to be of the applicative type if
a. it is a transitive clause that contains an overt marker to distinguish
it from other clause types and/or is a systematic alternative to some
other available clause type, and
b. a non-patient dependent has the O function.
In Halkomelem (see Section 2.1.2 and Gerdts 1988a) such key grammatical
phenomena as relativization, clefting, and wh-question formation, grant
special privileges to dependents with any of the S/A/O functions. The S/O
category is privileged for other phenomena, including quantiWer Xoat and
possessor extraction. Given this system, it is not surprising that Halkomelem
has both a way of getting the agent in the S category (i.e. an antipassive
construction) and ways of getting dependents with various semantic roles in
the O category. The applicative construction is characterized by a verbal aYx
signalling that a non-patient has the O function, and the patient, if there is
one, does not. Since S/O is the most privileged syntactic function, ‘promo-
tion’ to O in an applicative construction is essentially functionally equivalent
to promotion to S via passive in an accusative language.
Halkomelem has four distinct applicative clause types, as illustrated by the
following examples.18
(2.70) a. ni ?ám-@s-t-@s kwu@ sqw@m@
y? ?@ kwu@ su’ám?
Aux give-Recip-Tr-3A Det dog Obl Det bone
‘He gave the dog the bone.’
b. ni q’w@
l-@Lc-t-@s L@ sLéni ?@ kwu@ s@plı́l
Aux bake-Ben-Tr-3A Det woman Obl Det bread
‘He baked the bread for the woman.’
c. ni u’weyk’w-mé?-t-@s kwu@ sqw@m@
y?
Aux startle-Stim-Tr-3A Det dog
‘He was startled by the dog.’
 Grammatical relations across languages 85

d. ni y@?é?w@?-n-@s-@s L@ sLéni
Aux come-Goal-Tr-3A Det woman
‘He came toward the woman.’
Examples such as (2.70a–b) are, of course, similar to the English double-
object construction (see Section 1.3.2), except with a recipient or benefactive
suYx on the verb. One other diVerence is that the recipient and beneWciary
objects are harder to analyse as indirect objects in Halkomelem because the
patient is clearly an oblique, or at least a non-object, by multiple criteria and
the recipient and beneWciary dependents are therefore the only objects. That
they are indeed members of the O category is indicated, among other things,
by the transitive suYx on the verbs, the A 3rd person suYx on the verb
(indicating that these are transitive clauses and that there is thus an O), and
the fact that, if they were 1st or 2nd person, the recipient and beneWciary
would be indexed like an O on the verb, with an object pronominal suYx, as
shown in (2.71), which summarizes the meanings of the full array of potential
morphological markers.

(2.71) O = BEN V HAS A & O O = 1ST PERSON A = 3RD PERSON PATIENT=OBLIQUE

ni θ y - c -θ -ám?s - s
e e e ?
e kwθ n -sn xw
e e e e

Aux fix -Ben-Tr-1O -3A Obl Det 1Pos-canoe

‘He fixed my canoe for me.’
Halkomelem also diVers from English in that it does not have an alternative
construction in which the recipient and beneWciary can be expressed as an
oblique (or indirect object). What this means is that the most basic transitive
clause type, which has the patient as O, is systematically ruled out for a verb
with a recipient or beneWciary dependent. This appears to reXect a promin-
ence hierarchy constraint that is, at some level of analysis, analogous to what
happens with verbs that have a 3rd person agent and a 1st person patient in
Southern Tiwa (see Section 2.2.1). In Southern Tiwa the potential active-
clause realization of such a verb is ruled out, because the A cannot be
outranked in terms of person prominence by the O. In Halkomelem the
beneWciary and recipient dependents appear to always be animate and the
patient appears to be typically inanimate. As Gerdts notes, there is an animacy
hierarchy eVect operative with the O category that is also manifested with the
stimulus and goal applicative constructions illustrated by (2.70c–d). Goal and
stimulus dependents can be alternatively expressed as obliques—and indeed
have to be if they are inanimate:
86 Grammatical relations across languages

(2.72) a. ni c@n c’@
q’ ?@ kwu@ sx̌t’@
k’w-s (Inanimate oblique)

Aux 1Subj astonished Obl Det carving-3Pos
‘I was astonished at his carving.’
b. ?*ni c@n c’@
q’-mé?-t kwu@ sx̌t’@
k’w-s (?*Inanimate O)
Aux 1Subj astonished-Stim-Tr Det carving-3Pos
‘I was astonished at his carving.’
c. ?? ni c@n c’@
q’ ?@ kwu@ sqw@m@
y? (??Animate oblique)
Aux 1Subj astonished Obl Det dog
‘I was astonished at the dog.’
d. ni c@n c’@
q’-mé?-t kwu@ sqw@m@
y? (Animate O)
Aux 1Subj astonished-Stim-Tr Det dog
‘I was astonished at the dog.’
(2.73) a. ni ném ?@ kwu@ stá?lu? (Inanimate oblique)
Aux go Obl Det river
‘He went to the river.’
b. *ni ném-n-@s-@s kwu@ stá?lu? (*Inanimate O)
Aux go-Goal-Tr-3A Det river
‘He went to the river.’
c. ni ném ?@-l’ John (Animate oblique)


Aux go Obl-Det John

‘He went to John.’
d. ni ném-n-@s-@s kwu@ John (Animate O)
Aux go-Goal-Tr-3A Det John
‘He went to John.’
The situation with syntactic functions for non-patient dependents of a verb
can be summed up as follows. Inanimate NPs cannot be an O, whereas
animate NPs are preferentially expressed as an O, with the strength of the
preference being determined by the inherent animacy of their semantic role
(Rec/Ben > Stim > Goal). That is to say, the prototypical goal is a place rather
than an animate participant; the recipient and beneWciary roles are necessarily
associated with animate participants; and the stimulus role is neutral with
respect to animacy. Ensuring that dependents with the higher cognitive
salience associated with animacy have higher grammatical salience, and,
therefore, maximal syntactic privilege, appears to be the main functional
motivation for the system. Since the semantic role of the O can vary consid-
erably, the applicative morphology on the verb serves to provide role infor-
mation in the same way that prepositions and case markers do for
constituents in dependent-marking languages.
 Grammatical relations across languages 87

Of crucial importance to understanding the grammatical correlates of

applicative verbs in Halkomelem is that with respect to the phenomena for
which only A/S/O dependents have special privileges, the oblique-marked
patient fails to behave like an O, unlike the non-patient O-marked depend-
ents, which show most O properties. For example, as noted in Section 2.1.2, a
quantiWer such as ‘all’ can only be Xoated (i.e. separated from the NP that it is
associated with) from an absolutive (S/O) dependent and only a core depend-
ent (A/S/O) can be put in contrastive focus in the cleft construction. The
beneWciary rather than the patient behaves like an O with respect to quantiWer
Xoat in (2.74a). The contrast between (2.74b) and (2.74c) shows that the
recipient—not the patient—behaves like a core dependent with respect to
clefting.
(2.74) a. m@
k’w niw ?ı́l@q-@Lc-t-?e n?
t
s kwu@ n@-mé?m@n?@ ?@ kwu@
all Aux.Lnk buy-Ben-Tr-1SgSubj Det 1Pos-oVspring Obl Det
q L@y?š@n?
w


shoe
‘I bought all my children shoes/*all the shoes for my children.’
b. nı́L kwu@ Bob ni xwày@m-@s-t-?e n? ?@ kwu@ n@-lél@m-?@
t
s

3Emph Det Bob Aux sell-Rec-Tr-1SgSubj Obl Det 1Pos-house-Pst

‘It was Bob that I sold my house to.’
c. *nı́L kwu@ n@-lél@m-?@ ni xwày@m-@s-t-?e n? kwu@ Bob
t
s

3Emph Det 1Pos-house-Pst Aux sell-Rec-Tr-1SgSubj Det Bob

‘It was my house that I sold to Bob.’
The non-patient dependent in every type of applicative construction shows
parallel properties. It behaves like an O rather than an oblique, not only with
respect to quantiWer Xoat and clefting, but also with respect to possessor
extraction, relativization, and wh-question formation.19
Although the non-patient objects have many of the O properties of the
language, they do diVer from patient Os with respect to certain phenomena,
such as reXexivization. If the agent and the patient are co-referential, a
person-neutral reXexive suYx is added to the verb, making it intransitive, as
shown in (2.75a–b). If a non-patient argument (normally expressed as an
oblique) is co-referential with the agent, the reXexive suYx cannot be added
to the verb (2.75c); instead, an emphatic oblique-marked pronominal form
must be used (2.75d).
(2.75) a. ni c @n xı́q’-@u@t
Aux 1Subj scratch-ReX
‘I scratched myself.’
88 Grammatical relations across languages

b. ni q’áy-u@t
Aux kill-ReX
‘He killed himself.’
c. *ni c @n čxwmé m?-u@t
t
s

Aux 1Subj speak-ReX

‘I spoke about myself.’
d. ni c@n čxwmé m? ?@-l’
t
?é nu@
t


s s

Aux 1Subj speak Obl-Det 1Emph

‘I spoke about myself.’
In all of the applicative constructions, the O-marked dependent does not
behave like a patient with respect to reXexivization. The reXexive suYx cannot
be added to an applicative verb, as shown for the benefactive case in the
following example.

(2.76) *ni c@n u@
y-@Lc-u@t ?@ kwu@ sn@
xw@L

Aux 1Subj make-Ben-ReX Obl Det canoe
‘I made myself a canoe.’
Non-patient Os also fail to behave like patient Os with respect to recipro-
calization, which works in essentially the same way as reXexivization, but with
a distinct suYx, and antipassive voice, which eVectively demotes patients to
oblique (see Section 2.2.2).
Although it has its own unique mix of complexities, which pose interesting
analytical challenges, Halkomelem displays properties that are not atypical
of applicative constructions cross-linguistically. In an applicative clause, a
non-patient is marked like a member of the O category and takes on many of
the syntactic privileges of the O category. In cases where the verb also has a
patient dependent, it has none of the O privileges and is marked like an
oblique. Halkomelem is, thus, an applicative language of an asymmetric type
(Bresnan and Moshi 1990, McGinnis 2002), like Chichewa (Alsina and
Mchombo 1990), Tzotzil (Aissen 1983), Chamorro (Gibson 1980), and many
other languages, including (many dialects of) English, whose double-object
construction is, from a typological perspective, a kind of applicative con-
struction, modulo the possible interpretation of the O1 as an ‘indirect object’
rather than O (see Section 1.3.2). In languages of the symmetric type, such as
Kinyarwanda (Kimenyi 1980, Dryer 1983) and other Bantu languages, such
as Chi-Mwi:ni (Kisseberth and Abasheikh 1977) and Kichaga (Bresnan and
Moshi 1990), both the patient and the non-patient dependents are bare NPs and
behave syntactically like an O in most ways (at least with certain combinations
of patient and non-patient). Kinyarwanda can have as many as three NPs with
 Grammatical relations across languages 89

the morphosyntactic properties and privileges of an O in one clause. The

examples in (2.77) and (2.78) (from Kimenyi 1980) illustrate the phenomenon.
With verbs such as ‘give’ and ‘ask’, the recipient/addressee cannot be an oblique
and does not require applicative marking, as shown in (2.77a) and (2.78a). An
instrumental can be an oblique (2.77a); but can also be an O, in which case an
instrumental marker must appear on the verb (2.77b). BeneWciaries can be
added relatively freely to clauses but cannot be oblique and require an appli-
cative suYx, as shown in (2.78a). Examples (2.77b) and (2.78a) show that there
can be as many as three non-subject bare NPs, all of which have such A/S/O or
O-restricted privileges as being able to be relativized, to be the subject in a
passive clause, and to be realized as an object pronominal preWx on the verb, the
latter possibility being illustrated by (2.78b).
(2.77) a. Úmwáana y-a-sab-ye umugóre ibı́ryo n’ı̂ntoki
child Subj-Pst-ask-Asp woman food with.hands
‘The child asked the woman for food with his hands.’
b. Úmwáana y-a-sab-iish-ye umugóre ibı́ryo intoki
child Subj-Pst-ask-Instr-Asp woman food hands
‘The child asked the woman for food with his hands.’
(2.78) a. Umugóre a-rá-hé-er-a umugabo ı́mbwa ibı́ryo
woman Subj-Pres-give-Ben-Asp man dog food
‘The woman is giving food to the dog for the man.’
b. Umugóre a-rá-bi-yı́-mu-he-er-a
woman Subj-Pres-it-it-him-give-Ben-Asp
‘The woman is giving it to it for him.’
Languages with applicative constructions also diVer with respect to the
range of non-patient dependents that can be expressed as an O. The English
applicative O is restricted to recipient-type semantic roles, as is the case for
Korean (Park 2002), for example. Bantu languages, particularly Kinyarwanda,
on the other hand, allow applicative Os to have a large range of semantic roles,
including not only recipient and beneWciary, but also instrumental, locative,
possessor of patient, and even ‘adverbial’ or adjunct roles of certain kinds.
Other languages fall in between these extremes in various ways, although
recipient applicative Os seem to always be at least one of the options. The use
of overt verbal applicative marking also varies considerably. Languages like
English and Korean, with only a kind of recipient applicative construction,
seem generally not to have a morpheme dedicated to announcing the
construction, whereas languages such as Kinyarwanda and Halkomelem,
with various kinds of non-patient Os tend to favour overt marking. Whereas,
Halkomelem has markers for recipient, benefactive, stimulus, and goal
90 Grammatical relations across languages

applicative Os, Kinyarwanda has distinct markers for instrumental, goal, and
benefactive Os and no overt marking for recipient Os. Other Bantu languages,
such as Chichewa, only have a single marker.
Partially independently of their syntactic behaviour, patient and non-
patient dependents in applicative constructions can be marked in diVerent
ways. In English, Korean, and Bantu languages they both look like objects,
that is they are bare NPs in English and Bantu and are both marked with an
accusative case morpheme in Korean. In Chamorro and Halkomelem only the
non-patient is a bare NP; the patient is marked with a general-purpose
oblique preposition. However, even in languages that mark both in the
same way, the applicative O never appears to be outranked by the patient O
in terms of grammatical prominence, as evidenced by ability to participate in
phenomena that are restricted to Os or dependents in the A/S/O categories.
Finally, languages diVer with respect to whether certain non-patient se-
mantic roles are obligatorily expressed as an O in an applicative construction
or not. In Halkomelem, as discussed above, animate non-patients other than
goal cannot be realized as obliques. In Tzotzil (Aissen 1983), recipients can
only be an applicative O, whereas beneWciaries can either be an oblique/
adjunct or an applicative O. In Chamorro, Korean, and English (for the
most part), applicative Os of all types have an alternate oblique (or indirect
object) realization. If there is a regularity, it seems to be that more animate or
more cognitively salient event participants (e.g. recipient/beneWciary > loca-
tion/instrument) are more likely to be restricted to an O realization.
Summarizing, some of the various possibilities for applicative construc-
tions and some languages instantiating them are as in Table 2.3.20

Table 2 .3. Some variable features of applicative constructions

A B C D E

Kinyarwanda yes some 3 yes yes

Kichaga yes some 1 yes yes
Chichewa no some 1 yes yes
Halkomelem no some 4 yes no
Tzotzil no some 1 no no
Chamorro no yes 1 yes no
English no yes 0 no yes
Korean no yes 0 no yes

Legend: A: symmetric; B: oblique/O alternations; C: verbal morphemes; D: recip/ben O þ other non-patient Os; E:
patient marked like O.
 Grammatical relations across languages 91

2.2.5 The Philippine voice system

Austronesian languages such as Tagalog, Cebuano, and Ilokano have often
been analysed as having a unique voice system. The reason for this is that
these languages generally have one primary privileged dependent in each
clause and multiple voices (each with an associated verbal morphology),
whereby dependents with various semantic roles can have the privileged
function. The morphological form of the verb identiWes, at least roughly,
the general semantic role of the privileged NP. Although there are clear
parallels with applicative-voice languages such as Halkomelem, discussed
above, and, as noted below, it is possible to treat the Philippine system as
one with some combination of applicative constructions and passive or
antipassive constructions that function to promote various dependents to
either O or S, the problem is that it is unclear that the primary privileged
function is appropriately characterized as S/O. The basic system can
be illustrated with the following Tagalog examples, adapted from Kroeger
(1993).

(2.79) a. bumili ang¼lalake ng¼isda sa¼tindahan

buy.Perf.Act ª¼man Gen¼Wsh Loc¼store
‘The man bought Wsh at the store.’
b. binili ng¼lalake ang¼isda sa¼tindahan
buy.Perf.Obj Gen¼man ª¼Wsh Loc¼store
‘The man bought the Wsh at the store.’
c. binilhan ng¼lalake ng¼isda ang¼tindahan
buy.Perf.Loc Gen¼man Gen¼Wsh ª¼store
‘The man bought Wsh at the store.’
d. ipinambili ng¼lalake ng¼isda ang¼pera
buy.Perf.Instr Gen¼man Gen¼Wsh ª¼money
‘The man bought Wsh with the money.’
e. ibinili ng¼lalake ng¼isda ang¼bata
buy.Perf.Ben Gen¼man Gen¼Wsh ª¼child
‘The man bought Wsh for the child.’
NPs are marked for case with one of three markers, for which diVerent
labels are used by diVerent analysts. Although there are various uses for the
sa case marker, it is generally associated with semantic roles such as location,
source, goal, and recipient, for which reason I use the gloss ‘Loc(ative)’,
following Foley and Van Valin (1984). The privileged NP, whose general
semantic role is indicated by the voice marking on the verb, is marked with
ang, which has been glossed in various ways by diVerent analysts, including
92 Grammatical relations across languages

‘nominative’, ‘focus’, and ‘absolutive’. However, since a particular analysis of

the construction is implicit in the usual choices, I simply use the arbitrary
symbol ª, which is meant to indicate only that this is the special or privileged
dependent. Dependents of verbs which are not marked with ang or sa are
marked with ng. I follow Kroeber and others in calling this ‘Gen(itive)’, since
it is also used to mark the possessor relation within NPs. It is not uncommon
in languages for the possessor-marking device to be extended to various
clause-level dependency relationships. In these examples, the verb root is
bili. Various combinations of inWxes, preWxes, and suYxes are used to indicate
aspect and voice. I make no attempt here to segment and individually
label the diVerent components of the verb stem. The key fact is that the
form of the verb correlates with which dependent is marked with ang.
Although there are idiosyncrasies of various kinds, the general system can
be summarized as follows. Active voice is used with an ang-marked agent
(2.79a). Objective voice is used with an ang-marked patient (2.79b). Locative,
instrumental, and benefactive voices are used with ang-marked NPs with a
generalized locative, instrumental, or beneWciary role, respectively (2.79c–e).
For the most part, grammatical relations are indicated by the combination of
information from the voice marking on the verb and the sparse case marking
information. Although the NP constituents can appear in various orders,
there are certain preferences, such as agent Wrst and, otherwise, ang-marked
NP last.
The phenomenon known as quantiWer Xoat illustrates how the ang-marked
dependent is the privileged one. Although certain quantiWers, such as lahat
‘all’, normally appear within the NP that they modify, for purposes of em-
phasis they can also be placed right after the verb, dislocated from their
associated NP. As the following examples show, in every voice lahat must be
interpreted as modifying the ang-marked NP (or ª dependent).
(2.80) a. sumusulat lahat ng¼mga¼bata ang¼mga¼liham
buy.Imperf.Obj all Gen¼Pl¼child ª¼Pl¼letter
‘The/some children are writing all the letters.’
* ‘All the children wrote the letters.’
b. sinusulat lahat ang¼mga¼bata ng¼mga¼liham
buy.Imperf.Act all ª¼Pl¼child Gen¼Pl¼letter
‘All the children are writing letters.’ Not: ‘all the letters’
c. binigyan lahat ng¼mga¼guro ng¼pera ang¼mga¼child
give.Perf.Loc all Gen¼Pl¼teacher Gen¼money ª¼Pl¼child
‘(The) teachers gave money to all the children.’
Not: ‘all the teachers/money.’
 Grammatical relations across languages 93

Relativization also works in terms of the same privileged constituent.

A relative clause in Tagalog is simply a clause with a missing constituent
that modiWes the NP that it immediately follows. The modiWed NP controls
the interpretation of the missing NP, as shown by the following examples.
(The ni and kay case markers are the allomorphs of ng and sa, respectively,
that are used with proper names.)

(2.81) a. ang=bata=ng [nanuksu ∅ kay=Josie]

= child=Lnk tease.Perf.Act Loc=Josie
‘The child that teased Josie.’

b. ang=bata-ng [tinuksu ∅ ni=Josie]

= child=Lnk tease.Perf.Obj Gen=Josie
‘The child that Josie teased.’
If the relative clause is in the active voice, the gap in the relative clause is necessarily
interpreted as the agent, which is the ª dependent, as shown by (2.81a). If the verb
of the relative clause is in any other voice, the gap must be interpreted as the ª
dependent in that voice, as shown for objective voice in (2.81b).
Other phenomena that are restricted to the ª dependent, like quantiWer
Xoat and relativization, include control of secondary predication, possessor
raising, number agreement, raising to object from an embedded clause, and
conjunction reduction. There are also phenomena that are sensitive to other
categories of dependents. Consider, for example, control of participial com-
plements of verbs such as ‘leave’, ‘catch’, and ‘Wnd’, which in Tagalog take the
form of a clause with a verb in a non-completive aspect and a ‘missing’
dependent, as illustrated by the following examples.

(2.82) a. iniwanan ko siya=ng

leave.Imperf.Loc 1SgGen 3Sg=Comp
[sumusulat ∅ ng=liham]
write.Imperf.Act Gen=letter
‘I left him writing the letter.’

b. iniwanan ko siya=ng
leave.Imperf.Loc 1SgGen 3Sg=Comp
[sinususulat ∅ ang=liham]
write.Imperf.Obj Gen =letter
‘I left him writing the letter.’
94 Grammatical relations across languages

c. inabutan ko si= Manuel

catch.Imperf.Loc 1SgGen = Manuel
na [hinahalikan ng=katulong ∅]
Comp kiss.Imperf.Obj Gen=maid
‘I caught Manuel being kissed by the maid.’
Although the controller has to be the ª dependent, the controllee can either be
the agent or the ª dependent. The controllee is the ª dependent and the agent
in an active voice clause in (2.82a), the agent and genitive dependent in an
objective voice clause in (2.82b), and the patient and ª dependent in an
objective voice clause in (2.82c). Under no circumstance can the controllee be
anything but an agent or the ª dependent. A similar constraint is operative with
respect to the controller of a participial adjunct construction (Kroeger 1993: 42).
As discussed in Section 1.3.3, there is a so-called adjunct-fronting construc-
tion that privileges the following class of constituents: adverbials, prepos-
itional phrases (expressing, for example, causal, instrumental, or beneWciary
roles), and NPs marked with locative case. The class of constituents excluded
from adjunct fronting consists of the ª dependent and all NPs marked with
genitive case, including some NPs with the instrumental role, agents with
verbs not in the active voice, and patients with verbs not in the objective voice.
With this in mind, the grammar of Tagalog seems to operate with the
syntactic categories shown in Figure 2.2.
The classiWcation of the ª dependent is the key question in some sense.
One approach, worked out in some detail for the very similar grammar of
Cebuano in Bell (1976, 1983; see also BloomWeld 1917, Perlmutter and Postal
1984) takes the ª dependent to be S/A, or subject, in an accusative language,
with all voices other than active being variations on passive voice. On the
accusative analysis, the ª dependent category is more speciWcally that of Wnal
(or surface) subject, whereas the a category in Figure 2.2 is that of any kind of
subject (Wnal or initial/logical). The genitive marking with the b category is
attributed to the multifunctionality of genitive case (¼ accusative case,
demoted agent case, instrumental case, etc.). It is unclear how the g category
might be accounted for.
Another approach takes languages of this sort to be basically ergative
languages (e.g. Dryer 1978, Payne 1982, De Guzman 1988, Cooreman et al.
1988, Gerdts 1988b, Blake 1990: ch. 7, Farrell 1994a: ch. 4). The ª dependent is
the S/O or absolutive NP. In all transitive clauses with voices in which it is
marked genitive, the agent is simply the A. The active voice is actually
antipassive voice (see Section 2.2.2), with the patient demoted, such that the
 Grammatical relations across languages 95

agent is the S. The instrumental, benefactive, and locative voices are simply
applicative constructions (see Section 2.2.4),21 with the patient demoted such
that NPs with various semantic roles can be the O. Under this approach, the
phenomena restricted to the ª dependent, are restricted to the S/O category,
the a category in Figure 2.2 is that of A/S/O. The genitive marking with the b
category is attributed to the multifunctionality of genitive case (¼ ergative
case, demoted patient case, instrumental case, etc.). It is unclear how the
g category might be accounted for.
A third general approach takes languages of the Tagalog kind to be neither
accusative nor ergative. There is one main distinction: ª vs. other depend-
ents. The various voices are, in some sense, diVerent varieties of an inverse
voice of the kind suggested for Jarawara in Section 2.2.3, inasmuch as the
agent is expressed as a core constituent rather than an oblique in the voices in
which it is not the ª dependent. The ª dependent is called either the topic
(Schachter 1976), the subject (Kroeger 1993), or the pragmatic pivot (Foley
and Van Valin 1984). Relativization, quantiWer Xoat, raising to object, etc., are
constrained to work with the subject or pragmatic pivot (¼the ª dependent
as indicated by the voice of the verb). All other constituents of the clause are
either core arguments (¼the b category in Figure 2.2) or non-core constitu-
ents, including obliques and adjuncts (¼the g category). Adjunct fronting
works only for non-core constituents. Genitive marking is used for non-
subject core dependents of the verb. The category a is that of agent/S/A.
In any case, whether Philippine languages are ultimately amenable to a
standard accusative or ergative analysis or not, the complexity of the voice
system and its interaction with case marking and the unique ways in which
the categories of syntactically privileged constituents are constituted pose
interesting challenges for typologies of grammatical relations and theories
of syntax.

relative clause pivot; quantifier float pivot;

number agreement controller; controller of secondary predicate;
conjunction-reduction pivot; raising-to-object pivot

non- patient PP
dependent non- agent instrumental NP locative NP
α β γ

controller of genitive marking adjunct fronting

participial adjunct;
controllee in participial complement

Figure 2.2. Tagalog syntactic categories

96 Grammatical relations across languages

2.3 Function splitting and quasi-subjects/objects

As should be clear from the discussion in the previous sections of voice con-
structions and the typological parameters of grammatical-relation marking
systems, deWning the role categories that the grammars of languages operate
with is a non-trivial matter. In an ideal system, the head-marking, dependent-
marking, and constituent-order devices of a language would identify a uni-
form set of grammatical relations in a consistent way and syntactic phenom-
ena such as relative clause formation, quantiWer Xoat, control, raising, and
conjunction reduction would operate straightforwardly in terms of the same
categories. Natural languages, however, tend to be more complex—presum-
ably for good reasons that remain somewhat obscure. Often, alternative
relation-marking devices mark the same dependent of a verb as having
diVerent grammatical relations. In Choctaw, for example, under certain
conditions an NP that is case-marked nominative (S/A) can be indexed on
the verb with the aYx that is used for the O of a transitive clause (see Section
2.1.3). Morphologically ergative languages, in fact, generally employ relation-
marking systems that are partially at odds with each other (see Section 2.1.2).
Even more remarkably, in one voice construction in Jarawara the preverbal
pronominal indexing mechanism seems to identify as A and O the same
dependents that the constituent-order and gender-agreement mechanisms
identify as O and A, respectively (see Section 2.2.3). It seems, however, that
when all of the marking information is taken into account and interpreted, a
coherent system of grammatical relations can invariably be discerned, al-
though sometimes there may be viable alternative analyses. Even so, the
syntactic processes of a language may not operate entirely in terms of the
same system of grammatical relations. For example, in Kamaiurá (see Section
2.1.3) the morphological marking devices indicate that the main categories are
A/Sa and O/So, with subjects of intransitives splitting into two distinct classes.
For the purposes of relative clause formation, however, there are diVerent
strategies to relativize A, S, O, and oblique functions. Thus, S is treated as a
uniform category, distinct from both A and O. Similarly, possessor raising is
constrained to occur with S/O rather than just O/So dependents.
An even more complex issue is that grammatical phenomena within and
across languages work in terms of complex and hybrid categories of grammat-
ical relations, some of which are quite diVerent than those that the primary
marking systems ever indicate. In Halkomelem (see Sections 2.1.2 and 2.2.1),
although certain grammatical phenomena are constrained to occur with such
syntactic functions as S/O or any A/S/O function, which the marking devices of
the language explicitly indicate, others are not. For example, the pivot in the
 Grammatical relations across languages 97

raising to object construction must be either the S/A of the embedded clause or
the passive oblique agent, suggesting that the apparently oblique agent is
subject-like to some extent and that the category of relevance to the phenom-
enon is diVerent from and more complex than any of the categories that the
head-marking and dependent-marking devices indicate. Similarly, in Tagalog
(Section 2.2.5) most syntactic phenomena are restricted to the ang-marked
dependent (generally considered to be the S/A or S/O). However, the controller
of participial adjuncts, for example, can be either the ang-marked dependent or
a genitive agent. It turns out that, across languages, demoted agents frequently
have at least some S/A properties. Thus, a subject-sensitive function can some-
times be split between two diVerent dependents. In what follows, the general
kind of a dependent that has a restricted subset of the morphosyntactic prop-
erties of a subject, is referred to as a quasi-subject. This term is not meant to
designate a particular grammatical relation; it is merely a theory-independent
and analysis-neutral label for a type of dependent that has subject syntactic
privileges to some limited extent.
Similarly, the ‘demoted’ patient in applicative constructions (see Section
2.2.4) is O-like to varying degrees across languages. In certain Bantu lan-
guages, such as Kinyarwanda, it generally has both the morphological mark-
ing and syntactic privileges of an O. In Halkomelem, it appears to have no O
properties at all, although it also diVers partially from similarly marked non-
patient obliques. In English, on the other hand, although the patient (or
second object) in the double-object construction is not O-like in that it
cannot be placed right-adjacent to the verb and generally cannot be the
subject in a passive clause (at least for most speakers—see Section 1.3.2), it
is like a prototypical patient O in that it is not marked by a preposition and,
unlike oblique complements (2.83b) and like Os (2.83a), it can, at least under
certain circumstances (e.g. Bolinger 1971: ch. 4), precede a preposition func-
tioning as a so-called particle (i.e. a discontinuous part of a compound or
‘phrasal’ verb), as shown by (2.83c).
(2.83) a. I want you to give the exams back to them.
b. *I want you to give the exams to them back.
c. I want you to give them the exams back.
Although there are various analyses and labels that have been used for the
second object in this construction, it is, from a descriptive and typological
perspective, a quasi-object, since it has a restricted set of the O-deWning
properties of the language.
Even outside of the standard voice and ‘relation-changing’ constructions,
in many languages the dependents of certain verb classes can be marked in
98 Grammatical relations across languages

ways that are at odds to varying degrees with what their overall grammatical
behaviour suggests about their syntactic function. In some cases, as with
Icelandic dative subjects (often called ‘quirky subjects’), discussed brieXy in
Section 1.2, the syntactic behaviour of the dependent in question is system-
atically subject-like, in spite of the dependent-marking or head-marking
morphology. Icelandic dative (and accusative or genitive) subjects have
most of the many S/A syntactic privileges of the language. They essentially
only fail to behave like subjects with respect to morphological marking, that is
case marking and verb agreement. In other languages, certain similar kinds of
dative-marked NPs may have only a small set of the various syntactic privil-
eges of the prototypical subject. Although it is not entirely clear how to draw a
line or whether one should be drawn, this latter type, considered to be a
quasi-subject, is distinguished here from the Icelandic type, which is called an
oblique subject (or dative subject, accusative subject, etc.).22 Although
possibly less common and certainly less well-studied, the same kind of
variation appears to occur with objects. Some languages use alternatives to
O-marking with dependents that otherwise have essentially all the syntactic
privileges of objects and are thus oblique objects, whereas other languages
have quasi-objects, often with certain verbs or verb classes, that have a much
more restricted set of object properties and may or may not be marked like an
oblique.

2.3.1 Quasi-subjects and oblique subjects

Beyond the passive agents of some languages, there are various kinds of
syntactic constructions that could be characterized as having quasi-subjects.
In English, for example, the P-marked location-designating NPs in the fol-
lowing sentences occupy the preverbal slot normally reserved for S/A depend-
ents, whereas the postverbal NP determines the presence or absence of the 3rd
singular S/A-indexing -s suYx.
(2.84) a. At the top of that mountain sits an eagle’s nest.
b. In this grave lie the remains of a great man.
In this so-called locative inversion construction (Coopmans 1989, Levin
and Rappaport Hovav 1995: ch. 6, Bresnan 1994, Ura 2000: ch. 5), both
dependents are quasi-subjects, by virtue of the fact that each has at least one
S/A property and neither has all of the S/A properties. Numerous positions
have been taken on if either of the dependents is the ‘true’ subject. Any analysis
that attempts to portray either as the true subject runs into diYculties with the
fact that neither, for example, can undergo subject-Aux inversion (2.85), be
construed with a ‘Xoated’ quantiWer (2.86), for which there is an S/A
 Grammatical relations across languages 99

constraint, or be the subject of a raising verb such as seem (2.87), a possibility

restricted to the S/A of the embedded inWnitival phrase (see Section 2.2.1).
(2.85) a. Did these children see the movie?
b. * Does at the top of the mountain sit an eagle’s nest?
c. * Does an eagle’s nest at the top of the mountain sit?
(2.86) a. These children both watched the movies.
b. * In these graves both lie those great men.
(2.87) a. These children seemed [ØA to like the movie].
b. ??At the top of the mountain seems [Ø to sit an eagle’s nest].
c. *An eagle’s nest seems [at the top of the mountain to sit Ø].
Many languages have an analogous construction in which the locative
dependent has varying degrees of subject properties, with Bantu languages
such as Chichewa apparently having an oblique subject in the Icelandic sense
(Bresnan and Kanerva 1989). English also has a locative PP subject in a
copular non-verbal predicate construction that only superWcially resembles
the preverbal locative inversion quasi-subject, and which might be designated
an oblique subject, as shown by the following examples:
(2.88) a. Behind the couch is a good place to hide.
b. Is behind the couch a good place to hide?
c. Behind the couch and under the bed are both good places to hide.
d. Behind the couch seems [Øs to be a good place to hide].
Probably the most well-studied kind of quasi-subject is the verb or verb-
class restricted dative kind that is found in many languages of diVerent families
and types, including Germanic languages (Allen 1995, Sigurðsson 2002), Rus-
sian and Polish (Dziwirek 1994, Franks 1995, Moore and Perlmutter 2000),
Marathi, Newari, and Hindi ( Joshi 1993, Rosen and Wali 1989, Mohanan 1994,
Bickel and Yādava 2000), French and Italian (Perlmutter 1984, Belletti and Rizzi
1988, Legendre 1989a), Choctaw (Davies 1986), K’ekchi (Berinstein 1990), and
Korean and Japanese (Ura 1999, 2000). Although these languages are often
claimed to have dative subjects in the Icelandic sense, the dependent type in
question in many of these languages has relatively few syntactic subject prop-
erties, as noted by Moore and Perlmutter (2000).
In addition to a certain class of psychological verbs—discussed brieXy in
Section 1.2—whose experiencer and stimulus dependents are always marked
dative and nominative, respectively, but which alternate freely between sub-
ject and non-subject (Barðdal 2001). Icelandic has verbs and other predicates
of certain classes that have a dative, accusative, or genitive dependent that is a
100 Grammatical relations across languages

non-alternating subject (i.e. S or A). Most of these are either passivized verbs
or psychological verbs of some kind. The following examples (adapted from
Sigurðsson 2002) illustrate a few kinds of dative subjects, to which attention is
limited here.
(2.89) a.Mér höfðu alltaf virst stelpurnar vera gáfaðar.
1SgDat had.3Pl always seemed girls.FemPlNom to.be intelligent
‘The girls had always seemed to me to be intelligent.’
b. Mér var hjálpað.
1SgDat was helped
‘I was helped.’
c. ÓlaW leiddist.
Olaf.Dat bored
‘Olaf was bored.’
Neither the case morphology nor the verb agreement marks the preverbal NP in
clauses like these as the S/A dependent. In (2.89a), for example, the postverbal
nominative NP rather than the 1st person dative subject is indexed by the
agreement morphology on the auxiliary verb. When there is no nominative NP,
the predicate (auxiliary/verb/adjective complex) is in a non-agreeing or 3rd
singular form. The following examples highlight the agreement contrast be-
tween dative and nominative subjects with a predicate that alternates between
dative and nominative subjects, correlative with a diVerence in meaning.
(2.90) a. Strákarnir höfðu verið illir.
boys.MascPlNom had.3Pl been bad.MascPlNom
‘The boys had been angry.’
b. Strákunuum hafði verið illt.
boys.Dat had.3Sg been bad.NeutSg
‘The boys had been ill/felt badly.’
Since agents of volitional action-denoting predicates are invariably agreement
controllers in nominative form, as are the experiencers of many verbs of
psychological experience, it is perhaps tempting to consider the preverbal
dative NPs in examples like (2.89) and (2.90b) to be topicalized or fronted
constituents in clauses with no subject, or with a postverbal subject in the case
of (2.89a). However, with respect to numerous syntactic phenomena (as many
as sixteen according to Sigurðsson, including reXexive antecedence, subject-
verb inversion, raising, complement control, and conjunction reduction) only
the preverbal NP in such clauses behaves like a subject. For example, like the
nominative A of a verb like lesa ‘read’(2.91), a dative subject of an embedded
verb can be the subject of a raising verb such as byrja ‘begin’ (2.92).
 Grammatical relations across languages 101

(2.91) a. Ólafur las bókina.

Olaf.Nom read book.Acc
‘Olaf read the book.’

9 9
b. Ólafur byrja i [∅ a lesa bókina].
boys.Dat began A to read book.Acc
‘Olaf began to read the book.’
(2.92) a. Ólafi leiddist.
Olaf.Dat bored
‘Olaf was bored.’

9 9 9
b. Ólafi byrja i [a ∅ lei ast].
Olaf.Dat began Comp S bore
‘Olaf began to get bored.’
Similarly, it is possible to have a dative subject as the pivot in a controlled
complement clause (2.93a) or a conjunction-reduction structure (2.93b). As
in English, the pivot in such constructions has to be the S/A dependent.

9 9 9
(2.93) a. Ég vona ist [til a ∅ verda hjálpa ].
1SgNom hoped Comp SDAT was helped
‘I hoped to be helped.’

9 9 9 9
b. Ég haf i miki a gera] og [∅ var samt ekki hjálpa ].
1SgNom had much to do and SDAT was still not helped
‘I had much to do and was still not helped.’

As noted by Sigurðsson, even though German has certain superWcially

similar dative experiencer constructions, as illustrated by (2.94), the dative
NP generally lacks syntactic subject properties, as illustrated—in brief—for
complement control and conjunction reduction by the examples in (2.95).
(2.94) a. Mir ist kalt.
1SgDat is cold
‘I’m cold.’
b. Mir wurde geholfen.
1SgDat was helped
‘I was helped.’
102 Grammatical relations across languages

(2.95) a. *Ich hoffte [∅ geholfen zu werden].

1SgNom hoped SDAT helped to be
‘I hoped to be helped.’

b. *[Ich hatte viel zu tun] und [∅ wurde trotzdem nicht geholfen].

I had much to do and SDAT was still not helped
‘I had much to do and was still not helped.’

Unlike in Icelandic, such dative NPs in German appear to be non-subjects, or,

at most, quasi-subjects.
The experiencer dependent of certain psychological verbs in French is not a
subject according to the dependent-marking, head-marking, and constituent-
order conventions of the language. With respect to these conventions, it
behaves like an indirect object (or dative oblique) as shown by the following
examples (with essentially literal English glosses).
(2.96) a. Pierre a téléphoné à Marie.
Pierre telephoned to Marie. (i.e. ‘called Marie’)
b. Pierre lui a téléphoné.
Pierre 3SgDat-telephoned.
(2.97) a. Cette femme plait à Pierre.
This woman pleases to Pierre. (i.e. ‘is pleasing to Pierre’)
b. Cette femme lui plait.
This woman 3SgDat-pleases.
(2.98) a. Elle manque à sa famille.
She misses to her family. (i.e. ‘Her family misses her.’)
b. Elle lui manque.
She 3SgDat-misses. (i.e. ‘He/she misses her.’)
Both the addressee NP of a verb such as téléphoner and the experiencer NP of
verbs such as plaire ‘please’ and manquer ‘miss’ are marked with a prepos-
ition, are placed postverbally, and, when pronominal, are realized as a dative
verbal aYx/clitic (which the standard orthography shows as a preverbal
word). The usual analysis is that dative/à-marked dependents such as these
are indirect objects. They fail to behave like subjects with respect to such S-/A-
restricted phenomena as conjunction reduction, raising, and complement
control. For example, as in English, only the preverbal NP in a sentence
such as (2.96a) can be an omitted element (i.e. pivot) in the second of two
 Grammatical relations across languages 103

conjoined clauses, under the control of the S/A of the Wrst clause, as shown by
(2.99). Similarly in an iniWnitival phrase embedded under a verb such as
vouloir ‘want,’ the pivot, which is controlled by the subject of vouloir, can only
be the S/A of the inWnitival verb (2.101). With verbs in the plaire class, the
stimulus NP can be a conjunction-reduction pivot or a controllee, but the
dative experiencer cannot, as shown by (2.100) and (2.102).

(2.99) a. [Pierre-S est entré dans un café] et [Ø-S a téléphoné à Marie].

‘Pierre went into a café and telephoned to Marie.’
b. [Pierre-S est entré dans un café] et [Marie lui/*Ø a téléphoné].
‘Pierre went into a café and Marie 3SgDat/*Ø-telephoned.’
(2.100) a. [Il-S est bien mignon] et [Ø-NOM plait à Corinne].
‘He is very cute and pleases to Corinne.’
b. [Il-S a plusieurs amis] et [cela lui/*Ø plait].
‘He has many friends and that 3SgDat/*Ø-pleases.’
(2.101) a. Pierre veut [Ø-S téléphoner à Marie].
‘Pierre wants to telephone to Marie.’
b. *Pierre veut [Marie téléphoner Ø-DAT ].
‘Pierre wants Marie to telephone (to him).’
(2.102) a. Pierre veut [Ø-NOM plaire à Marie].
‘Pierre wants to please to Marie.’
b. *Pierre veut [Marie plaire Ø-DAT ].
‘Pierre wants Marie to please (him).’
The experiencer dependent of plaire, manquer, and other similar verbs
in French is clearly not a dative/oblique subject in the Icelandic sense. It
is, however, a quasi-subject, inasmuch as it has ‘subject’ properties with
respect to control of missing subjects in certain kinds of adverbial phrases.
As shown by the following examples (adapted from Legendre 1989a), the
controller of ‘before/after’ inWnitival phrases with a missing subject, which
ordinarily must be an S/A, can also be the dative dependent of a verb in the
plaire class.
controller must be S of téléphoner

(2.103) a. Marie-S lui a téléphoné [avant de ∅-S s’enfuir à l’etranger].

‘Marie 3SgDat-telephoned before fleeing abroad.’ (i.e. ‘before
she/*he fled’)
104 Grammatical relations across languages

controller can be either Nominative or Dative dependent of plaire

b. Que cette femme lui plaise [avant même de ∅-S lui avior parlé...]
‘That this woman should 3SgDat-please before even 3SgDat-having
spoken.’ (i.e. ‘That this woman should be pleasing to him before her
having spoken to him’ or ‘before his having spoken to her’).
Legendre also shows that the dative experiencer of verbs in the plaire class
can be controllers for certain other kinds of participial or inWnitival adverbial
phrases, for which the constraint otherwise works in term of the notion ‘subject’.
Similar kinds of limited syntactic subject properties (generally, ability to
control adverbial phrases and reXexive pronouns) characterize the dative
dependents of certain classes of verbs in languages such as Polish (Dziwirek
1994), Russian (Moore and Perlmutter 2000), and Italian (Perlmutter 1984).
The extent to which these quasi-subjects are subject-like varies somewhat,
however. In Italian, for example, the dative quasi-subject can not only be a
controller for inWnitival and participial adverbial phrases, but can also
antecede certain reXexives and is preferably placed in the preverbal
subject position (Belletti and Rizzi 1988), unlike in French. Choctaw (Davies
1986) is even closer to the Icelandic end of the quasi-subject spectrum.
The dative quasi-subject in Choctaw triggers indirect object agreement
on the verb and cannot be the pivot in a complement control construction
(unlike in Icelandic, for example—see (2.93)). However, it is subject-like in that
it has nominative dependent marking and can be a controller for the purposes
of reXexives, switch-reference marking, and complement control.

2.3.2 Quasi-objects and oblique objects

Direct objects are sometimes marked in diVerent ways, correlative with certain
semantic distinctions. For example, in many languages overt morphological
O marking can alternate with no marking, depending on relative salience in
terms of either person/animacy or deWniteness (Bossong 1991, Aissen 2003).
By way of illustration, the Australian ergative language Pitjantjatjara uses the
absolutive case marker -nya on O dependents if they are pronouns or proper
names (2.104a) but not if they are common NPs (2.104b).
(2.104) a. tjitji-ngku Billy-nya/ngayu-nya nya-ngu
child-Erg Billy-Abs/1Sg-Abs see-Pst
‘The child saw Billy/me.’
b. Billy-lu tjitji nya-ngu
Billy-Erg child see-Pst
‘Billy saw the child.’ (Aissen 2003)
 Grammatical relations across languages 105

There is a similar constraint in Romanian, which obligatorily case marks only

animate-referring proper or pronominal NPs (Farkas 1978, Dobrovie-Sorin
1994) and certain Bantu languages, in which the obligatoriness of O head
marking is contingent on whether the O is animate-referring or, in some
cases, human-referring (Morimoto 2002). In other languages, such as Hebrew
(Givón 1978) and Turkish (Enç 1991), only NPs whose referents are high in
deWniteness or speciWcity are overtly case-marked. Although these kinds of
marking alternations have a semantic dimension, in general they do not
indicate syntactic function distinctions.
In Spanish, human-referring NPs that are also speciWc are marked with the
same preposition that is used for indirect objects (IOs) (e.g. Zagona 2002:
ch. 1). The inanimate-referring O (la casa ‘the house’) in (2.105a) and (2.105b)
is a bare NP, whereas the human-referring O in (2.105b) is marked with the
preposition a ‘to’, like the IO in (2.105a).
(2.105) a. Marta le vendió la casa a una mujer.
Marta 3SgDat sold the house to a woman
‘Marta sold the house to the woman.’
b. Marta vio la casa / vio a una mujer.
Marta saw the house saw to a woman
‘Marta saw the house/saw a woman.’
The a-marked O is apparently a ‘dative direct object’ or oblique object,
analogous to the Icelandic dative subject, rather than either an indirect object
or a quasi-object, which is to say that it is morphologically marked like
an indirect object/oblique while having all the syntactic privileges of an O.23
O-dependents are distinguished from obliques and indirect objects by their
ability to be the S of a passive clause (2.106) and the pivot in the tough-
movement construction (2.107) or the lo que pseudocleft construction (2.108).
(2.106) a. La casa fue vendida/vista por Marta.
‘The house was sold/seen by Marta.’
b. *La mujer fue vendida la casa.
‘The woman was sold the house.’
(2.107) a. Esas casas son difı́ciles de vender/ver.
‘These houses are diYcult to sell/see.’
b. *Esas mujeres son difı́ciles de (les) vender casas (a).
‘These women are diYcult to (3PlDat) sell houses (to).’
(2.108) a. Lo que vendi/vi fue una casa.
‘What I sold/saw was a house.’
b. *Lo que (le) vendi la casa fue a una mujer.
‘What I (3SgDat) sold the house was to a woman.’
106 Grammatical relations across languages

Human-referring Os that are a-marked behave like their nonhuman-referring

counterparts rather than like IOs with respect to these O-deWning syntactic
properties:
(2.109) a. La mujer fue vista por muchas personas.
‘The woman was seen by many people.’
b. Los soldados a veces son difı́ciles de ver.
‘Soldiers are sometimes diYcult to see.’
c. Lo que vi fue a una mujer.
‘What I saw was a woman.’
Brazilian Portuguese (Farrell 2004) provides a clearer example of a kind of
quasi-object. Two of several classes of verbs designating psychological or emo-
tional states or processes can be distinguished according to whether the stimu-
lus-referring dependent is expressed as the O or as an oblique marked with de ‘of,
from’, which is either a preposition or a preWxal marker on a determiner, as
illustrated by the following examples (with more or less literal translations).
(2.110) a. Eu amo/adoro/odeio aquela música.
‘I love/adore/hate that song.’
b. Eu cansei/enjoei daquela música.
‘I tired/sickened of-that song.’
The two kinds of non-subject dependents systematically diVer with respect to
the O-deWning properties of the language. The bare NP type can be expressed
as an accusative pronominal clitic (see Section 2.1.1), as shown by (2.111a), can
be alternatively expressed as the subject of a passive clause (2.112a), can be the
pivot in the o que pseudocleft construction (2.113a), the tough-movement
construction (2.114a), and the reXexive/reciprocal se construction (2.115a)
(see Section 1.3.2). The de-marked stimulus dependent of verbs such as cansar
and enjoar, like other oblique dependents, has none of these O-deWning
properties, as shown by the (b)-examples in the following contrasting sen-
tence pairs.
(2.111) a. Minha mãe o amava.
‘My mother 3SgMasc-loved.’
b. *Minha mãe o cansou.
‘My mother 3SgMasc-tired.’
(2.112) a. Aquela música foi adorada por todo mundo.
‘That song was adored by everyone.’
b. *Aquela música foi enjoada por todo mundo.
‘That song was sickened (of) by everyone.’
 Grammatical relations across languages 107

(2.113) a. O que odiei foi a comida.

‘What I hated was the food.’
b. *O que cansei foi da comida.
‘What I tired was of-the food.’
(2.114) a. Pessoas como essas são difı́ceis de amar.
‘People like those are hard to love.’
b. *Este lugar é difı́cil de cansar.
‘This place is hard to tire (of).’
(2.115) a. Meus pais se amam muito.
‘My parents 3Rec-love a lot.’
b. *Meus pais se cansaram muito.
‘My parents 3Rec-tired a lot.’
(i.e. *‘got really tired of each other’, but OK: ‘got really tired’24)
There is a third class of psychological verbs, including at least gostar ‘like’
and precisar ‘need,’25 whose stimulus dependent is obligatorily de-marked (at
least when it is expressed in the default postverbal position) like the stimulus
of cansar ‘tire’ and enjoar ‘sicken’, as illustrated by (2.116a). With this class of
verbs, the de-marked dependent behaves like an oblique with respect to
passivization (2.116b) and the accusative pronominal clitic (2.116c),26 but
like an O with respect to the other O-constrained phenomena (2.116d–f).
(2.116) a. Eu gosto daquilo/*aquilo.
‘I like of-that/*that.’
b. *Aquela música foi gostada por todo mundo.
‘That music was liked by everyone.’
c. *Minha mãe o gostou.
‘My mother 3SgAcc-liked.’
d. O que gostei foi da comida.
‘What I liked was of-the food.’
e. Este lugar é difı́cil de não gostar.
‘This place is hard not to like.’
f. A gente se gosta muito.
‘We 3Rec-like a lot.’ (i.e. ‘We really like each other.’)
To be more precise about the phenomena in question, only the passive and
accusative clitic constraints are strictly O-speciWc constraints. The others do,
however, distinguish between the categories O and oblique, and in most cases
O, IO, and oblique. The pivot of the o que psedocleft construction, for example,
can be any A/S/O dependent (e.g. O que me irrita é a desonestidade ‘What
irritates me is dishonesty’), but not an IO (*O que vendi isso foi a ele ‘What I sold
108 Grammatical relations across languages

this was to him’) or any oblique (*O que a meia desapareceu foi da gaveta ‘What
the socks disappeared was from the drawer’). The reXexive/reciprocal pivot is
restricted to the category O/IO (see Section 2.1.1). The tough-movement con-
struction works on a split-intransitive basis, excluding A, IO, and oblique pivots,
but allowing both O pivots and suYciently patient-like S pivots (e.g. Essa
mancha vai ser difı́cil de sair ‘This stain is going to be hard to come out’ vs.
*Essa mulher vai ser difı́cil de gritar ‘This woman is going to be hard to shout’).
Thus, the categories of dependent types relevant to the constraints on the
phenomena in question are as shown in Table 2.4. The de-marked dependent
of verbs in the gostar class is a quasi-object, inasmuch as it is both O-like and
oblique-like in terms of its syntactic privileges and morphological properties.
Indonesian has a somewhat similar kind of quasi-object, with a large class
of verbs of cognition and emotion, including sengan ‘like, happy’, sayang ‘love,
pity’, lupa ‘forget’, and takut ‘afraid’ (Musgrave 2001, to appear). Unlike in
Brazilian Portuguese, the stimulus argument with these verbs can, in general,
be alternatively realized as a prepositionally-marked oblique, as shown by
(2.117a), a ‘true’ O with an applicative marker on the verb (2.117b), or a quasi-
object, which is unmarked (2.117c).
(2.117) a. Ali sengan dengan rumah itu(oblique, P-marked stimulus)
Ali like/happy with house that
‘Ali likes the house.’
b. Ali sengan-i rumah itu (stimulus ¼ applicative O)
Ali like/happy-Appl house that
‘Ali likes the house.’
c. Ali sengan rumah itu (stimulus ¼ quasi-O)
Ali like/happy house that
‘Ali likes the house.’
As summarized in Table 2.5, the applicative O with verbs in the sengan class
has all the O or A/S/O properties of the language, including lack of a
prepositional marker and the potential to be construed with a ‘Xoated’

Table 2.4. Grammatical properties of dependent types in Brazilian Portuguese

S/A SO O Quasi-O IO Oblique

O que pseudocleft pivot yes yes yes yes no no

Tough-movement pivot no yes yes yes no no
Reciprocal se pivot NA NA yes yes yes no
Passive subject potential NA NA yes no no no
Accusative promominal clitic no no yes no no no
Not marked with preposition yes yes yes no no no
 Grammatical relations across languages 109

Table 2.5. Grammatical properties of stimulus non-subjects in Indonesian

O/OAPPL Quasi-O Oblique

Controller of Xoated quatiWer yes yes no

Relativization potential yes yes no
Not P-marked yes yes no
Co-occurrence with transitive verbal preWx yes no no
‘Passive’ subject potential yes no no

quantiWer, to be relativized, to co-occur with a transitive preWx on the verb,

and to be realized as an S in a construction analogous to the English or
Portuguese passive construcion. By contrast, the prepositionally-marked
oblique realization of the same dependent has no O or A/S/O properties
and the non-applicative ‘quasi-O’ has only three such properties.
The following examples illustrate the quantiWer-Xoat phenomenon,
whereby the quantiWer semuanya ‘all’ can appear in clause-Wnal position,
separated from the NP that it modiWes.

floated quantifier modifies either subject or O or quasi-O

(2.118) a. anak-anak itu suka(i) gula-gula itu semuanya

child-PI that like(.Appl) sweet-PI that all
‘All the childern like the sweets/The childen like all the sweets.’

floated quantifier modifies subject but not comitative oblique

b. orang-orang Sasak datang dengan anak-anaknya semunaya

man-PI Sasak come with child-PI all
‘All the Sasak people came with their childen.’
* ‘The Sasak people came with all their children.’

floated quantifier modifies subject but not stimulus oblique

c. anak-anak itu suka dengan gula-gula itu semunaya

child-PI that like with sweet-PI that all
‘All the childen like the sweets /*The children like all the sweets.’

The distinction between quasi-objects and ‘true’ Os can be illustrated with

the passive-like construction that in some ways resembles the so-called
110 Grammatical relations across languages

objective-voice construction of languages such as Tagalog (see Section 2.2.5),

for which reason the voice preWx is glossed with Obj here. In an active-voice
clause, such as (2.119a), the agent precedes the verb and the patient follows. In
the objective voice, the patient precedes the verb and the agent follows,
possibly—in some cases—marked with a preposition (2.119b).
(2.119) a. dia baca buku itu
3Sg read book that
‘He/she read the book.’
b. buku itu di-baca (oleh) Ali
book that Obj-read by Ali
‘The book was read by Ali.’
With the psychological verbs whose stimulus dependent alternates between
oblique-marking and no marking, as illustrated for takut ‘afraid’ by (2.120a),
the objective-voice construction can only be used if the applicative suYx
appears on the verb, as shown by the contrast between (2.120b) and (2.120c).
Thus, only the applicative O with verbs in this class can be ‘promoted’ to
subject.
(2.120) a. saya takut (dengan) lelaki itu
1Sg afraid with man that
‘I’m very afraid of that man.’
b. *lelaki itu di-takut saya
man that Obj-afraid 1Sg
‘That man is feared by me.’
c. lelaki itu di-takut-i saya
man that Obj-afraid-Appl 1Sg
‘That man is feared by me.’
Although there are many languages that either have idiosyncratically case-
marked objects or systematic alternative options for case marking with certain
verb classes, the extent to which the marking is indicative of quasi-objecthood
is not always clear. Another kind of apparent quasi-object is the genitive-
marked object in Polish (Dziwirek 1994) and Russian (Franks 1995), which
appears as an alternative to or in place of an accusative NP in clauses that have
been negated. In the Uralic language Ostyak (Nikolaeva 1999) the verb can
either be marked for object agreement or not. The object in a clause with no
verb agreement is a kind of quasi-object, inasmuch as it does not have all the
syntactic properties and privileges of the agreement-triggering object. As with
 Grammatical relations across languages 111

quasi-subjects, the inconsistent syntactic behaviour of these diVerent kinds of

objects poses a challenge to theories of grammatical relations, which are
generally structured in such a way as to yield discrete rather than continuous
categories.
 3

Relational Grammar

Relational Grammar (RG) (Perlmutter 1983, Perlmutter and Rosen 1984, Blake
1990, Postal and Joseph 1990) is a theory of syntax that is built on the idea that
grammatical relations such as subject, direct object, and indirect object are
primitive (i.e. basic and indeWnable) concepts in terms of which clause
structure in all languages is organized. Its use of multiple levels of clause
structure, whereby an NP with the patient role, for example, can be the direct
object at an initial or ‘logical’ level and the subject at the Wnal level, make it
well-suited to describing the voice and relational-alternation constructions
that are so pivotal to an understanding of the syntactic phenomena in
most languages, as discussed in Section 2.2, as well as such ubiquitous
phenomena as raising to subject, possessor raising, causativization, noun
incorporation, quantiWer Xoat, and control of adverbial phrases and comple-
ment clauses, which are generally constrained in terms of relational
categories such as subject and direct object. Coupled with its general user-
friendliness, that is its relative lack of obscure and complex formalism, its
cross-linguistic adaptability has made it particularly popular among Weld-
workers, who have produced accessible and theoretically-informed descrip-
tive work of enduring interest on a wide array of typologically and genetically
diverse languages.

3.1 Basic design of the theory

Constituents are assumed to bear a relation within the phrase that contains
them. In a simple transitive clause, for example, there are two NPs and a verb.
The verb bears the relation of predicate (or P). The NPs bear the relations
subject and direct object. Consistent with the typological research concerning
a grammatical-relation hierarchy in the sense of Keenan and Comrie (1977; see
Section 1.3.2), a set of core relations, known as ‘term’ relations, are assumed to
be ranked relative to each other and to outrank all non-term relations. The
numbers 1, 2, and 3 designate the term relations, that is subject, direct object,
and indirect object, and indicate their ranking. The relations 1 and 2 consti-
 Relational Grammar 113

tute the class of ‘nuclear terms’. The 2 and 3 relations constitute the class of
object terms. More generally, the dependent (or non-P) relations are categor-
ized and hierarchically-ranked as shown in Figure 3.1.
Although semantic roles such as agent and patient are not included in the
inventory of grammatical relations, the individual oblique relations, such as
loc(ative) and instr(umental), are, for all intents and purposes, simply se-
mantic roles. They have the labels of standard semantic roles and constitute a
set of similarly uncertain size. Moreover, unlike the chômeur relation and the
term relations, they are purely semantically deWned and cannot be revalued to.
That is to say, although a dependent with the initial instrumental relation, for
example, can revalue to 2 and therefore be a Wnal 2, an initial 2 cannot be a
Wnal instrumental, as guaranteed by the Oblique Law in (3.26) below.
The non-oblique relations can, in principle, be borne by a dependent with
any semantic role. However, it is generally assumed that at the initial level (or
‘stratum’), there is a semantic basis for assigning the term relations as well as
the oblique realtions. The prototypical agent, patient, and recipient depend-
ents are initial 1, 2, and 3, respectively. Although this idea is codiWed in the
Uniform Alignment Hypothesis (Perlmutter and Postal 1984), the precise
details of how initial relations are determined from verb meanings within
particular languages and across languages are given little attention and the
meaningfulness and viability of any such universal principle are controversial
(Rosen 1984).
The laws governing clause structure yield a set of possible construction
types, deWned by the nature of the revaluations, if any, that take place.
Individual languages utilize possibly diVerent subsets of the constructions
made available by universal grammar. The passive construction, for example,
is deWned by a detransitivizing revaluation of 2 to 1.1 The RG characterization
of the passive construction in Halkomelem, for example, is illustrated in
Figure 3.2, which shows the analysis of example (2.40b) from Section 2.2.1
(with the auxiliary omitted for the sake of simplicity). The formal represen-
tation of clause structure is a graph-theoretic object consisting of nodes that
are linked by arcs labelled for grammatical relation and stratum. A phrase or
word that is a constituent of a larger phrase, identiWed by a node labeled a, is
said to ‘head’ an arc whose tail is a. In the diagram on the left in Figure 3.2, tu@
scé Lt@n ‘the salmon’ heads two arcs with tail a. Thus, tu@ scé Lt@n is a con-
t
s
t
s

stituent of clause a. One of the arcs is labelled for the grammatical relation 2
and the coordinate c1, indicating that this NP is the initial 2 of the clause. The
other arc is labelled for the grammatical relation 1 and the coordinate c2,
indicating that this NP revalues to 1 and is, hence, the Wnal 1 of the clause
(since there are no arcs with coordinate c3). The other constituents of the
114 Relational Grammar

term non-term
nuclear
1 > 2 >3 > loc, instr, ben, … chômeur

object oblique

grammatical relations

Figure 3.1. Typology of dependent (or ‘nominal’) grammatical relations

P 1 c1
c1,2 c2
chô c2 1
c1
2 initial P 1 2
final P chô 1
q’w l- t- m
e e e θ s éni ] [tθ scé t n]
e e e e q⬘w l- t- m
e e e e θ e s éni ] [tθ scé t n]
e e
bake-Tr-Intr Obl Det woman Det salmon
‘The salmon was baked by the woman.’

Figure 3 .2. RG analysis of a simple passive clause in Halkomelem

clause are the verb, which heads a P(redicate)-arc with coordinates c1 and c2
and is, thus, the initial and Wnal P, and the oblique-marked u@ sLéni? ‘the
woman’, which heads an initial-stratum 1-arc (coordinate c1) and a Wnal-
stratum chô-arc. The key information about the grammatical relations and
strata of the constituents of the clause is more perspicuously summarized in
the tabular stratal diagram on the right in Figure 3.2.
The grammatical rules of the language, including rules having to do with
constituent order, derivational and inXectional verbal morphology, case mark-
ing, and agreement, are formulated in terms of the relational and stratal
information. This is a passive clause because there is a revaluation from 2 to 1
from a transitive stratum, that is a stratum with both a 1-arc and a 2-arc. Since
there can be at most one dependent bearing a given term relation in a given
stratum, the intial 1 revalues to chômeur. The combination of transitive and
intransitive suYxes on the verb register the passive revaluation. The verb
has zero agreement morphology because the Wnal 1 is third person and the
Wnal stratum is intransitive. The NP u@ sLéni? is marked for oblique case with the
oblique preposition because all non-term dependents are marked in this way.
Naturally, provided that no laws governing clause structure are violated, a
clause can have multiple strata with various kinds of revaluations. The
 Relational Grammar 115

following Halkomelem example illustrates a clause type generally assumed to

involve both advancement to 2 (in an applicative construction—see Section
2.2.4) and passive revaluation.

initial P 3 1 2
P 2 1 chô
Wnal P 1 chô chô

(3.1) ?i ?am?-@s-t-@m tu@ John ?@-l: ’ Mary ?@ kwu@ šckı́ ks t


s

Aux give-Rec-Tr-Intr Det John Obl-Det Mary Obl Det vanilla

‘John is being given vanilla extract by Mary.’
The recipient applicative morpheme on the verb registers the revaluation of
the initial 3 to 2, which induces demotion of the initial 2 to chômeur. The
intermediate stratum being transitive enables the passive-voice advancement
of the recipient dependent to 1. The verb, thus, bears the transitive and
intransitive morphemes that register passive as well as the recipient applica-
tive morphology. Since the agent and patient are both Wnal non-terms, both
are marked with the oblique preposition.
Some of the most basic (and largely uncontroversial) laws are as follows
(see Perlmutter and Postal 1983 for more formal characterizations).
(3.2) a. Stratal Uniqueness Law
A stratum can have at most one of each type of term arc.
b. Oblique Law
If a dependent bears an oblique relation R, it bears R in the initial
stratum.
c. Final 1 Law
The Wnal stratum of a clause has a 1-arc.
d. Motivated Chômage Law
A dependent may only bear the chômeur relation if it bears a
relation R in some sratum (ci ) and another dependent bears R in
the subsequent stratum (ciþ1 ).
The Stratal Uniqueness Law is meant to account for the fact that languages
typically do not have two dependents in a clause with identical syntactic
privileges and morphological marking properties. The Oblique Law, as
noted above, prohibits the revaluing of any dependent to an oblique relation,
the idea being that revaluations simply change syntactic privileges without
substantially aVecting meaning and ‘instrumental’, for example, is a relation
that reXects an aspect of meaning. The Final 1 Law accounts for the fact that
116 Relational Grammar

the sole argument of a single-argument verb is necessarily realized as the

subject (To do nothing won’t help), unless a ‘dummy’ pronoun functioning as
subject is inserted (It won’t help to do nothing). Chômage is the device that
saves structures with revaluations from violating either the Stratal Uniqueness
Law or the Oblique Law. If the 2 of a clause with an agent and a patient
revalues to 1, the initial 1 must revalue. Since it cannot revalue to an oblique
relation, it revalues to chômeur—a syntactic function that any dependent of a
verb can, in principle, bear. The Motivated Chômage Law limits revaluation
to chômeur to cases where some other revaluation induces it. In (3.1), for
example, the agent and the patient are both Wnal chômeurs. In both cases,
their chômage is induced by a revlauation of the recipient (or initial 3).

3.2 Basic language types

It should be clear that the relations 1 and 2 correspond roughly to the concepts
S/A and O discussed in Section 2.1. However, the relativization of all gram-
matical relations to strata makes it possible and necessary to be more precise
about syntactic functions. An accusative language, for example, is one whose
case marking and/or verb agreement (possibly as well as other syntactic
phenomena) distinguish the Wnal 1 and Wnal 2 of clauses. Accusative languages
are presumably the most common type because marking systems generally
work in terms of Wnal relations and 1 and 2 are the most basic relational
notions. The notions ‘ergative’ and ‘absolutive’, by contrast, are more com-
plex, since they are deWned in terms of other notions:
(3.3) a. Ergative
x bears the ergative relation in stratum s, if x bears the 1 relation and
s is transitive.
b. Absolutive
x bears the absolutive relation in stratum s, if x bears the lowest-
ranking nuclear term relation in s.
A morphologically ergative language is, thus, one in which case marking and/
or agreement work in terms of the Wnal ergative and Wnal absolutive of
clauses. For example, in the following monostratal transitive Halkomelem
clause, the verb bears the suYx -@s, which indexes the Wnal ergative (i.e. the
Wnal 1 of a Wnally transitive clause).

initial=final P 2 1
(3.4) ni q’wáqw-@u-ám?š-@s tu@ John
Aux club-Tr-1O-3A Det John
‘John clubbed me.’
 Relational Grammar 117

Although John is also the Wnal 1 of example (3.1), the verb of this clause does
not bear the suYx -@s, because the Wnal stratum of a passive clause is
intransitive and the Wnal 1 is, therefore, the Wnal absolutive.
From an RG perspective it is unsurprising that so-called ergative languages
often have some morphological marking and other syntactic phenomena that
work on an accusative or some other non-ergative basis, since notions such as
‘Wnal 1’ (¼ S/A) and ‘Wnal 2’ (¼ O) are always available, even if more complex
relational notions may be appealed to for certain purposes. In Halkomelem,
for example, even though such syntactic phenomena as quantiWer Xoat and
possessor extraction are constrained to work with the Wnal absolutive, that is
S/O (see Section 2.1.2), use of the Wrst-person suYx on the verb in (3.4) works
on an accusative basis, that is it only ever indexes the Wnal 2 (¼ O). What
remains most surprising about ergative languages from an RG perspective is
that in languages whose syntax works primarily on an ergative basis, such as
Dyirbal (see Section 2.1.2), the Wnal absolutive of a clause is the one that is
most privileged syntactically. Given a relational hierarchy on which 1 outranks
2, the ergative relation (which encompasses only 1s) should at least not be
outranked by absolutive (which encompasses 2s).
Split-intransitive languages are characterized as having marking systems that
are sensitive to initial as well as Wnal relations. The basic idea is that such languages
overtly display a universal basic distinction between two types of intransitive
verbs (Perlmutter 1978, 1989). Patient-like dependents of initially intransitive
verbs are initial 2s, whereas agent-like dependents of initially intransitive verbs
are intial 1s. By way of illustration, consider the following examples of transitive
and intransitive clauses in Choctaw discussed in Section 2.1.3.2
(3.5) a. chi-bashli-li-tok is-sa-bashli-tok
2sgO-cut-1sgA-past 2sgA-1sgO-cut-past
‘I cut you.’ ‘You cut me.’
A-marking O-marking
b. hilha-li-tok sa-hohchafoh
dance-1sgA-past 1sgO-be hungry
‘I danced.’ ‘I’m hungry’.
The two types of intransitive clause are assumed to have the following
relational structures:
initial 2 P
initial=final P 1 final 1 P
(3.6) a. hilha-li-tok b. sa-hohchafoh
(unergative clause) (unaccusative clause)
118 Relational Grammar

Given this analysis, verbs such as ‘be hungry’ show O marking for the
dependent because it is a direct object at the initial level. The rules for
head-marking on verbs can be formulated as follows:
(3.7) Head marking in Choctaw
a. O marking is used for dependents that bear the 2 relation (in any
stratum) and a Wnal nuclear term relation.
b. A marking is used for dependents that bear the initial and Wnal 1
relation.
Since languages may have morphosyntactic rules keyed in various ways to the
distinction beween intransitive verbs with an initial 2 (unaccusatives) and
intransitive verbs with an initial 1 (unergatives), there is no reason to expect
so-called split-intransitive languages to have a fundamentally diVerent syntax
than other languages. All languages are assumed to have two main classes of
intransitive verbs. Split-intransitive or active languages are simply languages
that key some aspect of their main grammatical-relation marking mechan-
isms (usually head marking) to the distinction. The fact that the syntax of
some languages, such as Acehnese (see Section 2.1.3), appears to use no other
distinction between core grammatical-relations than initial/Wnal 1 vs. initial 2
is certainly not ruled out as a logical possibility; but it is (appropriately)
rather unexpected.
Summarizing, in RG all languages are assumed to have the same
basic clause structure and the same grammatical relations. Morphosyntactic
phenomena, however, can work in terms of categories that can be deWned
in diVerent ways, taking into account stratal and relational information.
The three main kinds of languages in terms of grammatical-
relation systems deWne their two main relational categories as shown in
Table 3.1.

3.3 Some case studies

3.3.1 Dative quasi-subjects
One of the main advantages of RG is that the revaluation mechanism makes it
possible to account for the fact that dependents of a clause sometimes behave
only partially like a subject or object, indirect object, etc. The quasi-subject
phenomenon, in particular, has provided one of the main motivations for
multistratal representations of grammatical relations (Perlmutter 1982). Con-
sider, for example, the hybrid category S=A=Obla (i.e. the category consisting of
S/A and oblique agents), which is needed to characterize the raising-to-object
 Relational Grammar 119

Table 3.1. RG deWnitions of main syntactic categories in three language types

Accusative Ergative Split-intransitive

S/A ¼ O¼ A¼ S/O ¼ A=Sa ¼ O=So ¼

Final 1 Final 2 Final 1 of Lowest-ranking Initial and Initial 2 and
Wnally transi- Wnal nuclear Wnal 1 Wnal nuclear
tive clause term term

pivots in Halkomelem (see Section 2.2.1), for example. Since oblique agents
are always initial 1s that have been demoted to chômeur and the S/A category
is that of Wnal 1s and there are no other kinds of dependents that head a 1-arc
in Halkomelem, raising-to-object pivots are a natural class: dependents that
head a 1-arc (in some stratum). In languages such as French, Italian, Polish,
and Choctaw, which have classes of indirect objects with only certain subject
properties, there is also a straightforward analysis. The quasi-subject indirect
object is an initial 1 that has been demoted to 3, as shown in the following
stratal diagram for the plaire ‘please’ clause in a French sentence with two
possible controllers of an adverbial phrase.
(3.8) controller can be either Nominative or Dative dependent of plaire

Que cette femme lui plaise [avant même de Ø-s lui avoir parlé...

‘That this woman should 3SgDat-please before even 3SgDat-having

spoken.’ (i.e. ‘That this woman should be pleasing to him before her having
spoken to him’ or ‘before his having spoken to her’)
initial 2 1 P
2 3 P
Wnal 1 3 P
cette lui plaire
femme
The controller has to (minimally) be a subject (i.e. head a 1-arc in some stratum).
It can be cette femme, because this is the Wnal 1 and has all or most subject
privileges; but it can also be lui, which is both a Wnal term and an initial 1. Lui is a
quasi-subject, that is only manifests a restricted range of subject privileges,
because most syntactic phenomena in French are keyed to Wnal relations.
3.3.2 Philippine voice alternations
The complex grammar of Philippine languages is amenable to a straightfor-
ward treatment if subject revaluations of the 1 to 3 kind, which yields quasi-
120 Relational Grammar

subjects, are allowed. Recall from Section 2.2.5 that grammatical phenomena
are, by and large, sensitive to four main grammatical-relation categories. The
most important is that of what is called here the ª (or ang-marked) depen-
dent, which can have virtually any semantic role, contingent on the choice of
voice-marking morphology on the verb. Most grammatical phenomena are
contrained to work with only the ª dependent. Another category consists of
both the ª dependent and the non-ª or ‘demoted’ agent. Only dependents
in this category can be controllees in participial complements or controllers of
participial adjuncts. Instrument-referring NPs and non-ª agents and pa-
tients constitute the category of genitive-marked dependents. Finally the
category consisting of PPs and locative case-marked NPs, including goals
and recipients, is distinguished by the ability of its members to undergo
adjunct fronting. The situation concerning the syntactic phenomena related
to these categories is summarized in Figure 2.2.
The key assumptions for a successful analysis are as follows:
. The ª dependent is always the Wnal 1, implying that the multiple voices
indicate diVerent kinds of 1s or advancements to 1, as in Bell’s (1983)
analysis of Cebuano, for example.
. Recipients and other locative NPs are simply obliques, like PPs, rather than
initial/Wnal 3s.
. Initial 1s displaced by advancements to 1 demote to 3.
. Instrumental NPs are initial/Wnal 3s (if they do not advance to 1).
. Patients are initial 2s (that can advance to 1 but otherwise don’t revalue).
Given this set of assumptions, most syntactic phenomena, including, for
example, relativization, quantiWer Xoat, number agreement, and conjunction
reduction, work in terms of the Wnal 1 category (S/A). These phenomena
therefore work as in many languages, particularly the most common accusa-
tive type, which Tagalog would be on this analysis. The controller of a
participial adjunct and the controllee in a participial adjunct must be either
the ª dependent or a demoted agent in a non-active voice (i.e. S/A/Agent).
On the proposed analysis, this category is deWned as dependents that bear the
1 relation in any stratum (whether initial or Wnal). Genitive marking is used
for Wnal objects, that is 2s or 3s, including deomoted agents. Adjunct fronting,
which doesn’t work with the ª dependent and genitive-marked dependents,
is restricted to Wnal non-terms.
There is really only one non-standard assumption in this version of the
classic RG analaysis, which is that initial 3-hood can vary considerably across
languages, insofar as instrumental NPs are claimed to be initial 3s in Tagalog
and recipients, for example, are not.3 Since there appears to be no viable
 Relational Grammar 121

universal semantic characterization of the category ‘initial 3’, in any case,

let alone an empirical basis for the claim that all languages even utilize the 3
relation (see Section 1.3.2), this seems to be an unproblematic assumption.4
By way of example of how Tagalog is seen as working under this analysis,
consider the proposed analyis of a complex clause with an advancement to 1
(from an oblique relation) in the main clause and control of a participial
complement in which an advancement from 2 to 1 has occurred, that is
(2.82b), repeated here as (3.9), with a literal (non-idiomatic) gloss that reXects
the analysis.

(3.9) initial P 1 Loc

final P 3 1

iniwanan ko siya=ng
leave.Imperf.Loc 1SgGen 3Sg=Comp
[sinususulat Ø ang=liham]
write.Imperf.Obj Gen =letter
initial P 1 2
final P 3 1
‘I left him writing the letter.’
Literally: ‘He was left (to) me the letter being written.’

The controller is the Wnal 1 of the main clause. The controllee is the initial
1/Wnal 3 of the participial complement. The main clause verb is in the
locative voice, signalling the advancement to 1 from a (generalized) locative
oblique. The initial 1 (or agent) is marked genitive because it is a Wnal object
(or Wnal term other than 1). In the participial complement the verb is in the
objective voice due to the advancement to 1 of the intial 2 (or patient); even
though it is a Wnal 3, the agent is eligible to be a controllee by virtue of its
initial 1-hood. Thus, both the quasi-subject properties of the ‘demoted’ agent
and its superWcial non-subject term properties are straightforwardly
accounted for.

3.3.3 A/O-reversal in Jarawara

The syntactic inverse voice phenomenon of the kind described for Jarawara in
Section 2.2.3 is also explicable in terms of simple revaluations. The idea is that,
122 Relational Grammar

in the inverse voice, the initial 2 advances to 1 and, rather than being demoted
to chômeur, as it would be in the cross-linguistically more common passive
construction, the initial 1 is demoted to 2, as shown in the following analysis
of examples (2.66), repeated here as (3.10).

initial=final 1 2 P Direct voice

(3.10) a. Mioto Watati awa-ka
Mioto(Masc) Watati(Fem) see-DeclMasc
‘Mioto saw Watati.’
initial 2 1 P
Wnal 1 2 P Inverse voice
b. Watati Mioto hi-wa hi-ke
Watati(Fem) Mioto(Masc) Inv-see Inv-DeclFem
‘Mioto saw Watati.’ (or, ‘Watati was seen by Mioto.’)
The basic facts are that in the inverse-voice construction the patient (¼initial
2) has virtually all of the syntactic properties and privileges that the agent (¼
initial/Wnal 1) has in the direct-voice construction (with respect to constituent
order, verb/mood gender agreement, conjunction reduction, etc.) and the
demoted agent in the inverse voice, nevertheless, has the properties of the Wnal
2 of a transitive clause. The reversal analysis accounts for the object-like
properties of the agent in the inverse-voice construction, while also account-
ing for the subjecthood of the patient. Beyond this, the Wnal 2 is a quasi-object
and the Wnal 1 is a quasi-subject in the inverse voice, insofar as the former
behaves like a 1 and the latter behaves like a 2 with respect to one phenom-
enon: preverbal pronominal positions. In essence, the order is patient-
referring pronoun Wrst, agent-referring pronoun second, independently of
voice. One need only say that the preverbal pronominal placement constraint,
unlike the other morphosyntactic phenomena in the language, are sensitive
only to initial syntactic relations.

3.3.4 Unaccusativity and the alignment problem

Relational Grammar is probably most well known for the pioneering work of
Perlmutter (1978, 1989) and others (e.g. Rosen 1984, 1988, Harris 1982) on what
would become known as the unaccusative hypothesis,5 that is the hypo-
thesis that the subject of some intransitive verbs is an initial 2 rather than an
initial 1, which plays a critical role in the RG approach to split-intransitive
languages, as noted in Section 3.2. It has long been clear that many languages
have subclasses of verbs, generally deWnable in semantic terms, and that there is
 Relational Grammar 123

a semantic dimension to even the common distinction between two super-

categories of intransitive verbs that is used in many languages. As argued
persuasively in Rosen (1984), there are good reasons for abandoning the
original idea that there might be a universal semantic basis for the distinction
between unaccusative intransitive verbs (those with an initial 2) and unergative
intransitive verbs (those with an initial 1). Typological work on split-intransi-
tive languages (e.g. Mithun 1991, Dixon 1994: ch. 4) also makes clear that the
unaccusative/unergative distinction is not semantically predictable across lan-
guages, since these kinds of languages use diVerent semantic features to draw
the distinction. The question for the proponents of the unaccusative hypoth-
esis then is how the decision is made language-internally as to which verbs take
initial 2s and which take initial 1s. Given that lexical semantic features appear to
be clearly relevant, the idea would have to be that languages diVer with respect
to which lexical semantic features matter. In some languages, for example, only
the single dependents of stative verbs are initial 2s, whereas in other languages
the single dependents of telic (i.e. change of state) verbs are initial 2s, and so
forth. For some languages, the choice may even be partially idiosyncratic.
However, since the rationale for assigning diVerent initial grammatical
relations to the dependents of diVerent classes of verbs is that these two
types of dependents behave diVerently, one might argue that the distinctive
behaviours are explicable directly in terms of the semantic features that would
have to be said to motivate the unaccusative/unergative distinction in the Wrst
place, obviating the need for an initial syntactic-function diVerence. This tack
is taken in a number of studies (e.g. Centineo 1986, Van Valin 1990, McClure
1990, Zaenen 1993, Kishimoto 1996, Yang 1996, Aranovich 2000). The key
problem is that of unaccusative mismatches. Suppose, for example, there are
two classes of intransitive verbs in a given language: one indexes the S with
characteristic O-marking and the other with A-marking and the diVerence
hinges on whether or not the dependent is construed as controlling the action.
One could either say that the choice of head marking depends on the semantic
factor or that the semantic factor determines whether the dependent is an
initial 1 or 2 and the head marking depends on the initial grammatical
relation. If there is only one syntactic phenomenon at stake or all phenomena
at stake work on precisely the same basis, either analysis seems at least viable.
If, however, there are multiple phenomena at stake and at least some are keyed
to diVerent semantic features of verbs, then the motivation for claiming an
unaccusative/unergative distinction appears to evaporate. The choice of
which of the distinct phenomena reveals the basis for the unaccusative/
unergative distinction would be essentially arbitrary. Moreover, at least
some phenomena would have to be explained directly in semanatic terms
124 Relational Grammar

and, indeed, all apparently could. The syntactic unaccusative hypothesis

would seem, therefore, to be superXuous at best.
Making the problem more concrete, consider a contrast between Acehnese,
described in Section 2.1.3, and English. In Acehnese, for example, pronominal
proclitics are used for subjects of transitive verbs and of intransitive verbs
denoting controlled actions (¼ members of the descriptive category A=Sa ),
whereas pronominal enclitics are used for direct objects and subjects of
intransitive verbs denoting non-controlled actions (¼ members of the de-
scriptive category O=So ). The only other syntactic phenomena that appear to
draw a distinction among A/S/O dependents work on the same basis. That is
to say, possessor raising is possible for O=So dependents only and the pivot of
the so-called control construction with verbs like ‘want’ must be an A=Sa
dependent. One could say, essentially, that only core dependents whose
referents are interpreted as being in control of the action designated by the
predicate can be pivots in the ‘want’ construction or pronominal proclitics,
whereas pronominal encliticization and possessor raising are constrained to
work with core dependents whose referents are not so interpreted.6 On the
other hand, one could say that ‘control’ is the basis for the unaccusative vs.
unergative distinction in this language, that is that in-control dependents are
initial 1s and other core dependents are initial 2s. The constraints on the
phenomena in question could then be formulated in terms of intitial gram-
matical relations, that is only initial 2s can undergo possessor raising and
determine pronominal enclitics, for example. The only motivation, however,
for engaging the unaccusative hypothesis would be that it limits the role of
semantics in grammar. One might claim, for example, that the meanings of
verbs are only relevant for determining initial grammatical relations.
In English, in contrast to Acehnese, most of the phenomena that might be
claimed to motivate unaccusativity work on somewhat diVerent bases.7 For
example, only a subset of intransitive verbs can form participial adjectives,
which can otherwise only be formed from verbs that can have a patient/direct
object, that is transitive verbs (Bresnan 1982b, Rappaport and Levin 1988,
Farrell 1994a: 125–8, 1994b):
(3.11) a. a recently arisen national security problem (intransitive verb)
b. a lapsed candidacy (intransitive verb)
c. a fallen angel (intransitive verb)
d. a rebuilt engine (transitive verb)
e. a recently announced result (transitive verb)
f. *a recently laughed person (intransitive verb)
g. *a shone star (intransitive verb)
 Relational Grammar 125

At the same time, only a subset of intransitive verbs, including many that
form participial adjectives, can appear with an expletive there subject:
(3.12) a. There arose a controversy over the use of this test.(intransitive verb)
b. There lapsed three years before his return. (intransitive verb)
b. There fell from the sky an angel wrapped in light.(intransitive verb)
c. *There laughed a child. (intransitive verb)
d. *There announced the results a celebrity. (transitive verb)
e. *There rebuilt some engines a local mechanic. (transitive verb)
The problem is that the two classes of intransitive verbs identiWed by these
phenomena only partially overlap. Although the full story is undoubtedly
more complex, there insertion seems to be largely restricted to occurring in
sentences with intransitive verbs in which existence or appearance is some-
how at stake, including the state of existence (exist), coming into existence or
appearing (fall from the sky), or going out of existence (lapse). Participial-
adjective formation on the other hand seems to be restricted to telic intransi-
tive verbs with a patient dependent, that is those that denote a change of state
or location for a participant in an event. The two phenomena, thus, work with
diVerent verbs in many cases and in diVerent ways with many of the same
verbs. Fall for example, does not felicitously occur with an expletive there
subject in a context where the denoted falling does not have an appearance or
disappearance meaning (*There fell a wrench out of my tool belt) even though
this meaning of fall is compatible with participial adjective formation (I got
down on my knees to look for the fallen wrench). Moreover, certain verbs that
allow there insertion do not allow participial adjective formation and vice-
versa:
(3.13) telic and atelic intransitive verbs with existence/appearance at stake
a. There shone down on me the full rays of a bright October moon.
b. *The shone rays of the bright October moon left me dazzled.
b. There came a time when nothing worked anymore.
c. *The come time turned out to be very diVerent.
d. There exist very good reasons for doing it that way.
e. *The existed reasons are the same.
(3.14) telic intransitive verbs with no existence/appearance at stake
a. She didn’t want to touch the fainted soldier.
b. *There suddenly fainted a soldier.
c. We walked across the frozen lake.
d. *There recently froze a lake on the other side of the county.
126 Relational Grammar

In short, one could claim that the collective set of intransitive verbs that either
work with there insertion or that form participial adjectives constitutes the
class of unaccusative verbs, that is verbs that have an initial 2 and not an initial
1. In that case, however, the initial 2 status of the dependent of each class of
verbs would not be suYcient to explain the contrasting membership of the
two sub-classes and to account for the constraints on participial-adjective
formation and there insertion. Additional semantic factors such as telicity and
existence or appearance would need to be appealed to. But even if it were
assumed that all intransitive verbs simply have an initial/Wnal 1 the semantic
factors that are relevant to an understanding of the phenomena appear to
suYce to deWne the sub-classes and to account for the constraints. One could,
on the other hand, claim, somewhat arbitrarily, that only one of the sub-
classes is unaccusative. For example, it might be claimed that only the patient
of a telic intransitive verb can be an initial 2. The constraint on participial-
adjective formation could then be stated in terms of initial 2-hood. However,
this account ultimately just masks what is a semantic constraint on participial
adjective formation, inasmuch as no general explanatory beneWt accrues from
the claim that the class of verbs in question has an initial 2. Moreover, since
initial 2-hood cannot be appealed to for the constraint on there insertion, it is
still necessary to make use of semantic factors to characterize a grammatical
constraint. There is, thus, no clear motivation for the unaccusative hypothesis
internal to English.
Language-internal ‘unaccusative’ mismatches of the kind found in English
appear to be quite common, as in Dutch (Zaenen 1993), French (Legendre
1989b), (Levin and Rappaport Hovav 1995), Japanese (Kishimoto 1996), and
Spanish (Rex 2001).8 As noted in Section 2.1.3, the O-marking for intransitive
predicates that appears to announce unaccusativity in Kamaiurá yields an
unaccusative mismatch: stativity is criterial for O marking on intransitive
verbs, whereas agentivity (or participant control) is criteral for O marking on
nominalized verbs. The problem is exacerbated by the fact that there are
equally clear cross-linguistic mismatches with respect to the semantic basis for
split intransitivity (see especially Mithun 1991). In the absence of a substantive
theory of how initial 2-hood is determined either universally or in particular
languages, which RG has not developed, it is diYcult to see how it might be
claimed that the general phenomenon of intranstitive verb classes is illumin-
ated in an interesting way by the idea that some intransitive verbs have
subjects that are ‘underlying’ direct objects.
The larger problem, of course, is that the general basis for determining the
initial relations of the dependents of verbs in general remains uncertain.
There obviously would have to be both at least some universal semantic
 Relational Grammar 127

constraints of the kind envisioned but not spelled out by the Universal
Alignment Hypothesis (Perlmutter and Postal 1984) and language-speciWc
semantic constraints. It may well be that an explicit theory of the interface
between lexical semantics and grammatical relations would indeed yield a
viable version of the unaccusative hypothesis. However, articulating such a
theory remains an unmet challenge for RG.

3.3.5 Icelandic dative subjects

Dative/oblique subject constructions in which the apparent Wnal subject only
fails to behave like a subject with respect to case marking and verb agreement
constitute another kind of challenge for RG. Consider, for example, the fact
that there is a relatively large class of psychological verbs in Icelandic that can
have either the stimulus or experiencer as the apparent Wnal subject, as
illustrated by henta ‘please’ in the following examples from Barðdal (2001),
discussed in Section 1.2.
(3.15) a. Ég veit að þetta mun henta mér.
I know that this.Nom will.3Sg please 1SgDat
‘I know that this will be pleasing to me.’
b. Ég veit að mér mun henta þetta.
I know that 1SgDat will.3Sg please this.Nom
‘I know that I’ll be pleased with this.’
It might be tempting to utilize some version of the standard RG ‘inversion’
analysis, as illustrated in Section 3.3.1 for French. The problem, however, is
that the Icelandic constructions have quite diVerent properties than the dative
experiencer constructions of French. According to Barðdal (2001) the nom-
inative stimulus in the construction illustrated by (3.15a) has all the subject-
deWning properties of the language (conjunction reduction pivot, initial
position in main declarative and subordinate clauses, raising pivot, controller
of inWnitival clauses, reXexive controller, auxiliary inversion pivot, etc.) and
the dative experiencer has none. In the construction illustrated by (3.15b), on
the other hand, the dative experiencer alone has all the subject-deWning
properties, except that it is in the dative case and does not determine verb
agreement, which is restricted to work with the nominative dependent.
Based on the assumptions that the sentences are paraphrases and that
initial grammatical relations are determined semantically in a consistent
way across paraphrases (and across languages), one possible RG revaluation
analysis would have inversion (1–3 demotion) occurring in both (3.15a) and
(3.15b). The diVerence between the two would be that (3.15b) is an ‘imper-
128 Relational Grammar

sonal’ inversion construction, in which the stimulus is demoted to Wnal

chômeur by a phonologically null expletive pronoun (pro), as shown in the
following stratal diagrams:
initial 2 P 1 (Inversion)
2 P 3
Wnal 1 P 3
(3.16) a. þetta mun henta mér
this.Nom will.3Sg please 1SgDat
‘This will be pleasing to me.’
initial 1 P 2 (Impersonal inversion)
3 P 2
3 P 1
Wnal 1 3 P chô
b. pro mér mun henta þetta
1SgDat will.3Sg please this.Nom
‘I’ll be pleased with this.’
Under this analysis, all the subject-deWning phenomena as well as case
morphology and verb agreement could be said to be determined by the
Wnal relations, with the proviso that a dependent that is retired to chômeur
by an expletive pronoun bears the case of this pronoun and determines verb
agreement via a ‘brother-in-law’ relation with the Wnal 1 (Aissen 1988). The
fact that the dative experiencer in (3.16a) has no subject properties could be
attributed to the fact that it is a Wnal 3. Although there would be no evidence
supporting its initial 1-hood in this construction, there would also be no
counterevidence, since Wnal relations would be what matters for syntactic
phenomena. One problem is that the subject-deWning properties of the
language could not be said to be properties of Wnal 1s, since in (3.16b) it is
the Wnal 3 that has these properties. What one would have to say instead is that
what are generally considered to be the properties of ‘true’ or Wnal subjects,
are, rather, the properties of the non-expletive highest-ranking Wnal term.9
This approach is problematic in that it depends on positing an otherwise
unmotivated phonologically null expletive pronoun and requires recognizing
a novel (apparently Icelandic-speciWc) grammatical-relation category, that is
non-expletive highest-ranking Wnal term. In essence, it is necessary to take the
position that although the null expletive pronoun counts as being a depen-
dent of the clause for the purposes of meeting the demands of certain putative
laws of clause structure,10 it must be said not to count for purposes of
determining the highest-ranking Wnal grammatical relation.
 Relational Grammar 129

If one gives up the usual RG assumption that dative experiencers are

necessarily initial 1s, a simpler analysis emerges, as illustrated by the following
stratal diagrams:

initial/Wnal 1 P 3
(3.17) a. þetta mun henta mér
this.Nom will please 1SgDat
‘This will be pleasing to me.’
initial 3 P 1
Wnal 1 P 2
b. mér mun henta þetta
1SgDat will please this.Nom
‘I’ll be pleased with this.’
The idea is that the preverbal nominative stimulus construction is monostratal,
as shown in (3.17a), whereas the preverbal dative experiencer construction
involves 3–1 advancement, with the demoted initial 1 remaining a Wnal term.
This analysis dispenses with the phonologically null expletive pronoun and
allows the subject-sensitive syntactic phenomena (conjunction reduction, rais-
ing, constituent order, etc.) to be restricted to the Wnal 1. What makes Icelandic
unusual is mainly that case marking and agreement may be determined by initial
rather than Wnal relations. That is to say, dative case marking is used for initial 3s
and nominative case marking and verb agreement are keyed to the initial 1.
The problem with this analysis is that it is unclear how one might also
account for the standard passive construction (with prototypical agent and
patient dependents), in which the demoted 1 (or agent) is marked by the
preposition af ‘by’ rather than being marked nominative and the promoted 2
(or patient) is marked nominative rather than accusative (Zaenen et al. 1985,
Van Valin 1991, Minger 2002):

initial/Wnal 1 P 2
(3.18) a. Lögreglan tok Siggu fast
Police.Nom took.3Pl Siggu.Acc fast
‘The police arrested Sigga.’
initial 2 P 1
Wnal 1 P chô
b. Sigga var tekin föst af lögreglunni
Sigga.Nom was.3Sg taken fast by police.Dat
‘Sigga was arrested by the police.’
130 Relational Grammar

Since it is apparently not the case that initial relations systematically deter-
mine case marking, why they should apparently be relevant in the case of
alternations that occur with certain classes of verbs, such as psychological
verbs of the henta type, for example, remains a mystery.

3.4 Other relational theories

RG evolved from an attempt to build an alternative to classic Transform-
ational Grammar (see Chapter 5) in which syntactic functions such as subject,
direct object, and indirect object are central rather than epiphonemal. Al-
though phrase-structure representations of sentences built around lexical
cateories such as noun, verb, and preposition were abandoned, the ‘trans-
formation’ idea, which involved primarily changing the position within a
sentence of noun phrases (NPs) and other constituents, was retained in the
form of revaluations. Many of the phenomena that are of central concern to
transformational grammarians (passive, raising, complement control, causa-
tivization, etc.) are relatively easy to reconceptualize in terms of operations on
grammatical relations. Passive, for example, can be thought of as involving
promotion of the direct object to subject rather than movement of a VP-
internal NP to a VP-external position.

3.4.1 Lexical-Functional Grammar

Lexical-Functional Grammar (LFG) (Bresnan 1982a, 2000, Falk 2001) also
developed as an alternative to classic Transformational Grammar. Like RG,
LFG grants syntactic functions (called ‘grammatical functions’) a central role
and has developed for them an elaborate representational system independent
of phrase-structure trees. Unlike RG, however, LFG maintains phrase-struc-
ture representations of constituent structure and linear precedence, while
abandoning the idea of transformations as such. For each sentence, there is
assumed to be a single ‘functional’ structure (without revaluations) and a
single phrase-structure representation (without movements or other trans-
formations). Passive, for example, is conceived of as a purely morhpo-lexical
operation: a passivized verb has a diVerent functional structure and a diVerent
associated phrase structure than those of the base verb.
Focusing just on the assignment of grammatical functions to clausal con-
stituents, and putting aside the rather elaborate formal representation in
terms of functional structures as well as constituent structures, grammatical
relations are characterized essentially as follows. First, the lexical semantics of
words (for which no standard LFG-speciWc representational schema exists,
 Relational Grammar 131

although JackendoV ’s (e.g. 1990) model is often assumed) yield semantic roles
(or ‘theta-roles’) of the familiar kind for predicates. The argument structure
of a predicate consists of a list of the arguments from the conceptual structure
assigned an intrinsic grammatical-function feature and ordered according to
the position of their associated semantic role on a hierarchy of the following
kind (based on Falk 2001: 104).
(3.19) Thematic hierarchy
Agent > BeneWciary/Patient > Instrument > Theme > Path/Location
All arguments are mapped to a grammatical function, of which there are four:
subject, object, objectu (aka 2nd object), and obliqueu , where the theta
notation on object and oblique is a variable ranging over semantic roles.
Thus, it is possible to have obliqueloc , obliqueinstr , etc. The mapping of
semantic roles to grammatical functions for a typical three-place active-
voice verb such as put as used in I put my socks in the drawer is as follows.
(3.20) Mapping from argument structure to functional structure
put <Agent, Patient, Location>
j j j
Arg Struct: [
o] [
r] [
o]
j j j
Funct Struct: SUBJ OBJ OBLloc
The intrinsic grammatical function classiWcations of the arguments follows
default speciWcations, that can sometimes be overridden. The patient or
theme (see Section 1.3.4 on the diVerence) is associated with the semantically
unrestricted function feature [
r], for which reason it can in principle be
either subject or object. All other arguments are [
o] by default, that is
unable to be object. The mapping algorithm is as follows.
(3.21) Argument to functional structure mapping algorithm
a. A [
o] argument maps to SUBJ if it is the highest-ranking argu-
ment on the thematic hierarchy.
b. The [
r] argument maps to subject if possible and otherwise to
OBJ.
c. Any [þo] argument is mapped to OBJu.
d. Any remaining arguments are mapped to OBLu.
e. There can only be at most one SUBJ and one OBJ.
In the case of (3.20), the patient cannot map to subject because there can be
only one subject and the agent, being [
o] and the thematically highest-
ranking argument must map to subject. Therefore the patient maps to object.
132 Relational Grammar

The patient of put can map to subject only if the agent is taken out of the
picture. The eVect of passivization is to do just that. The thematically highest-
ranking argument is suppressed for mapping purposes (and therefore left
implicit or expressed as an adjunct):
(3.22) passive put <Agent, Patient, Location>
Ø j j
Arg Struct: [
r] [
o]
j j
Funct Struct: SUBJ OBLloc
The double-object construction, and applicative constructions more gen-
erally, allow for alternatives to the patient ¼ [
r] default. The idea is that the
recipient (or goal) argument of verbs such as give is construed as a beneWciary
when not realized as a PP and is therefore intrinsically classiWed as a [
r]
argument, for which reason the theme is classiWed as [þo]:
(3.23) give <Agent, BeneWciary, Theme>
j j j
Arg Struct: [
o] [
r] [þo]
j j j
Funct Struct: SUBJ OBJ OBJtheme
Naturally, the beneWciary/recipient is mapped to subject in a passive con-
struction built on double-object give (The boy was given a bath), as it is the
only [
r] argument.
One nice result of the feature-based deWnitions of grammatical functions
(i.e. [
o,
r] ¼ SUBJ; [ þ o,
r] ¼ OBJu ; [ þ o, þ r] ¼ OBJu ; [---o, þ r]
¼ OBLu ) is that LFG has a built-in mechanism for dealing with quasi-
object phenomena in languages (at least to some extent). The O2 in the
English double-object construction, for example, is object-like by virtue of
its [þo] status but doesn’t have all the properties of an object, since it is like
an oblique in being [þr]. It cannot be the subject of a passive sentence
because a [
r] intrinsic classiWcation would be necessary.11 Consider, also,
the quasi-objects of verbs in the gostar ‘like’ class in Brazilian Portuguese
discussed in Section 2.3.2. The stimulus dependent of these verbs is marked
with a preposition, unlike direct objects, and fails to behave like an object with
respect to passivization and accusative pronominal cliticization. However, it
behaves like a direct object with respect to reXexivization, pseudoclefting, and
tough movement. Assuming (as in the Role and Reference Grammar schema
discussed in Section 4.1.1) that the experiencer role, being beneWciary-like, is
higher than theme on the thematic hierarchy and that what I have been calling
 Relational Grammar 133

‘stimulus’ is encompassed in a generalized theme category, the grammatical

function assignment for verbs such as gostar can be characterized as in (3.24).
Gostar is diVerent from similar psych verbs that take a direct object theme in
that its theme argument is lexically stipulated to be intrinsically [þo] rather
than [
r], for which reason it maps to objectu.
(3.24) gostar <Exp, Theme>
j j
Arg Struct: [
o] [þo]
j j
Funct Struct: SUBJ OBJtheme
It follows without stipulation that verbs such as gostar cannot undergo
passivization, since suppression of the experiencer argument leaves no possi-
bility for a subject (i.e. the [þo] theme argument cannot map to subject,
which is necessarily [
o]). One need only assume that accusative pronouns
must have the object function (i.e. must be [þo,
r], whereas pseudoclefting,
reXexivization, and tough movement work for any [þo] argument, that is
either object or objectu ).
It is somewhat less clear how LFG might deal with the more complex problem
of quasi-subject phenomena, since there is no subjectu function, for example. It
is, however, possible for constraints on grammatical phenomena to be formu-
lated in terms of argument structure rather than simply functional structure.
Dative arguments typically show subject properties only in constructions in
which they are associated with the highest-ranking semantic role. For example,
French verbs in the class of plaire ‘please/like’ (see Sections 2.3.1 and 3.3.1) are
stative verbs with experiencer and theme arguments (as with gostar in Brazilian
Portuguese). The diVerence between the two languages is simply that in French
the experiencer argument has a special lexical speciWcation that causes it to be
mapped to oblique, for which reason the [
r] theme is mapped to subject, as
shown in (3.25).
(3.25) plaire <Exp, Theme>
j j
Arg Struct: [þr] [
r]
j j
Funct Struct: OBLexp SUBJ
The quasi-subject behaviour of the experiencer argument can be attributed to
the fact that the controller of adverbial phrases must simply be the highest
ranking agrument on some grammatical-relation hierarchy (i.e. either the
thematic hierarchy or the function hierarchy, which puts subject Wrst).
134 Relational Grammar

3.4.2 Arc Pair Grammar

Arc Pair Grammar (APG) (Johnson and Postal 1980, Postal 1986, 1989) is a
kind of evolutionary alternative to RG. In essence, APG focuses on the
formalism of relational networks (with labelled arcs) as a device for represent-
ing linguistic structure, while maintaining most of the basic RG assumptions
about clause structure and grammatical relations. Various new relations are
posited, so as to deal with such things as coordination, anaphora, the internal
relational structure of words with inXectional morphemes, and such con-
stituents as PPs, which RG, in practice, simply treats as nondistinct represen-
tationally from other so-called ‘nominal’ dependents of clauses. Primitive
relations between arcs (particularly, ‘sponsor’ and ‘erase’12) are also explicity
recognized and represented, in such a way as to allow for new or alternative
conceptualizations of relational ‘laws’ and various kinds of grammatical
phenomena. Additionally, a larger set of term relations is posited (e.g. Postal
1990), including 4, 5, and 6, largely in order to deal with quasi-object and
other quasi-term phenomena in languages, without resorting to demotion to
oblique relations. The daunting formalism, which requires abandoning sim-
ple stratal diagrams in favour of relational networks that include many more
arcs than those of classic RG as well as various kinds of arrows connecting arcs
(for the sponsor and erase relations, for example), seems to have made APG
less appealing than RG to descriptively-oriented theoreticians and Weldwor-
kers. With the notable exception of Aissen’s (1987) study of Tzotzil, APG has
not been widely used to produce comprehensive grammatical descriptions of
languages in the same way that RG has.
 4

Role and Reference Grammar

Role and Reference Grammar (RRG) (Foley and Van Valin 1984, Van Valin and
La Polla 1997, Van Valin 1993, to appear) provides an interesting contrast to
Relational Grammar precisely because it focuses considerable attention on the
issue that RG needs to address most but doesn’t, that is the details of a
presumed relationship between verb meaning and grammatical relations.
Furthermore, while drawing a clear line between semantic roles, which are
ascribed considerable importance, and syntactic functions, the latter are held
to be deWned and used in language-speciWc and even construction-speciWc
ways that are subject to considerable variation and the former are ascribed a
relatively central role in accounting for grammatical phenomena. Whereas RG
attempts to provide a framework for analysing the syntactic component of
grammar, which is assumed to be autonomous in the sense of Chomsky (1957),
RRG maintains that syntax cannot be properly understood without an inte-
grated account of the semantic and pragmatic functions of language. The
design of RRG from its inception has been informed by a concern for account-
ing for the grammatical properties of typologically and genetically diverse
languages, considered on their own terms. It has been utilized in grammatical
descriptions of a wide variety of languages and its central ideas have been
widely appealed to in theory-neutral typological and descriptive work.

4.1 Basic design of the theory

4.1.1 Lexical decomposition and semantic roles
The meanings of lexical items such as verbs and adjectives and the clauses
containing them are explicitly represented in RRG using a lexical-semantic
decomposition model akin in certain respects to those developed in or for
Localist Case Theory (Anderson 1971, and see Section 1.3.4), the Natural
Semantic Metalanguage framework (Wierzbicka 1996),1 and Generative
Grammar in various forms (McCawley 1971, Gruber 1965, JackendoV 1972,
1990, Pinker 1989). The idea is that lexical items can be decomposed into
logical structures (LSs) consisting primarily of combinations of primitive
136 Role and Reference Grammar

predicates (e.g. do’, see’, be-at’, dead’), predicate modiWers such as CAUSE,
BECOME, NOT and &, and arguments, which are typically lexical variables
that are instantiated by NPs at the sentence level—but may also take the form
of speciWc predicates. The relationship between predicates and arguments is
indicated in the traditional predicate-calculus format used in formal seman-
tics: an LS of the form predicate’ (x, y) indicates a predicate with two
arguments ordered in a signiWcant way, that is (Wrst, second), such that feel’
(x, y), for example, means x feels y. By way of illustration, the lexical LSs of the
adjective broken, the verb break in its accomplishment and causative senses,
the verb eat, and the verb take, are as follows (with examples sentences and
translations of the LSs from the metalanguage into more or less idiomatic
English—for the more complicated cases).
(4.1) a. LS of broken: broken’ (x)
Example: The window is broken.
b. LS of accomplishment break: BECOME broken’ (x)
Example: The window broke.
c. LS of causative break: [do’ (x, Ø)] CAUSE [BECOME broken’ (y)]
Translation: x does something (unspeciWed ¼ Ø) that causes y to
become broken.
Example: The boy broke the window.
d. LS of eat: do’ (x, [eat’ (x, y)])
Example: The children are eating pizza.
e. LS of take: [do’ (x, Ø)] CAUSE [BECOME NOT have’ (y, z) &
BECOME have’ (x, z)]
Translation: x does something that causes (i) y not to have z and
then (ii) x to have z.
Example: The boy took the money from his mother.
The theory of LSs is articulated in such a way as to give meanings in a
format that can be applied across languages and to yield distinctions con-
cerning Aktionsart (types of action) in the sense of Vendler (1967) and
semantic roles (or cases/thematic relations) in the sense of Gruber (1965)
and Fillmore (1968). Stative verbs/adjectives have the simplest kind of LS, that
is predicate’ (x), as in the case of broken (4.1a), or predicate’ (x,y), as in the
case of like (like’ (x, y)).2 Activity verbs are formally indicated in the meta-
language by the generalized activity predicate do’, as with eat in (4.1d) or the
single-argument verb laugh (do’ (x, [laugh’ (x)]) ). Accomplishment verbs
are deWned by the addition of the BECOME modiWer to a state LS, as in the
case of break in (4.1b) or learn (BECOME know’ (x, y) ). Causative verbs are
deWned by the inclusion of the CAUSE modiWer, as in the case of take in (4.1e).
 Role and Reference Grammar 137

In principle, any two LSs can be related by CAUSE. For example, scare, as in
Earthquakes scare me, involves an unspeciWed activity that causes a feeling
state ([do’ (x, Ø)] CAUSE [feel’ (y, afraid’)]). Although there are certain other
modiWers that can be included in LSs and other patterns that yield further
Aktionsart distinctions, the basic idea of the metalanguage and how it can be
used to deWne verb classes should be clear.
At the most Wne-grained level of analysis, a large variety of semantic roles
can be recognized in terms of relations between basic kinds of predicates and
their arguments, as shown in Table 4.1 (adapted from Van Valin and La Polla
1997: 115). In general, the semantic role of an NP can be identiWed from the
lexical decomposition, since NPs either instantiate a variable argument of a
basic state or basic activity predicate or of such a predicate embedded within a
more complex LS. Thus, the window in The boy broke the window has the role
of patient because it expresses the only argument of the predicate broken’ (x),
which happens to be modiWed by BECOME in a sub-LS of a complex
causative LS, as shown in (4.1c).
In essence, by combining information about basic predicate type (activity
vs. state), argument status (only vs. Wrst vs. second), and other semantic
factors (e.g. motion vs. creation vs. consumption vs. sound emission, etc.),
numerous semantic roles can be distinguished. However, these semantic roles
can eVectively be collapsed into a smaller set of distributionally contrastive
roles, as in (4.2), where one of the possible semantic roles can be chosen as a
mnemonic category label (or prototype) for each class of noncontrastive
roles.
(4.2) a. pred’ (x)
single argument of state predicate ¼ patient , entity
b. pred’ (x, y)
second argument of pred’ ¼ theme , stimulus, content, possessed,
target, consumed, creation, implement, etc.
c. pred’ (x, y)
Wrst argument of pred’ ¼ location , perceiver, possessor, cognizer,
emoter, etc.
d. do’ (x, y)
Wrst argument of do’ ¼ effector , l-emitter, s-emitter, creator,
consumer, user, etc.
In addition to these basic classes of roles, there are a few other more speciWc
roles that need to be deWned, because they demonstrably play a role in the
grammars of languages. For example, the role of agent, conceived of as a
volitional, purposeful instigator of an action is distinguished from that of
138 Role and Reference Grammar

Table 4 .1. Some possible semantic roles in relation to verb types and LSs
Type of verb Example LS Semantic roles

State verbs
state or condition broken’ (x) x ¼ patient
existence exist’ (x) x ¼ entity
pure location be-loc’ (x, y) x ¼ location; y ¼ theme
perception hear’ (x, y) x ¼ perceiver; y ¼ stimulus
cognition know’ (x, y) x ¼ cognizer; y ¼ content
possession have’ (x, y) x ¼ possessor; y ¼possessed
emotion love’ (x, y) x ¼ emoter; y ¼ target
inner experience feel’ (x, y) x ¼ experiencer; y ¼ sensation
Activity verbs
unspeciWed action do’ (x, Ø) x ¼ effector
motion do’ (x, [walk’ (x)]) x ¼ mover
light emission do’ (x, [shine’ (x)]) x ¼ l-emitter
sound emission do’ (x, [babble’ (x)]) x ¼ s-emitter
creation do’ (x, [make’ (x, y)]) x ¼ creator; y ¼ creation
consumption do’ (x, [eat’ (x, (y) )]) x ¼ consumer; y ¼ consumed
use do’ (x, [use’ (x, (y) )]) x ¼ user; y ¼ implement

eVector by the element DO in LSs. Thus, the verb murder, for example, whose
eVector must be interpreted as wilful and purposeful would have the lexical
representation DO (x, [do’ (x, Ø)] CAUSE [BECOME dead’ (y)]) (i.e. x acts
wilfully and purposefully in doing something that causes y to become dead).
The verb kill does not require that its subject be construed as an agent (in the
relevant sense), as evidenced by the possibility of The poison killed the rats.
Thus, kill does not have a lexically represented DO in its LS, although, as with
many verbs, its eVector can be interpreted as an agent in particular contexts
of use.
The traditional roles of recipient, source, and goal are basically special
versions of the Wrst argument of pred’ (x, y), that is variants of what can be
considered a location-type role, as can be seen from the LSs for put, give,
remove, and receive:
(4.3) a. put: [do’ (x, Ø)] CAUSE [BECOME be-loc’ (y, z)]
The man (eVector) put the beer (theme) in the refrigerator
(location).
b. give: [do’ (x, Ø)] CAUSE [BECOME have’ (y, z)]
The man (eVector) gave the letter (possessed) to the boss
(possessor).
 Role and Reference Grammar 139

c. remove: [do’ (x, Ø)] CAUSE [BECOME NOT be-loc’ (y, z)]
The man (eVector) removed the beer (theme) from the reWgerator
(location).
d. receive: BECOME have’ (x, y) & NOT have’ (z, y)
The man (possessor) received a ticket (possessed) from the police-
man (possessor).
A recipient, that is a typically animate participant that comes to have some-
thing, is the possessor (¼ Wrst) argument of BECOME have’ (y, z), as with the
object of to in (4.3b) and the subject in (4.3d). A goal, that is a typically
inanimate entity to which something moves, is the location (¼ Wrst) argu-
ment of BECOME be-loc’ (y, z), as with the object of in in (4.3a). A source is
the location or possessor (¼ Wrst argument) of BECOME NOT have’ (y, z), as
in (4.3d), or BECOME NOT be-loc’ (y, z), as in (4.3d). Other traditional roles
that arguably play a role in grammatical phenomena, such as instrument,
speaker, and addressee, can be deWned in similar ways, although these require
further elaborations of LSs (Van Valin and La Polla 1997: Ch. 3).

4.1.2 Macroroles
The most important semantic roles from the perspective of grammatical
phenomena in languages are the so-called macroroles, actor and undergoer.
Analogues of these roles, which have been pivotal in RRG since its inception
(e.g. Foley and van Valin 1984), include the two arguments on the action tier
in the lexical decomposition schema of JackendoV (1990), the proto-agent
and proto-patient of Dowty (1991), and the agent and theme of Delancey
(1991). Although these also correspond more or less to the initial subject and
direct object, respectively, of RG, they diVer in being considered semantic
roles. In some sense, at least for the subject and object relations, RRG has an
explicit account of the semantic constraints on initial relations envisioned by
the Universal Alignment Hypothesis of RG (see Section 3.3.4). However, in
recognizing the semantic factors at play in the determination of actor and
undergoer roles and in characterizing syntactic functions as being fundamen-
tally diVerent, the RRG conception of grammatical relations diVers sign-
iWcantly from that of RG. Moreover, the RRG assignment of the undergoer
macrorole, in particular, doesn’t always correspond to standard RG assump-
tions about what are initial direct objects in English and other languages.
Several factors play a role in determining which arguments of a verb or
other lexical item are assigned which macroroles. The Wrst is a hierarchy of
markedness for actor and undergoer assignment, which ranks semantic roles
with respect to their relative distance from agent and patient, as shown in
140 Role and Reference Grammar

Figure 4.1. The general principle is that the markedness of actor macrorole
assignment increases in correlation with distance from agent on the hierarchy,
whereas the markedness of undergoer macrorole assignment increases in
correlation with distance from the patient role. If a verb has one semantic
argument and one macrorole to assign, there are only two basic scenarios,
which makes the choice between actor and undergoer predictable. The single
argument will be the patient (or entity) argument of a state predicate, as with
die (BECOME dead’ (x)) or exist (exist’ (x)), in which case it is an undergoer,
by virtue of the maximal closeness to the prototype for undergoer. Otherwise,
a single-argument verb has an activity LS, as with laugh (do’ (x), [laugh’ (x)]),
in which case it is assigned the actor macrorole, by virtue of the closeness to
the prototype for actor (and, of course, the distance from the prototype for
undergoer). The principle can be formulated as follows:
(4.4) Default macrorole assignment for verbs with one macrorole
a. If the verb has an activity predicate in its LS, the macrorole is actor.
b. Otherwise, the macrorole is undergoer.
If a verb has exactly two (distinct) semantic arguments that are both assigned
macroroles, it is also predictable from the hierarchy which one will be actor
and which one will be undergoer. For example, hear has two semantic
arguments, that is perceiver and stimulus (x and y respectively in the LS
hear’ (x, y)). The perceiver is the actor, since it is a less marked choice for
actor than the stimulus is. Similarly, in the case of causative break (4.1c), the
effector is actor and the patient is undergoer, since the eVector is a less marked
actor than the patient and the patient is a less marked undergoer.
A second factor bearing on the assignment of macroroles is M(acrorole)
transitivity, which is strongly inXuenced by but doesn’t necessarily correlate

Markedness of
Actor assignment Markedness of
Undergoer assignment

Arg of DO 1st Arg of 1st Arg of 2nd Arg of Arg of state

do´ (x, …) pred´ (x, y) pred´ (x, y) pred´ (x)
AGENT EFFECTOR LOCATION THEME PATIENT
consumer possessor possessed entity
creator recipient stimulus
mover experiencer consumed
user cognizer implement
etc. etc. etc.

Figure 4.1. Actor–undergoer hierarchy and semantic roles

 Role and Reference Grammar 141

perfectly with number of LS arguments. Although the number of macroroles

a verb may assign cannot exceed the number of its LS arguments, its argu-
ments need not all be aligned with macroroles. The verb run, for example, can
have both mover and goal arguments, as in the case of The boy ran to the store,
the LS of which is do’ (boy, [run’ (boy)]) & BECOME be-loc’ (store, boy).
The mover (eVector-type) argument is the actor, since the Wrst argument of
do’ is always assigned the actor macrorole.3 However, the goal argument,
which is expressed as the oblique PP to the store, is not assigned a macrorole.
Verbs with a goal argument and an eVector/theme argument are typically M-
intransitive in English, since English is a satellite-framed language in the sense
of Talmy (2000), that is a language in which the goal or path argument of
verbs of motion is typically overtly expressed outside the verb in a PP, or
‘satellite’. The general schema for macrorole expression across languages is
that macroroles are assigned maximally and macrorole arguments are ‘direct’,
that is NPs rather than PPs in a language like English and NPs assigned one of
the primary cases in a typical case-marking language. Thus, by default, a verb
with two LS arguments assigns two macroroles to (bare) NPs. However,
certain classes of verbs, as with English verbs of directed motion, and even
individual verbs can have special M-transitivity features that lead to diVerent
patterns of argument expression. Moreover, the passive-voice construction, to
which we return in Section 4.2.1, also involves a mismatch between between
macroroles and direct core arguments.
The default for verbs with only a state predicate is that M-transitivity is
determined by number of LS arguments. Love and hate, for example, have
only a two-argument state verb in their LSs (love’ (x, y) and hate’ (x, y) ) and
therefore have actor and undergoer macroroles, assigned to the Wrst and
second arguments respectively. However, a verb such as belong must be
lexically marked in a way that its quasi-synonym have does not, that is it
must be speciWed to take only one macrorole (MR1):
(4.5) a. have: have’ (x, y)
The neighbour has a new car.
(actor ¼ possessor; undergoer ¼ possessed)
b. belong: have’ (x, y) [MR1]
This car belongs to the neighbour.
(undergoer ¼ possessed)
Since there is no activity verb in its LS, the single macrorole of belong is
necessarily undergoer, in conformance with (4.4b). The possessed argument is
the one assigned the macrorole, since it is the least marked undergoer, or most
patient-like, as shown on the actor–undergoer hierarchy. The possessor of
142 Role and Reference Grammar

belong is expressed as an ‘oblique’ core argument, marked with to, which is the
default preposition for a non-macrorole Wrst argument of a state predicate.
Finally, more marked macrorole choices can be made. Particularly in the
case of 3-argument verbs, languages and verb classes within languages can
diVer with respect to how the undergoer macrorole is assigned. In Brazilian
Portuguese, for example, the verb dar ‘give’, like other similar verbs, only
allows the least-marked choice for the undergoer (i.e. the second argument of
BECOME have’ (x, y)), whereas in Halkomelem, the recipient (or Wrst
argument of BECOME have’ (x, y)) is necessarily the undergoer:4
(4.6) ‘give’: [do’ (x, Ø)] CAUSE [BECOME have’ (y, z)]
a. [O homem]x deu [um presente]z para [a mãe dele]y
‘The man gave a present to his mother.’ Brazilian Portuguese
x ¼ actor/eVector; z ¼ undergoer/possessed; y ¼ oblique recipient
b. ni ?ám-@s-t-@sx [kwu@ sqw@m@
y?]y ?@ [kwu@ su’ám?]z
Aux give-Recip-Tr-3 Det dog Obl Det bone
‘He gave the dog the bone.’ Halkomelem
x ¼ actor/eVector; y ¼ undergoer/recipient; z ¼ oblique possessed
In English, on the other hand, such verbs must be said to allow the undergoer
macrorole to be assigned to either the recipient (I gave the dog a bone) or the
theme-type argument (I gave a bone to the dog), for which reason sentences
with macrorole assignments analogous to both (4.6a) and (4.6b) are possible
expressions of the same LS. The English double-object construction diVers
from the Halkomelem recipient-O construction only in that it has the theme-
type argument expressed as a ‘direct’ non-macrorole core argument.
In some cases, the general rules can be overridden with speciWc verbs.
The verb allow in English, for example, only makes available the marked
linking of undergoer (This country allows us many privileges vs. *This country
allows many privileges to us), for which reason it must be lexically
speciWed that the U(ndergoer) role is necessarily associated with a particular
argument:
(4.7) allow : [do’ (x, Ø)] NOT CAUSE [NOT have’ (y, z)] [U ¼ y]
[This country]x allows usy [many privileges]z .
x ¼ actor/eVector; y ¼ undergoer/possessor; z ¼ non-macrorole core
argument
Although there is a clear connection between the traditional notion of direct
object and undergoer, these notions only partially overlap. Clearly, primary O
privileges are generally associated with the undergoer of a 2-macrorole clause.
An undergoer, however, can also be associated with subject privileges—both
 Role and Reference Grammar 143

in intransitive clauses, as noted above, and in passive clauses, to be discussed

in Section 4.2.1.

4.1.3 Clause structure and syntactic functions

As in Transformational Grammar, the constituents of clauses are classiWed by
the lexical category of their head (NP vs. PP, etc.) and are organized in
hierachical structures that indicate their constituency, their linear order, and
the nature of the relations between them. The verb and its arguments consti-
tute the ‘core’ of the clause, organized around the nucleus, which is typically a
verb. Adjuncts, such as adverbial phrases and PPs that indicate locative and
temporal settings are outside the core, in the ‘periphery’ of the clause. The
main distinctions drawn for arguments are macrorole vs. non-macrorole,
actor vs. undergoer, and direct vs. oblique. Since, at most, two arguments
can be linked to macroroles, one of the arguments of a verb with three
arguments will be a non-macrorole argument, as in the case of the example
from Kamaiurá in Figure 4.2.
Kamaiurá distinguishes the macrorole arguments from other arguments
(to some extent) with both dependent marking (postpositions) and head
marking (agreement) and constituent order. The default constituent order is:
macrorole argument(s) pre-nucleus; non-macrorole argument(s) post-nu-
cleus. Only macrorole arguments are also direct; non-macrorole arguments
are systematically marked with a postposition and only these cannot be
indexed by agreement morphology on the verb. As noted in Section 2.1.3
only one of the macrorole arguments is typically indexed on the verb; but
which one is indexed depends primarily on person (1st > 2nd > 3rd).5 Only in
case the person ranking of the two macrorole arguments is the same, does the
actor take precedence over the undergoer. Thus, in RRG the actor and under-
goer semantic roles may be appealed to directly in the formulation of gram-
matical constraints. In the case of verb agreement in Kamaiurá, they are
essentially surrogates for the traditional subject and direct object syntactic
functions. Instead of saying subject > object for agreement with 3rd person
dependents in a transitive clause (as one would in RG); one can say actor >
undergoer. There are, however, other cases in which the notions are not
interchangeable.
The concept ‘non-macrorole argument’ corresponds roughly to the trad-
itional concept ‘oblique argument’, although a distinction is also drawn
between direct non-macrorole arguments (e.g. dative and certain accusative
case-marked NPs or ‘bare’ NPs) and other non-macrorole arguments. Only
the latter are considered oblique.6 Unlike in RG and in traditional grammar,
there is no clear equivalent of the indirect object syntactic function in RRG.
144 Role and Reference Grammar

SENTENCE

CLAUSE

PERIPHERY CORE

ADV ARG ARG NUC ARG

PRED
PP

NP NP V NP P
ikuwe morerekwara ywyrapara o-me’eη kara’iwa upe
yesterday chief bow 3A-give foreigner to

Actor Undergoer

yesterday´ ([do´(morerekwara, Ø)] CAUSE [BECOME have´ (kara’iwa, ywyrapara)])

‘The chief gave the bow to the foreigner yesterday.’

Figure 4 .2. A sentence with a 3-argument verb and an adjunct in Kamaiurá

The analysis of a 3-argument clause with a source argument in Kamaiurá (or

any other language) would not diVer signiWcantly from that of the analysis
shown in Figure 4.2. The source argument would simply be a non-macrorole
argument expressed as a PP with the ‘from’ postposition. Hence, an oblique
vs. indirect object distinction is not drawn. This works well for Kamaiurá,
since as noted in Section 1.3.2, the only diVerence between recipient argu-
ments and source arguments seems to be the choice of postposition (upe ‘to’
vs. wi ‘from’), which can be attributed, in part, to the relevance of LS details
for postposition selection: wi for source (¼ Wrst argument of BECOME NOT
have’/be-loc’ (x,y)) and upe, presumably, as default—or ‘elsewhere’ choice—
for a non-macrorole Wrst argument of any pred’ (x, y), given that it is used for
various semantic roles, including beneWciary, animate goal, addressee, and
actor of passive verb—see example (2.40a). For a language such as Brazilian
Portuguese which distinguishes so-called indirect objects from other oblique
arguments less by choice of preposition than by the possibility of being
expressed as a dative pronominal clitic or being a reXexive/reciprocal pivot
(see Section 1.3.2), it is not entirely clear how the privileged class of argument
types might be speciWed.7
RRG also takes a nontraditional approach to the notion ‘subject’. In essence,
phenomena such as reXexivization, relativization, conjunction reduction, verb
agreement, raising, etc. are viewed as being possibly restricted to working with
subclasses of arguments that in many (indeed most) languages cannot be
strictly semantically characterized. These subclasses can be deWned in diVerent
 Role and Reference Grammar 145

ways, both across languages and on a construction-speciWc basis within a given

language. Two kinds of elements are recognized with respect to syntactic
phenomena: pivots and controllers. A controller determines the form or
interpretation of some other constituent of a clause; a pivot is a grammatically
privileged argument type that is not a controller, such as an argument that is
omitted in a particular construction and whose intepretation is controlled by
some other constituent. Consider, for example, the phenomenon of conjunc-
tion reduction in Dyirbal (see Sections 2.1.2 and 2.2.2). As the following
example (¼ (2.11)) illustrates, the O (or undergoer) of the Wrst clause, which
is overtly expressed, is understood as having the same referential value as the S
(in this case, the actor) of the second clause, which is not overtly expressed.
(4.8)
[ŋuma yabu-ŋgu bura-n] [∅ banaga-nyu]
father.Abs mother-Erg see-Nonfut S return-Nonfut
‘Mother saw father and he/*she returned.’

The undergoer of the Wrst clause is the controller of conjunction reduction

and the actor of the second is the pivot. In Dyirbal, both controller and pivot
in this construction are restricted to macrorole arguments. However, there is a
further restriction, as only one of the two macrorole arguments in a clause
with two can ever be controller or pivot. Which of the two is predictable—but
not entirely semantically predictable. It is necessary to take into account both
clause type and the actor vs. undergoer distinction. In an M-intransitive,
clause the single macrorole argument (either actor or undergoer) can be a
controller or pivot for conjunction reduction. These privileges extend only to
the undergoer in an active-voice clause, as illustrated by (4.8) for controller,
and the actor in an antipassive clause, as illustrated by the following example
(¼ (2.47)), again for controller.

(4.9)
[yabu bural-ŋa-nyu ŋuma-gu] [∅ babaga-nyu]
mother.Abs see-Antipass-Nonfut father-Dat S return-Nonfut
‘Mother saw father and she/*he returned.’

‘Pivot’ and ‘controller’ are quite diVerent from ‘subject’, because they are
phenomenon-speciWc relations. The controller of conjunction reduction, for
example, may be restricted to one type of argument, whereas a relative clause
pivot may be restricted to quite another. Although one can talk about the
conjunction-reduction pivot in Dyirbal, it makes no sense to talk about
the pivot in Dyirbal. However, since the same kind of voice-contingent
146 Role and Reference Grammar

neutralization of semantic roles is involved in the choice of pivot and con-

troller for conjunction reduction and in the choice of controllers and pivots in
various other grammatical phenomena in Dyirbal, it is possible to generalize
over the construction-speciWc pivots and controllers. For each clause type, it is
possible and useful to identify a default privileged syntactic argument (PSA),
which corresponds more closely to the traditional subject notion, albeit
relativized to speciWc languages. By way of example, the undergoer can be
said to be the default PSA in an active clause in Dyirbal, whereas the actor is
the default PSA in an antipassive clause, as shown in Figure 4.3. In an
antipassive clause, the undergoer macrorole is not assigned. The LS argument
that would be the undergoer in the active clause is realized as a non-macrorole
direct core arument, which is assigned the default dative case. The general-
ization concerning the default PSA across construction types in Dyirbal is that
it is the lowest-ranking macrorole argument (where actor outranks under-
goer). In an antipassive or other M-intransitive clause, there is only one
macrorole argument, which therefore is the lowest-ranking, whether it is
undergoer or actor; in an M-transitive clause, the undergoer is always the
lowest-ranking macrorole argument.
The PSA is characterized as a default, since the pivots and controllers that
the notion PSA generalizes over are phenomenon-speciWc. Since absolutive
case is assigned to the default PSA, the undergoer is marked absolutive in
(4.8) and the actor is marked absolutive in (4.9). It is also the default PSA that
is necessarily the pivot and controller in conjunction reduction. However, as

SENTENCE
SENTENCE
CLAUSE
CLAUSE
CORE
CORE

ARG NUC ARG

ARG ARG NUC
PRED
PRED
NP V NP
NP NP V
uma yabu gu buran yabu bural anyu umagu

voice? – active DAT

Actor Undergoer voice? – antipassive
PSA = Undergoer PSA = Actor Actor Undergoer
see´(yabu, uma)
see´(yabu, uma)

Figure 4.3. Default PSA choice in active and antipassive clauses in Dyirbal
 Role and Reference Grammar 147

noted in Section 2.1.2, pronominal case assignment, which works on an

accusative basis, treats the actor in both active and antipassive clauses in the
same way: nominative case is assigned to the macrorole argument (either
actor or undergoer) of M-intransitive clauses and to the actor in M-transtitive
clauses. Thus, nominative case assignment works in terms of a phenomenon-
speciWc PSA: the highest-ranking macrorole argument.
Not only can the choice of default PSA diVer across languages—an issue
addressed further in Section 4.2—but so can whether there even is a default
PSA. In Acehnese, for example, the grammatical phenomena of the language
are either essentially open to all core arguments, as in the case of the raising
construction (Van Valin and La Polla 1997: 259), or are, from an RRG
perspective, only semantically restricted. Possessor raising, with incorpor-
ation of the head noun, is restricted to the O or So, as shown by the following
examples, repeated from Section 2.1.3, with the relevant details of the RRG
analysis.

(4.10) a. seunang [até lôn]x

happy liver 1Sg
‘I’m happy.’ (Literally, ‘My liver (is) happy.’)
happy’ (x); (x ¼ patient & undergoer)
b. lôn seunang-até Possessor-raising host is undergoer
1Sg happy-liver
‘I’m happy.’ (Literally, ‘I’m liver-happy.’)
(4.11) a. ka lô nx -tët [rumoh gopnyan]y
Asp 1SgA-burn house 3Sg
‘I burned his/her house.’
[do’ (x, Ø)] CAUSE [BECOME burned’ (y)]
(y ¼ patient & undergoer)
b. gopnyan ka lôn¼tët-rumoh Possessor-raising host is undergoer
3Sg Asp 1SgA¼burn-house
‘I burned his/her house.’ (Literally, ‘Him/her, I house-burned.’)
(4.12) *gopnyan ka aneuk-woe *Possessor-raising host is actor
3Sg Asp child-return
‘His/her child returned.’
LS of woe: do’ (x, Ø) & BECOME be-back’ (x)
(x ¼ eVector/patient & actor)
Possessor raising can be said to be restricted to the undergoer argument,
which is to say the traditional O of transitive clauses and the undergoer S of
intransitive clauses. As noted in Section 2.1.3, argument-indexing pronominal
148 Role and Reference Grammar

forms are constrained by the A/Sa vs. O/So distinction and control of
inWnitival complements of verbs such as ‘want’ is constrained to work with
A/Sa pivots. Inasmuch as the O is always undergoer, the A is always actor and
the So vs. Sa distinction appears to be reducible to actor S vs. undergoer S
(Van Valin and La Polla 1997), the only distinction among arguments that
needs to be drawn for grammatical purposes is actor vs. undergoer. Since this
is viewed as a semantic distinction in RRG, no phenomena work in terms of a
privileged syntactic argument. Thus, from an RRG perspective, there are no
syntactic functions in Acehnese.8

4.2 Basic language types

Role and Reference Grammar’s characterization of the traditional subject
relation as a phenomenon-speciWc privileged argument type whose deWniton
can vary within and across languages yields a multi-faceted typology of
languages. The default privileged argument for a language can be semantically
deWned (actor only, for example) or partially syntactically deWned (actor or S,
for example). Moreover, since diVerent grammatical phenomena can work in
terms of diVerently deWned pivots and/or controllers, languages can have one
kind of privileged argument for head marking and dependent marking and
another for such syntactic phenomena as relativization and control of reXex-
ives, or one kind for head marking, another kind for dependent marking, and
several diVerent kinds for other syntactic phenomena. Before turning to the
question of how the traditional typology of grammatical-relation systems (see
Secton 3.2) is characterized in RRG, it is useful to have a better understanding
of how the main voice constructions are analysed.

4.2.1 Voice constructions across languages

The antipassive construction in Dyribal (see Figure 4.3) illustrates one of the
basic kinds of voice construction. There are two eVects: the default PSA is a
diVerent macrorole argument than in the active construction, that is actor
rather than undergoer, and the argument that would be the undergoer in the
active construction has a diVerent coding and status: it is a non-macrorole
dative core argument rather than an undergoer with absolutive case. Proto-
typical voice constructions, such as passive and antipassive, are characterized
in terms of which of these two kinds of functions it has:
(4.13) a. PSA modulation:
The default PSA is an argument of a diVerent kind than in the
active voice.
 Role and Reference Grammar 149

b. Argument modulation:
One of the active-voice macrorole arguments has a noncanonical
realization.
The antipassive construction in Dyirbal involves both PSA modulation and
argument modulation.
A syntactically ergative language can have both passive and antipassive
constructions; but only the antipassive involves PSA modulation. In Inuit,
for example, the undergoer in an antipassive construction is in instrumental
case, as is the actor in a passive construction, as illustrated by the following
examples adapted from Woodbury (1977),9 with an RRG analysis.
(4.14) a. Gimmix -p miiraqy kii-vaa Active clause
dog-erg child.Abs bite-Ind.Tr.3SgA3gO
‘The dog bit the child.’
default PSA: Undergoer; Macrorole arguments: Actor and under-
goer are direct core arguments (DCAs) (¼ default)
b. Gimmix miiraqy -mik kii-a-voq Antipassive clause
dog.Abs child-Instr bite-Antipass-Ind.Intr.3SgS
‘The dog bit a child.’
default PSA: Actor; Macrorole arguments: Actor only
c. Miiraqy gimmix -mik kit-tsip-puq Passive clause
child.Abs dog-Instr bite-Pass-Ind.Intr.3SgS
‘The child was bitten by a dog.’
default PSA: Undergoer; Macrorole arguments: Undergoer is a
DCA and actor is an adjunct
Invariant aspects of these three clauses:
LS of kii ‘bite’: do’ (x, [bite’ (x, y])
x ¼ actor; if there is an undergoer, y ¼ undergoer
As in Dyirbal, the antipassive construction has the actor marked absolutive
and functioning as the default PSA, whereas the default undergoer is realized
as a non-macrorole oblique argument assigned instrumental case. There is,
thus, both PSA modulation, that is the PSA is the actor rather than the
undergoer, and argument modulation, that is the undergoer macrorole is
suppressed and the argument that has this macrorole in the active voice is
noncanonically coded as an oblique argument with instrumental case. The
actor in the passive construction is analysed as an adjunct, that is a non-core
macrorole argument with instrumental case marking. There is, however, no
PSA modulation in the passive construction as the undergoer is the default
PSA, just as in the active construction.
150 Role and Reference Grammar

Unlike in Inuit, in English the passive construction involves both PSA

modulation and argument modulation, as illustrated in Figure 4.4. In a
passive clause, the actor is realized as an adjunct (or is omitted) rather than
a canonical direct core argument and the undergoer rather than the actor is
the default PSA.10 This diVerence in the eVect of passivization correlates with
the diVerence between ergativity and accusativity. Active clauses in an ergative
language have an undergoer default PSA, whereas they have an actor default
PSA in an accusative language. For English the generalization about the
default PSA is that it is the highest-ranking core macrorole argument, that
is the only macrorole argument (actor or undergoer) in an M-intranstive
clause, the actor in an active M-transitive clause, and the undergoer in a
passive clause, in which the actor is not a core argument.
All of the cases discussed thus far involve argument modulation, with or
without PSA modulation. It is also possible to have PSA modulation without
argument modulation. The Jarawara inverse-voice construction discussed in
Sections 2.2.3 and 3.3.3 provides an example from an accusative language. The
key fact is that the PSA (for conjunction reduction, verb agreement, and
consitituent order, among other things) is the actor in an M-transitive active
clause. In the inverse voice, the PSA for the same syntactic phenomena is the
undergoer. The actor, however, remains a direct core argument, occurring in
the same position as the undergoer in an active clause, and does not receive
noncanonical coding. Thus, the analysis of the following sentences (¼ (2.59)),
would be as shown in Figure 4.5.11
(4.15) a. Mioto Watati awa-ka Direct voice
Mioto(Masc) Watati(Fem) see-DeclMasc
‘Mioto saw Watati.’
SENTENCE SENTENCE

CLAUSE CLAUSE

CORE CORE PERIPHERY

ARG NUC ARG ARG NUC PP

PRED PRED
P NP
NP V NP NP V
The students like this teacher This teacher is liked by the students
voice? – active Actor Undergoer voice? – passive Undergoer Actor
∴PSA = Actor ∴PSA = Undergoer
like´(students, teacher) like´(students, teacher)

Figure 4.4. Active vs. passive voice in English

 Role and Reference Grammar 151

SENTENCE SENTENCE

CLAUSE CLAUSE

CORE CORE

ARG ARG NUC ARG ARG NUC

PRED PRED

NP NP V NP NP V

Mioto Watati awa-ka Watati Mioto hi-wa hi-ke

voice? – direct Actor Undergoer voice? – inverse Actor Undergoer

∴PSA = Actor ∴PSA = Undergoer
see´(Mioto, Watati) see´(Mioto, Watati)

Figure 4.5. Direct vs. inverse voice in Jarawara

b. Watati Mioto hi-wa hi-ke Inverse voice

Watati(Fem) Mioto(Masc) Inv-see Inv-DeclFem
‘Mioto saw Watati.’ (or, ‘Watati was seen by Mioto.’)
For Jarawara, as for English and Dyirbal, the semantic roles actor and under-
goer are assigned in the same way and the choice of default PSA is keyed to
voice. In RRG terms, there is no reversal of grammatical relations in Jarawara,
since actor and undergoer are semantic roles, which cannot be reversed. The
inverse voice (really just a version of passive voice) simply involves allowing
the undergoer to be the PSA, without any correlative argument modulation.
Some of the key grammatical constraints for Jarawara can be formulated as
follows (DCA ¼ direct core argument).

(4.16) a. The preferred constituent order is default-PSA DCA V.

b. The conjunction-reduction pivot is the default PSA.
c. The default PSA controls verb/mood gender agreement.

The default PSA for Jarawara is not simply the highest-ranking core macro-
role argument, as in English and most accusative languages, as the undergoer
in an inverse-voice clause is the PSA in spite of the fact that the actor is also a
direct core argument and actor outranks undergoer. Thus, a deWnition of the
default PSA is necessarily voice-contingent in Jarawara.
In a syntactically ergative language, a construction with only PSA modu-
lation would have the actor as default PSA and an undergoer macrorole
associated with a direct core argument that is indistinguishable in terms of
152 Role and Reference Grammar

marking and syntactic privileges from an undergoer in an active M-transitive

clause. The Austronesian language Sama is claimed to have precisely this kind
of antipassive construction in Van Valin and La Polla (1997: ch. 7) and Van
Valin (to appear: ch. 4).

4.2.2 Ergativity, accusativity, and split intransitivity

In a syntactically ergative language, such as Dyirbal or Inuit, the default PSA is
the undergoer of M-transitive active clauses (and passive clauses if these occur)
and the single macrorole argument of M-intransitive clauses. The basic gen-
eralization is that the default PSA is the lowest-ranking macrorole argument,
where actor > undergoer. If there is an an antipassive construction, the actor is
the default PSA in this construction, whether or not the undergoer macrorole
is suppressed (although it typically is). In a syntactically accusative language,
such as English or Jarawara, the default PSA is the highest-ranking core
macrorole, that is the actor in M-transitive clauses and the only macrorole
argument in M-intransitive clauses. If there is a passive (or inverse-voice)
construction the undergoer or some other argument is the default PSA,
whether the actor is still realized as a core argument (as in Jarawara) or not
(as in most languages). If such a language has an antipassive construction, it is
not a PSA-modulation construction, as it would only involve noncanonical
status and coding for the argument that would be undergoer in an active
clause.12 A syntactically split-intransitive language, such as Acehnese, has
only a semantic preferred argument: actor for certain constructions and
phenomena and undergoer for others. A pure language of this type does not
have a PSA-modulation voice, since it does not have a PSA. Thus, the three
most distinct types of languages in terms of how grammatical relations are
treated can be characterized as in Table 4.2.
Naturally, at least certain of the coding properties (head marking or
dependent marking) work on the basis of the same distinctions in languages
of these three types. Thus, Dyirbal, for example, marks only the actor in an
active M-transitive clause with ergative case. However, as noted above, since
syntactic functions are phenomenon-speciWc, it is fully expected that particu-
lar phenomena in any given language might work in terms of either particular
semantic roles (agent, for example), semantic macroroles (actor, for ex-
ample), or syntactic functions deWned in various ways (highest-ranking
macrorole argument, for example).
More generally, by allowing construction-speciWc pivots and controllers to
be deWned in various ways rather than assuming that grammatical phenomena
work in terms of a Wxed set of primitive syntactic functions, RRG accounts for
 Role and Reference Grammar 153

Table 4.2. Language type and main grammatical-relation categories in RRG

RRG
Language type Traditional Active voice (PSA-modulation voice)a

Accusative S/A Highest-ranking core (Non-actor argument)

macrorole argument
O Undergoer
Ergative S/O Lowest-ranking macrorole (Actor)
argument
A Actor
Split-intransitive A=Sa Actor
O=So Undergoer
a
The PSA-modulation voice column is in parentheses because only in languages such as Jarawara and Sama in which
there is PSA modulation without argument modulation do the traditional S/A or S/O grammatical-relation
categories require a more complicated characterization.

the fact that languages tend not to be strictly limited to ideal ergative, accusa-
tive, or split-intransitive systems, either with respcect to agreement and case or
with respect to other syntactic phenomena. At the same time, the prevalence of
accusative languages and the tendency for most languages to have some
phenomena restricted to an S-/A-type category of arguments is explicable,
insofar as one of the key functional motivations for grammatical-relation
constraints is to privilege the cognitively most salient argument. It is well
known that animacy and agency play an important role in various facets of
diVerential privileging in grammatical organization (Silverstein 1976, Delancey
1981, Aissen 1999b). In a transitive clause, the more agent-like argument is
more likely to be animate-referring, which yields a preference for privileging
the actor. In an intransitive clause, the animacy/agency preference is less
important, since there usually is only one argument available to be privileged.
Hence, an S-/A-type category is a natural candidate for privileged syntactic
status. Since accusative languages have an S-/A-type category as the default
PSA, it is not surprising that these are more common than ergative languages.
However, given the explanatory centrality of the actor and undergoer roles in
RRG, it is somewhat surprising that few languages seem to have pivots and
controllers systematically deWned in the most basic terms, that is simply in
terms of the actor vs. undergoer distinction. It would seem that languages like
Acehnese ought to be much more common than they are.

4.3 Some case studies

By virtue of the fact that the actor macrorole is not necessarily the same as the
default PSA and syntactic phenomena can be keyed to either semantic roles,
154 Role and Reference Grammar

such as actor, or argument classes deWned in various ways, RRG provides a

means of explaining many of the phenomena that motivate multistratal repre-
sentations of grammatical relations in Relational Grammar. The notion ‘initial
subject’, for example corresponds essentially to the actor macrorole. The notion
‘Wrst subject’ corresponds essentially to ‘highest-ranking macrorole argument’.
Even though RRG doesn’t recognize object syntactic functions as such, the
notions ‘direct core argument’ and ‘undergoer’ can be used, in conjunction
with voice and transitivity distinctions, to deWne language-speciWc categories
corresponding to direct object, 2nd object (or 2 chômeur), and the like, as well
as to generalize over various combinations of such categories.

4.3.1 Preverbal pronominal slots in Jarawara

Consider the following direct- and inverse-voice clauses in Jarawara (¼ (2.64)
with an RRG analysis).
(4.17) a. abay meey otaax kaba-haro otaax ama-ke
Wsh(Masc) 3Nsg 1NsgExcl eat-TnsFem 1NsgExcl extent-DeclFem
‘We were eating Wsh.’
do’ (x, [eat’ (x, y)]); x ¼ Actor; y ¼ Undergoer
Direct voice; default PSA: Actor
b. abay meey otaax kaba-haro meey ama-ke
Wsh(Masc) 3Nsg 1NsgExcl eat-TnsFem 3Nsg extent-DeclFem
‘We were eating Wsh.’ (or, ‘Fish was being eaten by us.’)
do’ (x, [eat’ (x, y)]); x ¼ Actor; y ¼ Undergoer
Inverse voice; default PSA: Undergoer
Since the actor is realized as a 1st person pronoun, leaving only the undergoer to be
realized as a full NP and since gender agreement in clauses with both preverbal
pronominal slots Wlled is the default feminine, the only indication of voice in these
two sentences is the pronoun preceding the secondary verb ama. The constraint is
that this pronoun is necessarily a manifestation of the default PSA, that is the actor
in (4.17a) and the undergoer in (4.17b). Although most morphosyntactic phe-
nomena that distinguish among macrorole arguments are keyed to the default
PSA vs. DCA distinction (see (4.16)), which varies with voice, the order of the
preverbal pronominals (mee and otaa in these examples) is determined instead by
the actor vs. undergoer distinction, independently of voice. That is to say, the Wrst
slot is for the actor and the second for the undergoer.

4.3.2 Passive and switch reference in Seri

In the accusative language Seri (see Section 2.2.1), a diVerent-subject (DS)
marker is used to indicate that the ‘subject’ of two related clauses (main and
 Role and Reference Grammar 155

adjunct, for example) is diVerent. If both of the clauses in question are in the
active voice, the default PSA (¼ highest-ranking macrorole argument), which
also determines ‘subject’ agreement, for example, is what counts for switch
reference. For a clause in passive voice, however, the default PSA (¼ undergoer)
does not count for purposes of switch reference. Instead, the actor, which is not
overtly realized, is what counts, as can be seen from the following examples:
(4.18) a. [[mi-naiL kom m-po-ki:xk (*ta)-x] ?ata:p
2Pos-skin the 2SgSubj-Irr-wet DS-Aux mucus
ko-m-si-a: ?a¼?a]
3Obl-2SgSubj-Irr-be Aux¼Decl
‘If you wet your skin, you will be with mucus.’ (i.e. ‘get a cold’)
Embedded clause: [do’ (2Sg, Ø)] CAUSE [BECOME wet’ (mi-naiL)]
Active voice; default PSA ¼ 2Sg (Actor)
Main clause: be-loc’ (?ata:p, 2Sg) [MR1]
Active voice; default PSA ¼ 2Sg (Undergoer)
b. [m-yo-a:?-kašni [kokašni šo m-t-a?o ma]]
2SgSubj-Dist-Pass-bite snake a 2SgSubj-Real-see DS
‘You were bitten, after you had seen a snake.’
Main clause: do’ (Ø, [bite’ (Ø, 2Sg)])
Passive voice; default PSA ¼ 2Sg (Undergoer)
Embedded clause: see’ (2Sg, kokašni)
Active voice; default PSA ¼ 2Sg (Actor)
DS marking is not keyed to the default PSA, since the DS marker must appear in
(4.18b) in spite of the fact that the default PSA is the same in both clauses. It is
also not keyed to the actor macrorole, since the DS marker does not appear in
(4.18a) because the actor PSA of the embedded clause is the same as the under-
goer PSA of the main clause. The RRG analysis of passive clauses as M-transitive
provides an explanation. In both passive and active clauses, the actor is what
counts for switch reference, not the PSA. Only in M-intransitive clauses is the
distinction between actor and undergoer neutralized for switch reference; what
counts is the only macrorole argument. The generalization is that DS marking
occurs if and only if the highest-ranking macrorole argument of the two clauses
is diVerent.13 Whether or not the macrorole arguments are part of the core is
irrelevant, since the actor macrorole is not even realized syntactically in (4.18b),
as indicated by the Ø argument in the eVector-role slots in the LS.

4.3.3 Dative subjects and quasi-subjects

The phenomenon of dative subjects in a language such as Icelandic can be
accounted for again by distinguishing the default PSA from the privileged
156 Role and Reference Grammar

argument for certain speciWc phenomena. Consider, to begin with, the RRG
analysis of M-transitive active and passive clauses with prototypical actor and
undergoer:
(4.19) a. Lögreglan tok Siggu fast.
Police.Nom took.3Pl Siggu.Acc fast
‘The police arrested Sigga.’
b. Sigga var tekin föst af lögreglunni.
Sigga.Nom was.3Sg taken fast by police.Dat
‘Sigga was arrested by the police.’
In (4.19a) the actor is marked with nominative case, determines verb agree-
ment, appears in preverbal position, and is the PSA for conjunction reduc-
tion, raising, control, reXexive antecedence, etc. (hence, the default PSA for
this construction). The undergoer, on the other hand, is marked with accusa-
tive case and does not have PSA behaviour. In (4.19b), which is a PSA-
modulation and argument-modulation passive construction, the actor is
realized as an adjunct marked with the prepositon af and the undergoer is
marked with nominative case, determines verb agreement, and is the PSA for
essentially the same range of phenomena as the actor is in the active con-
struction. Thus, Icelandic appears to have a typical accusative grammar,
including a passive PSA-modulation construction and rules for case and
agreement of the following form (based on Van Valin 1991).
(4.20) a. Nominative and accusative case assignment
The highest-ranking core macrorole argument (where actor >
undergoer) is assigned nominative case; the other macrorole ar-
gument, if there is one, is assigned accusative case.
b. Verb agreement
The Wnite verb agrees with the highest-ranking core macrorole
argument, if there is one; otherwise it is in a non-agreeing 3Sg form.
Certain verbs which in English have a typical actor–undergoer syntactic
frame have the lower-ranking argument marked dative rather than accusative,
as in the case of hjálpa ‘help’, which can be analysed as M-intranstive by lexical
stiupulation:
(4.21) Égx hjálpaði honumy . do’ (x, [help’ (x, y)] [MR1]; x ¼ actor)
1SgNom helped.1Sg 3SgMascDat
‘I helped him.’
Since hjálpa has an activity predicate in its LS, the single macrorole is actor.
The other argument is realized as a direct core argument and marked with
dative case, which is the default case for non-macrorole core arguments. Such
 Role and Reference Grammar 157

verbs can be passivized, in which case there is both argument modulation (the
actor is omitted or realized as a non-core adjunct) and PSA modulation
(the non-macrorole DCA is realized as the PSA).
(4.22) Mér var hjálpað (af honum).
1SgDat was helped by 3SgMascDat
‘I was helped (by him).’
Since there is no core macrorole argument, there is no nominative case
assigned and the verb is in a non-agreeing (default 3Sg) form. The preverbal
dative argument is, however, the PSA with respect to the usual ‘subject’-
sensitive phenomena (see Section 3.3.5). Thus, the possibility of dative sub-
jects arises for two reasons. First, certain verbs with two or more arguments in
their LS are nevertheless M-intransitive, which opens up the possibility of a
dative-marked core argument. Second, although nominative case and verb
agreement are keyed to the highest-ranking core macrorole argument, the
default PSA is not limited in the same way. In the passive construction, a core
argument can be the PSA, whether it is a macrorole argument or not.
Consider now the class of alternating psychological verbs discussed in
Sections 1.2 and 3.3.5, illustrated by henta ‘please/like’ in (4.23), which have
either a dative or nominative PSA for the usual subject-sensitive phenomena,
correlated with preverbal constituent order, as shown by the conjunction
reduction examples from Barðdal (2001) in (4.24).

(4.23) a. Ég veit að þettay mun henta mérx .

I know that this.Nom will.3Sg please 1SgDat
‘I know that this will be pleasing to me.’
b. Ég veit að mérx mun henta þettay .
I know that 1SgDat will.3Sg please this.Nom
‘I know that I’ll be pleased with this.’
(4.24) a. [þetta fyrirkomulag er ágœtt] og [ __ mun lı́ka henta
mér ágœtlega].
‘This arrangement (Nom) is Wne and will also please me (Dat) Wne.’
b. [Ég er ánœgð] og [ __ mun örugglega henta þetta ágœtlega].
‘I (Nom) am happy and will surely be pleased with this (Nom).’
(4.25) henta: like’ (x, y) [MR 1]; y ¼ undergoer
Given an analysis of the LS of henta as shown in (4.25) with a stipulation of
M-intransitivity,14 it follows that the single macrorole is undergoer and that it
is (by default) associated with the second, or theme-type argument, in
conformance with the actor–undergoer hierarchy (Figure 4.1). It also follows
158 Role and Reference Grammar

from the general case-marking and verb-agreement rules that the undergoer is
nominative and determines verb agreement. The other argument is a non-
macrorole core argument assigned the default dative case. The question is why
either argument can be the PSA for conjunction reduction and the various
other subject-sensitive phenomena. Building on an idea from Van Valin
(1991), who proposes that the possibility of alternative PSAs in ditransitive
passive clauses is due to the possibility of satisfying either the general prin-
ciple in (4.26a) or a preference for an undergoer PSA in the passive construc-
tion, the following potentially conXicting universal principles may be said to
be active in diVerent ways in diVerent languages.15
(4.26) PSA selection principles
a. The highest-ranking core argument with respect to the actor end
of the actor–undergoer hierarchy is the preferred PSA.
b. A macrorole core argument is the preferred PSA.
Principle (4.26b) is simply a more general way of stating the undergoer PSA
preference in passive clauses, since the undergoer is the only core macrorole in
a passive construction. In English, for example, (4.26a) and (4.26b) are both
active principles. In an M-transitive clause with two macrorole arguments
both (4.26a) and (4.26b) are satisWed only by making the actor the PSA. In any
clause with only one macrorole core argument, (4.26b) takes precedence,
guaranteeing that the undergoer is the only (or at least the least-marked)
PSA choice in a passive clause and that an oblique core argument never
trumps the undergoer for PSA selection in an active M-intransitive clause
(thus, This belongs to me and not *To me belongs this). Where Icelandic diVers
from English is that in cases where it is only possible to satisfy either (4.26a) or
(4.26b), neither systematically takes precedence. Example (4.23a) has the only
macrorole argument as PSA, in conformance with (4.26b). On the other
hand, (4.23b) has the highest-ranking core argument as PSA, in conformance
with (4.26a). This ambivalence with respect to the relative dominance of PSA
selection principles is also manifested in passives of ditransitive verbs such as
‘give’ and ‘show’, as illustrated by the following examples (from Zaenen et al.
1985 via Van Valin 1991)
(4.27) a. Bı́larnir voru sýndir henni.
cars.Nom were shown 3SgFemDat
‘The cars were shown to her.’
b. Henni voru sýndir bı́larnir.
3SgFemDat were shown cars.Nom
‘She was shown the cars.’
 Role and Reference Grammar 159

In the passive voice, verbs such as sýna ‘show’ have two direct core arguments.
The undergoer, which invariably controls verb agreement and is marked
nominative because it is the only core macrorole argument, can be selected
as the default PSA, in conformance with (4.26b), in which case it is clause-
initial and has all the correlative typical PSA behavioural properties. The dative
core argument can alternatively be selected as the default PSA, in conformance
with (4.26a), since as a locative-type argument it is higher on the actor end of
the actor–undergoer hierarchy than the theme-type undergoer.
RRG provides an apparently viable and interesting approach to so-called
dative subjects in Icelandic. However, dative ‘quasi-subjects’ of the kind found
more commonly in languages provide a diVerent kind of challenge, since
these show subject privileges with respect to certain phenomena, without,
however being the default PSA. Consider, for example, how the plaire class
of French psych verbs, exempliWed by the following example, might be
analysed.
(4.28) Cette femmey luix plait.
This woman 3SgDat-pleases, i.e. ‘This woman is peasing to him/her.’
like’ (x, y) [MR1]; y ¼ undergoer
As with Icelandic henta, plaire can be analysed as an M-intransitive verb
with only a state predicate in its LS. The single macrorole is undergoer and
is assigned to the second argument. The Wrst argument is realized as an
oblique core argument, wich is marked with the preposition à ‘to’ or, when
pronominal, expressed as a dative clitic. The verb necessarily agrees with the
undergoer, that is cette femme in (4.28), because it is the highest-ranking
core macrorole argument. Indeed, as the default PSA in all French construc-
tions is the highest-ranking core macrorole argument, only the undergoer in
this construction is the pivot or controller for most syntactic phenomena.
Put diVerently, for most syntactic phenomena, PSA selection principle (4.26b)
takes precedence over (4.26a) when the two are in conXict. This has the
eVect of limiting the PSA to a core macrorole argument. The problem is
that when it comes to control of certain kinds of adverbial phrases, either
argument of verbs in the plaire class can function as the controller (see
(2.103b) and the discussion in Section 2.3.1). The fact that PSA selection
can be not only construction-speciWc but phenomenon-speciWc as well pro-
vides a potential explanation. For control of adverbial phrases it is necessary
to say that the PSA is not the default PSA. Rather, either (4.26a) or (4.26b) can
be satisWed, with neither taking precedence with respect to PSA selection
for this phenomenon. In essence, Icelandic diVers from French only in that
the default PSA in Icelandic (i.e. the privileged argument for most syntactic
160 Role and Reference Grammar

phenomena) can be determined either by (4.26a) or by (4.26b), whereas only

with respect to a particular phenomenon does French allow this kind of PSA
alternation.

4.3.4 Applicative objects vs. quasi-objects in Indonesian

As noted in Section 2.3.2, Indonesian has a class of verbs of psychological and
emotional experience that have three diVerent active-voice frames for the
theme-type argument (i.e. the target or stimulus of the psychological or
emotional state), which is not the PSA. It can be realized as a bare NP with
a verb that bears applicative morphology, in which case it has all the privileges
of a prototypical O, which is to say that it can co-occur with a transitive verbal
preWx, be the PSA in a passive (i.e. PSA-modulation voice) construction, be
the controller of a Xoated quantiWer, and be the pivot in a relative clause. It
can also be realized as a bare NP with a non-applicative form of the verb, in
which case it has only two O-deWning properties (other than being a bare
NP), that is it can control a Xoated quantiWer and be a relative-clause pivot.
Finally, it can be realized in a PP, in which case it has none of the O-deWning
properties. Given that applicative morphology generally indicates a speciWc
choice of undergoer or the addition of an undergoer argument (e.g. Roberts
1995), it seems clear that the applicative case should be analysed as being M-
transitive, as illustrated by the following example.
(4.29) Alix sengan-i [rumah itu]y
Ali like/happy-Appl house that
‘Ali likes the house.’
LS of sengani: like’ (x, y); x ¼ actor, y ¼ undergoer
If a verb in this class does not bear applicative morphology, it is M-intransi-
tive by default. Since these are stative verbs, the single macrorole is undergoer
by general principle (see (4.4)). One pattern that occurs in many languages
with such M-intransitive psych verbs is that the undergoer macrorole is
associated with the lowest-ranking argument on the actor–undergoer hier-
archy, leaving the other argument to be realized as either a dative-marked
direct core argument or an oblique argument, as in the case of English The
team matters to me (matter’ (1Sg, team)) or plaire in French and henta in
Icelandic (Section 4.3.3). Under the second option, which also has the under-
goer functioning as PSA by virtue of being the only macrorole argument, the
more marked undergoer linking occurs, leaving the theme-type argument to
be realized as a direct non-macrorole argument or an oblique argument, as
with English psych adjectives such as pleased in I’m pleased with the outcome
 Role and Reference Grammar 161

(pleased’ (1Sg, outcome)). Indonesian appears to systematically opt for the

latter kind of linking, while allowing either an oblique or a direct-core
realization of the theme-type argument:
(4.30) a. Alix sengan dengan [rumah itu]y
Ali like/happy with house that
‘Ali likes the house.’
LS of sengan: like’ (x, y); x ¼ undergoer; y ¼ oblique
b. Alix sengan [rumah itu]y
Ali like/happy house that
‘Ali likes the house.’
LS of sengan: like’ (x, y); x ¼ undergoer; y ¼ DCA
Now, the three diVerent realizations of the theme-type argument account for
the diVerences and similarities between the constructions, given the following
constraints on the syntactic phenomena in question.
(4.31) Phenomena restricted to M-transitive clauses and/or undergoer
a. The default PSA in the passive voice is the undergoer.
b. A transitive preWx appears only on M-transitive verbs.
Phenomena restricted to direct core arguments
c. Only a DCA can be a relativization pivot.
d. A Xoated quantiWer controller must be a DCA.
Thus, the O-like behavioural properties of the bare postverbal NPs in sen-
tences like (4.30b) and (4.29) can be attributed to their status as direct core
arguments.16 The fact that only the (4.29) paraphrase allows transitive mark-
ing or passive voice can be attributed to the fact that only in the applicative
case is the theme-type argument an undergoer (yielding an M-transitive
clause). Sentences like (4.30a) show no O properties since the theme-type
argument is neither an undergoer nor a DCA.

4.3.5 Quasi-objects in Brazilian Portuguese

The quasi-object properties of the theme-type argument of the small class of
verbs with gostar ‘like’ as prototype in Brazilian Portuguese (see Section 2.3.2)
are less amenable to a straightforward RRG analysis. As in the case of psych
verbs in Indonesian, gostar is a two-argument stative verb. Since its theme-
type argument is marked with the preposition de ‘of, from’, it appears to be an
oblique non-macrorole argument, as with the de-marked argument of a verb
such as cansar ‘tire’ and unlike the bare NP argument of such semantically
similar verbs as adorar ‘adore’:17
162 Role and Reference Grammar

(4.32) a. Eu gostei de tudo isso.

I liked of all that
‘I liked all of that.’
like’ (x, y) [MR1]; x ¼ undergoer, y ¼ oblique complement
b. Eu cansei de tudo isso.
I tired of all that
‘I got tired of all of that.’
BECOME tired’ (x, y) [MR1]; x ¼ undergoer, y ¼ oblique com-
plement
c. Eu adorei tudo isso.
I adored all that
‘I adored all of that.’ (i.e. ‘really liked’)
adore’ (x, y); x ¼ actor, y ¼ undergoer
The problem is that unlike the de-marked oblique argument of cansar and
oblique arguments in general, the apparently oblique argument of gostar is
O-like (or undergoer-like) in its ability to be a pivot in the reciprocal/reXexive
se construction, the pseudocleft construction, and the tough-movement con-
struction, although it is oblique-like in being marked by a preposition and in
its inability to be a passive-clause ‘subject’ or an accusative pronominal clitic
(see Table 2.4 and the associated examples and discussion).
One possible approach would be to claim that the de-marked argument of
gostar is actually a non-macrorole DCA in spite of the fact that it is de-marked
and even though the de-marked argument of cansar, for example, would have
to be analysed as an oblique argument. One might claim for example that
the theme-type argument of gostar is really an ablative case-marked NP, that
the theme-type argument of cansar is a PP with an ablative preposition,
and that ablative case-marked NPs are direct arguments. The idea then
would be that passive-subject and accusative-clitic potential is limited to
undergoers, whereas the other apparently O-sensitive phenomena are allowed
to work with DCAs, which would include undergoers as well as the de-marked
arguments of gostar-type verbs. In other words, essentially the same approach
to quasi-object phenomena taken for Indonesian could be extended to Bra-
zilian Portuguese.
This kind of analysis is problematic, to begin with, in that it is at odds with
the standard RRG assumption that ‘direct’ is a meaningful designation, i.e.
only bare NPs (as in Indonesian) and NPs case-marked with a core case, such
as nominative, accusative, or dative in a language with case marking, are
direct arguments. A bigger problem is that the de-marked argument of gostar
shows evidence of being a PP of the same kind as the de-marked argument of
 Role and Reference Grammar 163

verbs such as cansar. There is a general constraint on across-the-board

topicalization illustrated by the following examples.
(4.33) a. Aquele carro, eu queria muito mas não comprei.
‘That car, I really wanted __ but didn’t buy __.’
b. Naquela loja, eu vi muita coisa interesante mas não comprei nada.
In that store, I saw a lot of interesting things __ but didn’t buy
anything __.
c. *Naquela loja, eu gostava muito mas não comprei nada.
In that store, I really liked __ but didn’t buy anything __.
d. *Daquela música, eu adorava mas cansei.
Of that song, I adored __ but got tired __.
The sentence-initial topicalized phrase in (4.33a) functions as a direct core
argument (undergoer) of the verbs of both the conjoined clauses. Similarly,
the topicalized PP functions as an adjunct in both of the conjoined clauses in
(4.33b). Example (4.33c) is ungrammatical because, although the topicalized
PP can be interpreted as an adjunct of the second clause, it cannot be
interpreted as the ‘missing’ theme-type argument of gostar in the Wrst clause.
Example (4.33d) is ungrammatical because a de-marked topicalized phrase
can be interpreted as the PP argument of cansar in the second clause but not
as the direct core argument of adorar ‘adore’. By the same token, the topica-
lized bare NP in (4.33d) can function as an argument of adorar but not as the
PP complement of cansar. In short, an NP cannot count as the PP argument
of a conjunct and a PP cannot count as the NP argument of a conjunct in an
across-the-board topicalization construction. Now, a topicalized PP with the
preposition de as its nucleus can function as the argument of both a cansar
clause and a gostar clause, as illustrated by the following example, suggesting
that the de-marked argument of gostar has the same status as the de-marked
argument of cansar, which is presumably an oblique object of a preposition.
(4.34) a. Daquela música, eu gostava muito mas cansei.
‘Of that song, I really liked __ but got tired __.’
Another approach, which doesn’t require drawing a dubious distinction
between oblique and direct PPs, would be to analyse the theme-type argu-
ment of verbs in the gostar class as an oblique/PP argument with the under-
goer macrorole. That is to say, instead of lexically stipulating that gostar is
M-intransitive with its Wrst argument as undergoer, it could be assumed to
be M-transitive (which is the default for 2-argument stative verbs). Its Wrst
argument would have the actor macrorole and its second the undergoer role,
in the same way as adorar (4.32c) or other stative psych verbs, such as amar
164 Role and Reference Grammar

‘love’, odiar ‘hate’, and entender ‘understand’. It would simply need to be

stiplulated that the second argument is realized in a PP with de as nucleus, in
spite of having the undergoer macrorole. By way of illustration of how this
analysis would account for the facts, consider the tough movement construc-
tion, whose pivot can generally be an undergoer, whether it functions as a
non-PSA direct core argument (4.35a) or as the PSA of an M-intransitive
accomplishment verb such as sair ‘come out’ (4.35b) or the PSA of a passive
clause (4.35c). It cannot, however, be an actor of an M-intransitive activity verb
such as gritar ‘shout’ (4.35e), or a non-macrorole oblique argument (4.35f–g).
(4.35) a. Pessoas como essas são difı́ceis de amar/odiar/entender.
‘People like those are hard to love/hate/understand.’
b. Essa mancha vai ser difı́cil de sair.
‘This stain is going to be hard to come out.’
c. Carros como esses são difı́ceis de serem encontrados.
‘Cars like these are diYcult to be encountered.’
d. Este lugar é difı́cil de não gostar.
‘This place is diYcult not to like.’
e. *Pessoas como essas são difı́ceis de gritar.
‘People like these are hard to shout.’
f. *Este lugar é difı́cil de cansar.
‘This place is diYcult to get tired (of).’
g. *Pessoas como essas são difı́ceis de dizer isso.
‘People like these are hard to say this (to).’
The constraint appears to be simply that the tough movement pivot must be
linked to the undergoer macrorole. The possibility of (4.35d) then follows
from the analysis of gostar as a verb with an exceptional linking of the
undergoer macrorole with an oblique argument. Similarly, the oblique argu-
ment of gostar can be the pivot in the o que pseudocleft construction because
being a macrorole argument is criterial, and it can be the pivot in the
reXexive/reciprocal se construction because being an undergoer is suYcient.
The fact that it cannot be the default PSA in the passive construction or an
accusative pronominal clitic, as illustrated by the following examples, can be
attributed to the fact that only an undergoer that is a direct core argument can
be realized as a passive-voice PSA or an accusative clitic.
(4.36) a. *Eu a gostei.
‘I 3FemSg-liked.’
b. *Essa música foi gostada por todo mundo.
‘This song was liked by everyone.’
 Role and Reference Grammar 165

One potentially big problem with this second approach to an explanation

for the quasi-object phenomenon in Brazilian Portuguese is that it challenges
the fundamental RRG assumption that the undergoer macrorole must be
associated with a direct core argument (e.g. Van Valin and La Polla 1997:
147). This brings out an interesting diVerence between the actor and under-
goer macroroles. The criteria for determining the presence of the actor
macrorole are semantic. For example, a verb with an activity predicate in its
LS always has an actor macrorole (linked to the eVector-type argument),
whether it is realized as a direct core argument (the default) or as an oblique
argument or adjunct, as in passive clauses. With the undergoer macrorole, on
the other hand, there is both a semantic constraint and a syntactic one:
(4.37) The undergoer role must be associated with
a. an argument that is lower than the eVector-type argument with
respect to the actor end of the actor–undergoer hierarchy, and
b. a direct core argument.
To the extent that there is a good reason for (4.37b), which isn’t entirely clear,
this second suggested analysis of Brazilian Portuguese quasi-objects poses a
problem.

4.4 Other functional theories

RRG is an example of what has come to be known as the functional approach
to grammar. In fact, the line between so-called functional and formal ap-
proaches is blurred by the rigorous formalization of RRG on the one hand
and the increasing incorporation of aspects of conceptual semantic structure
and discoure-pragmatic roles into such formal theories as Lexical-Functional
Grammar (see Section 3.4.1) and the Minimalist Program of Transformational
Grammar (see Chapter 5). In any case, from the perspective of grammatical
relations, functional approaches begin with a model of the lexical (concep-
tual) semantics of predicates and attempt to account for grammatical phe-
nomena as much as possible in terms of the semantic roles implicated in such
a model and other aspects of lexical semantic structure. Syntactic functions
such as the traditional subject, object, and indirect object relations are either
themselves given a (quasi-)semantic characterization or are used sparingly
and deWned on language-speciWc and/or construction-speciWc bases. Among
such theories are (Localist) Case Grammar and Lexicase Grammar, discussed
in Sections 1.2 and 1.3.4, Systemic-Functional Grammar (Halliday 1973, 1985),
Functional-Typological Syntax (Givón 1984, 1997), Construction Grammar
(Goldberg 1995), and Radical Construction Grammar (Croft 2001). By way of
166 Role and Reference Grammar

illustration, the remainder of this section provides brief overviews of the

approaches to grammatical relations of two such theories: Functional Gram-
mar and Cognitive Grammar.

4.4.1 Functional Grammar

Functional Grammar (FG) (Dik 1978, 1980, 1997, Siewierska 1991) shares with
RRG the general functionalist perspective on grammar and, thus, a concern
for explicating grammatical phenomena in terms of related semantic,
pragmatic, and discourse factors. Moreover, FG uses a similar layered clause
structure, with the verb and its arguments forming the core of a structure with
outer layers for adjuncts and question and focus operators and the like.
Semantic roles, called ‘semantic functions’, also play a critical role in
grammatical organization. They are conceived of as grammatically relevant
semantic inferences from the intersection of primitive properties of predicates
and their arguments. Predicates are classiWed according to the kind of state
of aVairs that they designate (e.g. state, position, accomplishment, activity,
change, dynamic process) following a deWnitional schema that is reminiscent
of the Vendler/Dowty schema underlying the LSs of RRG, although diVerent
in various respects. Semantic functions are deWned as kinds of entities in-
volved in certain ways in states of aVairs of diVerent kinds. For example,
the following semantic functions are among those recognized (Dik 1997:
118–20).
(4.38) Agent the entity controlling an activity or accomplishment
John (Agent) read a book.
Positioner the entity controlling a position
John (Positioner) keeps his money in a shoe box.
Force the non-controlling entity that instigates a process
The earthquake (Force) moved the rock.
Processed the entity that undergoes a process
The rock (Processed) moved.
the entity primarily involved in a state
The rock (Zero) was near the tree.
ProcExp the entity that experiences an experienced process
John (ProcExp) realized that it was raining.
ZeroExp the entity that experiences an experienced state
John (ZeroExp) was happy.
Goal the entity aVected or eVected by the operation of
some controller or force
John kissed Mary (Goal).
 Role and Reference Grammar 167

Recipient the entity into whose possession something is trans-

ferred
John gave the book to Mary (Recipient).
Direction the entity into whose possession something is trans-
ferred
John drove the car to Chicago (Direction).
Semantic functions are ordered into contrastive generalized categories in a
way that is reminiscent of the macroroles of RRG, although there are three of
them and the correspondence of even the closest two is not perfect. The idea is
that there can be as many as three arguments of a predicate and when there
are two or more they are ranked A1 > A2 > A3, following an algorithm such as
the following (Dik 1997: 120).18
(4.39) (1) (2a) (2b)
Agent Goal(Exp) Recipient(Exp)
Postitioner Location
Force Direction
Processed(Exp) Source
Zero(Exp)
1. In all predicate frames, A1 has one of the functions in (1).
2. In two-place predicate frames, A2 has one of the functions in (2a–b).
3. In three-place predicate frames A2 has the function in (2a) and A3 has
one of the functions in (2b).
In some languages, grammatical constraints are understood as working dir-
ectly in terms of either speciWc semantic functions or the generalized semantic
role categories A1, A2, and A3. For example, since Choctaw has no passive
construction that allows an argument of type (2a–b) to be ‘promoted’ to
subject, nominative case (see Section 2.1.3) would be said to be assigned to the
A1. Halkomelem, on the other hand, restricts the reXexivization pivot to
the goal (i.e. what is generally called ‘patient’ or ‘theme’) (see Section 2.2.4).
The only syntactic functions posited are subject and object, which are
assumed to reXect diVerent perspectives on the same predicate frame. The
subject is taken to deWne the primary perspective for the state of aVairs
designated by a given predicate frame. The object deWnes the secondary
perspective. The subject function only occurs in languages that have a passive
construction, or something equivalent. What the passive construction does is
allow a state of aVairs to be depicted from the primary perspective of some
entity other than the A1 in a predicate frame. Similarly, the syntactic function
object only occurs in languages that have an applicative (or ‘dative shift’)
168 Role and Reference Grammar

construction, which allows a state of aVairs to be depicted from the secondary

perspective of some entity other than the A2 in a predicate frame. Thus, the
following English sentences have their NPs associated with semantic ‘macro-
roles and syntactic functions in the way illustrated.
A1(subject) A2(object) A3
(4.40) a. Martha gave the book to Sam.
A1(subject) A3(object) A2
b. Martha gave Sam the book.
A3(subject) A2 A1
c. Sam was given the book by Martha
The subject and object notions correspond closely to the Wnal subject
and object of Relational Grammar (at least for semantically transitive clauses
in English, which are the only kind in which either the subject or object
function are assigned in FG). There is no real analogue of the traditional
indirect object notion, since the A3 category includes location, direction
(¼ goal of RRG and other theories), and source, for example. The notion
‘A3 object’, however, corresponds to indirect object (in English at least) for
those who take the term indirect object to refer only to the O1 in the double-
object construction (see Section 1.3.2). The quasi-object properties of the
second NP in sentences such as (4.40b) can be attributed to its status as a
non-object A2.

4.4.2 Cognitive Grammar

Cognitive Grammar (CG) (Langacker 1987, 1991, 1999) uses what can be
characterized as a billiard-ball model of event conceptualization. The basic
idea is that we conceive of the typical events designated by verbs as involving
entities that act upon each other in a force-dynamic manner causing move-
ment or some analogous change of state to occur. From the complex concepts
evoked by sentences such as The boy killed the insect or The man removed the
branches from the lawn the same basic kind of image schema of action is at
play, as illustrated in Figure 4.6. A participant or entity does something to
another entity (i.e. there is a targeted force-dynamic action) and because of
this the targeted entiy undergoes a change of state or location. What diVers
crucially with kill vs. remove, for example, is the precise nature of the eVect of
the targeted action: motion of the aVected entity out of some place with
remove vs. a change from alive to dead for the aVected entity in the case of kill.
At a schematic level of conceptualization, however, the basic action chain is
the same: an action results in a change (i.e. movement relative to places or
states) for the participant targeted by the actor.
 Role and Reference Grammar 169

Tr trajector
Lm landmark
entity
state/location
Tr Lm
movement/change
force-dynamic action

ALIVE DEAD PLACE

Tr Lm Tr Lm

KILL REMOVE

Figure 4.6. Cognitive Grammar action-chain schema and verb meanings

An important further aspect of the conceptualization of events is perspective

in the Functional Grammar sense discussed above or Wgure-ground organiza-
tion in the sense of Talmy (2000). One of the participants in a multiple-
participant event is necessarily more salient than the others, because the
event is conceptualized primarily from the perspective of this participant.
The same state of aVairs is often amenable to alternative conceptualizations
in which only the choice of primary participant, for which the term trajector is
used, varies. For example, The painting is above the table designates the same
objective state of aVairs as The table is beneath the painting. The choice reXects
diVerent perspectives: the entity whose location is at stake (a trajector) relative
to some other entity (a landmark) is allowed to vary. The clause-level trajector,
the default value of which is the head of an action chain, is grammatically
privileged (i.e. has the traditional ‘subject’ behaviours and properties) because
of its conceptual prominence. A Wnal key aspect of event conceptualization,
which can have an eVect on grammatical phenomena and grammatical rela-
tions, is proWling. Only certain aspects of an overall conceptual image schema
may be in relief in a particular sentence frame. For example, in the case of The
window broke, the overall concept ‘x did something to y and because of this y
became broken’ is activated at some level (just as it is in the case of The boy
broke the window). However, only the ‘y became broken’ part of the concept is
proWled. Since only a participant that is part of a proWled action chain can be
the trajector, the undergoer of the change in the event designated by The
window broke is conceived of as the trajector, as shown in Figure 4.7.
The CG equivalents of the RRG semantic macroroles are ‘head of an action
chain’ (¼ actor) and ‘tail of an action chain’ (¼ undergoer). In the case of
transitive break, the subject is the head of the action chain and the object is the
170 Role and Reference Grammar

BROKEN BROKEN

Tr Lm Tr

BREAK (transitive) BREAK (intransitive)

Figure 4.7. ProWling and trajector choice with transitive and intransitive break

tail. In the case of intransitive break, the subject is the tail of an action chain
(albeit a non-proWled one). The traditional subject notion corresponds to the
clause-level trajector in CG, which is systematically related to but not fully
determined by position in action chains, as the choice of trajector is construc-
tion-speciWc and language-speciWc to some extent. The most active proWled
participant in an event image schema (the head of a proWled multiple-partici-
pant action chain, for example) is the default trajector in most languages.
Passive and inverse voice constructions involve alternative conceptualizations
in which a non-default participant is the trajector. Syntactically ergative lan-
guages have the passive sort of conceptualization of events as the default: the tail
of a multiple-participant action chain, that is the undergoer of the main action,
is the default trajector. Although the trajector in the conceptual structure of a
clause is the dependent that is traditionally analysed as the subject, this is a
conceptually (i.e. semantically) deWned role in CG. The trajector is the entity
that is most prominent in the conception of a state of aVairs, by virtue of a very
general process of Wgure-ground conceptualization that is manifested in other
cognitive domains.
The CG approach to grammatical relations is particularly eVective in
explaining the crosslinguistic variablility and diachronic instability of gram-
matical relations with verbs of cognitive, perceptual, and emotional experi-
ence, and the related quasi-subject behaviour of dative experiencers in various
languages. It is clear that an emotional experience such as that designated by
the verbs like and please in English does not involve a force-dynamic action in
the same way as transitive verbs such as remove or kill. However, from the
perspective of grammatical relations the distinction is not signiWcant: both
like and kill are transitive and their A and O NPs have similar properties and
grammatical privileges. Although there may be no physically manifested
action with like and please, we conceive of the feelings as involving one or
more actions in the domain of ception, that is the domain of perceiving,
feeling, and cognizing (Talmy 1996), which can be represented in action chain
image schemas with a single-lined dotted arrow. To begin with, the feeling
itself is conceived of as a mental action focused on some entity that is
 Role and Reference Grammar 171

analogous to the aVected entity in a physical action. Moreover, the stimulus,

which is conceived of as giving rise to the feeing, ‘acts’ on the experiencer,
causing, in some sense, the feeling. Inasmuch as either participant can, in
principle, be construed as the primary actor, action proWling and trajector
choice can go in either direction, yielding the diVerence between (stative)
please and like, as shown in Figure 4.8. In the case of like, only the mental
action of the sentient participant targeting the stimulus is proWled, with the
result that the experiencer of the emotion is chosen as trajector, as it is the
head of a proWled action chain. The stimulus landmark is the tail of a proWled
action chain, which behaves analogously to the tail of a proWled force-
dynamic action chain with respect to grammatical properties and behaviour.
The stimulus and experiencer have inverse properties in the case of please,
since only the conceived action of the stimulus is proWled.
The highly abstract nature of the domain of ception and the partial Wctivity
of the conceived actions involved facilitate alternative construals of the
primary direction of what is an inherently symmetrical action situation. As
noted in Croft (1993), a natural consequence of this state of aVairs is that there
is both considerable crosslinguistic variation in grammatical relations with
verbs of cognitive and emotional experience as well as the potential for
diachronic instability. The dative experiencer construction, for example,
illustrated by French plaire (4.41a) (see Section 2.3.1) and the Old English
etymological source of like (4.41b) (Allen 1986), can have the analysis shown
in Figure 4.9.

FEEL GOOD FEEL GOOD

Tr Lm Tr Lm

PLEASE LIKE

Figure 4.8. ProWling and trajector choice with English please and like

FEEL GOOD

Tr Lm Tr Lm domain of sight

Old English lician SHOW

French plaire
Figure 4.9. ProWling and trajector choice with dative constructions
172 Role and Reference Grammar

(4.41) a. Cette femme plait à Pierre.

This woman pleases to Pierre. (i.e. ‘is pleasing to Pierre’)
b. Ge noldon gode lician.
you.NomPl wouldn’t.PlSubj God.Dat like
‘God wouldn’t like you.’ or ‘You wouldn’t be pleasing to God.’
The dative/oblique marking of the experiencer can be attributed to proWling
of both conceived actions in the emotional experience. Because of its proWled
secondary action, the experiencer is marked in the same way as the ‘indirect
object’ dependent of verbs such as show (e.g. I showed the pictures to him)
which also is a sentient participant in a secondary receptive action. The
experiencer argument of the verb plaire in French, which has the same
conceptual schema as Old English lician, does not trigger subject–verb agree-
ment or behave like a subject with respect to most phenomena because it lacks
trajector status. However, its cognitive prominence due to its animacy and
sentience and its heading an action chain motivate its subject-like behaviour
with respect to adverbial control.
According to Allen, Old English lician, like Icelandic henta (see Sections 1.2
and 4.3.3), allowed an alternative conceptualization in which the experiencer
had trajector status without, however, losing its dative marking. Thus, the
same alternation with respect to choice of trajector that occurs with henta and
other psych verbs in Icelandic was apparently possible in Old English as well.
The only diVerence in the image-schematic conceptualization for the dative
subject construction in Icelandic and Old English would be that the experi-
encer would be the trajector, for which reason it would have typical subject
privileges. The nominative case marking of the stimulus in this alternative
construction, although not a necessary consequence, would be conceptually
motivated by the proWling of the action chain headed by the stimulus. Over
time in English, accusative case marking for the stimulus and nominative case
marking for the experiencer became possible, presumably by analogical pres-
sure from the prototypical transitive action-chain scenario, and the nomina-
tive–accusative frame, with the associated asymmetrical action proWling
shown for like in Figure 4.8 eventually became the only possibility.
 5

Transformational Grammar

From the classical models (Chomsky 1957, 1965) through Government-Binding

(GB) theory (Chomsky 1981, 1986) and the more recent Minimalist program
(Chomsky 1995), (Generative-)Transformational Grammar (TG) has con-
cerned itself primarily with an understanding of constituent structure and the
implications of constituent structure for grammatical phenomena, including
case marking, agreement, and other phenomena inwhich grammatical relations
such as subject and object are implicated. The main idea is that representations
of sentence structure primarily show the syntactic categories of elements (i.e.
noun, verb, adjective, noun phrase, verb phrase, etc.), relations of constituency
and containment (i.e. in what way elements, including words and phrases, are
contained within larger phrases), and the linear order of constituents. Unlike in
Relational Grammar, Role and Reference Grammar, and most other theories,
both syntactic functions and semantic roles can and often do have a deWning
conWgurational instantiation. Unlike in other theories in which constituent
structure plays a central role, such as Head-Driven Phrase-Structure Grammar
(Pollard and Sag 1994), mismatches between linear order and grammatical
relation and/or grammatical-relation sharing, splitting, or ambiguity can be
attributed, at least in part, to transformations, that is movements of constituents
within structures. Having been the most widely used theory of syntax for some
time, it has been employed in analyses of countless phenomena in many
languages and consequently has remained in considerable Xux. Thus any
detailed characterization of the theory’s approach to grammatical relations is
necessarily era- and movement-contingent to some extent.

5.1 Basic design of theory

5.1.1 Classical TG
The key idea of constituent-structure analyses of sentences is that phrases are
built from words and other lexical and grammatical items which give them
their categorical status and determine, to a large extent, what other elements
must or may be contained in the same phrase (or clause). In the so-called
174 Transformational Grammar

Aspects model of classical TG (Chomsky 1965), also known as ‘standard

theory’, sentences are constructed following language-speciWc phrase-
structure rules, which specify the possible kinds of phrases, their possible
immediate constituents, and the linear order of all elements. Words drawn
from the lexicon can be placed freely into structures generated by the phrase-
structure rules, provided that the lexically-speciWed combinatorial constraints
of the words themselves (so-called subcategorization rules and selectional
restrictions) are not violated. Thus, for example, the verb please is lexically
speciWed to occur with a following animate-referring NP in the same VP as it,
the verb put is lexically speciWed to occur with both a following NP and a PP
in the same VP as it, and so forth. Language-speciWc transformational rules,
some of which are optional and others of which are obligatory, allow sentence
structures constructed by the base component of the grammar (phrase-
structure rules plus lexicon), known as a deep structure, to be modiWed
in various ways, to yield a surface structure.
By way of example, the phrase-structure rules shown in (5.1) together with
the lexical speciWcations of the verb put and other relevant words, sanction a
deep structure such as shown in Figure 5.1. An obligatory transformational
rule of ‘aYx hopping’ moves all verbal inXectional aYxes to the right of the
verbal element on which they appear. Thus, the deep structure of a tensed
clause will always be changed in some way to produce a surface structure.1
Moreover, because there is an optional passive transformation, which may
apply prior to aYx hopping, a single deep structure can have alternative
surface structures associated with it. The meanings of related passive and
active clauses are the same, for all intents and purposes, because they share the
same deep structure, which is where meaning is encoded.
(5.1) Abbreviated set of phrase-structure rules
a. S ! NP Aux VP
A clause, S, consists of a N(oun) P(hrase) followed by an Aux and a
V(erb) P(hrase).
b. VP ! V (NP) (PP)
A VP consists of a V followed by an optional NP and an optional
P(repositional) P(hrase).
c. NP ! Det (AdjP) N (PP)
An NP consists of a determiner followed by an N which optionally
has a preceding Adjective phrase and/or a following PP.
d. PP ! P NP
A PP consists of a P(reposition) followed by an NP.
 Transformational Grammar 175

S Deep structure

NP Aux
VP
the boy PST may
V NP PP
Affix hopping put
the candy P NP
in
The boy may PST put the candy in the bag the bag
Passive
S Surface structure
of active clause
The candy PST may be -en put in the bag by the boy
NP Aux
VP
Affix hopping
the boy might
The candy may PST be put -en in the bag by the boy
V NP PP
put S Surface structure
P NP
the candy in of passive clause
the bag NP Aux
VP

the candy might be

V PP PP
put
P NP P NP
in by
the bag the boy

Figure 5.1. Classical TG analysis of active and passive clauses

e. Aux ! Tense (Modal) (have þ -en) (be þ -ing)

An Aux constituent consists of a Tense element followed optionally by
one or more verbal auxiliary elements and any associated aYxes.
Concepts such as subject, object, patient, actor, etc. are given no formal
status in the representational system. Nevertheless, as noted by Chomsky
(1965), it is possible to deWne certain traditional grammatical relations in
conWgurational terms, that is in terms of position in a canonical sentence
structure, as shown in (5.2). Moreover, notions such as actor (or ‘logical’
subject) and patient (or ‘logical’ object) can be deWned by taking into account
the distinction between deep and surface structure. Thus, for example, the
logical subject of a passive sentence is the NP immediately dominated by S at
deep structure and the surface subject is the logical object or patient, which is
the NP immediately dominated by VP at deep structure.
(5.2) DeWnitions of syntactic functions
a. Subject ¼ NP immediately dominated by S
b. Object ¼ NP immediately dominated by VP
c. Oblique ¼ NP immediately dominated by PP
176 Transformational Grammar

5.1.2 The GB approach

The GB model of TG was developed with a view to eliminating language-
speciWc rules in favour of universal constraints with a limited range of
interlanguage variation. With respect to grammatical relations, it introduces
more elaborate ways of dealing with semantic roles and oVers some new ways
of deWning syntactic functions, while maintaining the idea that traditional
subject and object functions are primarily deWned in conWgurational terms.
Consider the GB-style analysis of a basic kind of active clause shown in
Figure 5.2.2 Following the constraints of so-called X-bar theory (JackendoV
1977, Chomsky 1986: ch. 2), which is a universally valid set of constraints on
phrase structures, phrases such as VP and possibly subphrases such as V’ are
built around heads that determine their category. The clause is headed by
inXectional elements (I ¼ inXection), which ultimately are manifested on an
auxiliary or main verb by a transformational process not shown here.3 Three
basic kinds of non-head constituents are recognized: speciWers, which occur
in the position immediately dominated by an XP (or maximal projection of
any category X), such as the clausal subject; complements, which occur in the
minimal (or lowest) X’ containing the head; and adjuncts (¼ non-arguments
of the head) which are adjoined to a phrasal category (typically an X’, as in the
case of the temporal adverbial on Thursday). Beyond this classiWcation,
complements can be either direct, as in the case of the pictures in the active
clause or indirect as in the case of my mother, which occurs in a (virtual) PP.
The primary deWnition of subject is ‘the NP in the speciWer position of IP’
and the primary deWnition of object is ‘NP sister of the verb’ (where two
phrases are sisters if they share the node immediately above them in the tree).
Thus, my brother is the subject and the pictures is an object in the active clause
in Figure 5.2. There are, however, other ways of deWning grammatical rela-
tions. For example, it is assumed that all NPs must be assigned asbstract case,
which may or may not be manifested morphologically. If I contains a Wnite
tense it assigns ‘structural’ nominative case to its speciWer position, where a
structural case is one of the cases indicating a primary grammatical relation
and which is assigned in a structure-speciWc way. Thus, ‘subject’ can also be
deWned as the nominative-marked NP. Because structural case is assigned to
positions, case marking and position generally yield the same deWnitions of
grammatical relations. However, there are cases where they diverge. For
example, in an inWnitival clause such as for me to lie in It would be crazy for
me to lie the subject, me, which occupies the speciWer of IP position in its
clause, is marked accusative by the preceding for. Generally, prepositions, like
verbs, assign accusative (or some other) case to their complement. Thus,
 Transformational Grammar 177

IP

NPi I´

my brother I VP

3Sg Pst V´

NOM V´
PP
PP NPj
V P´
give P´
the pictures P NP
P NPk on
Ø
Thursday
my mother
ACC

inherent ACC
Argument structure of give: <agent i, [patient j, recipientk]>

Figure 5.2. GB-style analysis of an English double-object clause

Thursday gets case from on in My brother gave my mother the pictures

on Thursday. Verbs have a structural accusative case to assign, as well (pos-
sibly) as ‘inherent’ or semantically-restricted cases such as dative, ablative,
instrumental, accusative, etc. In the case of the double-object construction,
the recipient is expressed in a PP with a null or abstract preposition which
incorporates into the verb (Baker 1988b). Such prepositions, which would be
overt incorporated applicative morphemes in languages such as Kinyarwanda
and Halkomelem (see Section 2.2.4), are unable to assign case to their object.
The structural accusative case of the verb is therefore assigned to the recipient
NP, accounting for its primary object status, including its adjacency to the
verb. Thus, ‘the NP that receives structural accusative case’ is another deWni-
tion of object. The recipient could also be deWned as an indirect object, that is
an object (in terms of case and behaviour) that is underlyingly in a PP. The
second object in the double-object construction is assigned an inherent
accusative case, accounting for its quasi-object properties (see Section 2.3).
Naturally, in clauses without a recipient or in which the recipient (or other
non-patient argument) is case-marked by a preposition, the patient is the
object both by virtue of being an NP compement of the V and by virtue of
being assigned the verb’s structural accusative case.
The argument structure (i.e. semantic/syntactic valence) of a verb (e.g.
Marantz 1984, Zubizarreta 1987, Grimshaw 1990), which can be derived from a
decomposition of verb meaning or conceptual structure similar to the ‘logical
178 Transformational Grammar

structure’ of RRG (e.g. JackendoV 1990), speciWes how many semantic roles
(or ‘theta roles’) it has and in what way they must be related to NPs in the
syntactic structure. Although the theta-role labels are not critical, labels of a
generic kind are used here for ease of exposition. Square brackets are used to
indicate one critical distinction: internal vs. external theta role. Theta roles
within square brackets must be aligned with NPs that are internal to the
minimal V’ projection of the V. Italicization is used to indicate an argument
that is realized as an indirect (or oblique) complement. The Wrst theta role,
which is outside the square brackets must be aligned with a constituent that is
external to the minimal V’ projection of the V—the default case being the
speciWer position of IP. Thus, theta-role assignment yields potentially distinct
notions of grammatical relations (construed broadly). The external argument
corresponds to the actor of RRG and the initial (or ‘logical’) subject of RG.
The direct internal argument corresponds to the initial (or ‘logical’) direct
object of RG.

5.1.3 The Minimalist approach

The Minimalist approach within TG has a somewhat diVerent basic concep-
tion of grammatical relations, as it implements a new technical mechanism
for handling case and agreement (i.e. the case and agreement features of NPs
are said to be ‘checked’ with matching features on heads) and introduces a
more elaborate clausal architecture that has certain substantive ramiWcations.
The deWning idea of TG, that is that constituents can move within the clause
structure, continues to play a key role. An NP, for example, may move from
one position to another, in which case it leaves behind a coindexed trace. In
eVect, an NP and its traces conjointly occupy more than one position and,
thus, have the grammatical properties associated with each of the positions.
There are four main new ideas with respect to grammatical relations. First,
following Kuroda (1988) and Koopman and Sportiche (1991), the so-called
external theta role is assumed to be associated with a speciWer position within
the verb phrase. Thus, the position of the ‘logical’ subject is dissociated from
the canonical clausal subject position (i.e. speciWer of IP), allowing, in eVect,
diVerent potential manifestations of subjects. Second, following ideas from
Larson (1988), Hale and Keyser (1993), and others, the traditional VP is given a
much more elaborate structure, with more possibilities for distinguishing
grammatical relations, both in terms of positions and in terms of case
relationships. Third, a more elaborate structure is also posited for the extra-
VP part of the clause, yielding more positions for NPs to move to. Fourth, in
addition to allowing structural case features of NPs to be checked in a
canonical way (i.e. the NP moves to a speciWer position to check its case
 Transformational Grammar 179

with a head), NPs can check their case features by moving (or having their
features move) virtually or covertly. This is accomplished by having a level of
representation known as logical form (LF),4 which is an abstract representa-
tion of the syntactic/semantic structure of a sentence which need not match
the phonologically-produced linear order. In eVect, grammatical relations in
the form of case may be partly dissociated from overt structural position.
By way of example, consider the Minimalist-style analysis of the simple
English clause shown in Figure 5.3, which adopts Baker’s (1997) analysis of
semantic-role positions coupled with the so-called ‘bare phrase structure’ and
multiple-speciWer framework (e.g. Chomsky 1995, Ura 2000, 2001). Inter-
mediate levels in the phrases are dispensed with where not needed explicitly
to introduce arguments or adjuncts and otherwise need not be labelled, as
they are predictably X’. The basic schema, however, is that a head (X) can
combine with a complement and one or more speciWers and/or adjuncts to
form a phrase XP. Although semantic roles can be seen as being projected
from an argument-structure representation (and an underlying conceptual
structure) as in GB, a common alternative view is that such additional levels
of representation are superXuous, given that semantic roles are instantiated
conWgurationally in a uniform way within and across languages (Hale and
Keyser 1993, Baker 1997).5 In essence, the syntactic structure is itself a (more or
less schematic) representation of verb meaning, such that at least the general
semantic role types of the referents of NPs can be read oV the structure.6 The
VP can be split into at least two levels: an outer vP and an inner VP. The v
receives an interpretation like ‘cause’ when it has a ‘transitive’ VP comple-
ment and the NP originating in its speciWer position is interpreted as referring
to an agent(-like) participant. The inner VP can have both speciWer and
complement arguments. The referent of the NP that originates in the speciWer
of VP position is interpreted as having a patient-type role (prototypically, its
referent undergoes motion or change) and the PP in the complement position
has a goal or other locative-type role (prototypically, its referent speciWes the
state, location, or participant to or from which movement or change occurs).7
The base structure, with NPs in their semantic-role positions, can be
modiWed by movement processes in several ways. The V, for example,
moves overtly from its base position to join with the higher v and in some
languages and under certain conditions to the I node (often analysed as
simply T, for tense). The need to check case and/or other grammatical features
(collectively ‘nominal features’) is one of the primary motivations for NP
movement. A nominative case feature, for example, is located in the head of
IP and a structural accusative case feature is located in the head of v. The head
of IP can also have agreement features (person/number, for example) and a
180 Transformational Grammar

IP

NPi
my brother vP
I

NOM/3Sg agent
EPP NP
ti v VP

patient
LF NP
ACC the children
V PP
took
goal
P NP
to the park

Figure 5.3. Minimalist-style analysis of a basic transitive English clause

so-called EPP feature, both of which can in eVect force the presence of an NP
in the speciWer position of IP, independently of or conjointly with the case
feature.8 The EPP feature is a remnant of the GB Extended Projection
Principle (Chomsky 1981), according to which all clauses must have a subject
(an analogue of the Final 1 Law of Relational Grammar). Being closest, the
agent moves to the speciWer of IP position to check the case, person/number
(agreement), and EPP features of the head. Movement to check features
cannot be covert (i.e. at LF—indicated by the dashed line in the diagram) if
one or more of the features of the head is speciWed as being ‘strong’, as in the
case of the EPP feature in English, which forces the NP destined to be the
traditional subject to move to a speciWer position of IP. Since the accusative
case feature and any other nominal features of v are ‘weak’ in English, overt
movement of the patient to the speciWer of vP does not occur. Instead, the
accusative case feature is checked by covert or LF movement—technically
only the features of the NP move to v (Chomsky 1995, Ura 2000).
One reason for saying that the v contains nominal features (including
accusative case) that attract an NP is that some languages allow objects to
appear in positions that suggest fronting to a pre-v position. In Icelandic,
which is a language in which tensed verbs move to I overtly, an object can
precede a subject that has not been moved out of its base position, as
illustrated by the following examples (Ura 2000: 52).
(5.3) það borðuðu [vp ostinni margar mýs [VP ti ]]
there ate the cheese many mice
‘Many mice ate the cheese.’
 Transformational Grammar 181

The Icelandic object shift phenomenon can be accounted for by assuming that the
nominal features hypothesized for v must be checked and may (in some lan-
guages) be checked by an NP that has overtly moved to a second speciWer position
of vP. The claim that this kind of overt object movement cannot take place in
English is motivated by the contrast between the following examples, in conjunc-
tion with the assumption that the verb is necessarily in the v position overtly.
(5.4) a. My brother might [v p send the children to the park].
b. *My brother might [v p the children send to the park].
Nevertheless, because the basic template of clause structure is assumed to be
essentially invariant across languages, the idea that there is a feature-checking
relationship between v and the object is maintained. One notion of direct
object is ‘NP in a nominal feature-checking relationship (but not a semantic-
role relationship) with v.’9
In short, the basic structure of transitive clauses (in accusative languages) is as
shown in Figure 5.4, with V movement ignored. There are, however, several
provisos. First, languages may diVer with respect to matters of linear order (e.g.
head before or after complement and speciWer to the left or right of the head).
Second, languages may diVer with respect to whether NPs may, must, or cannot
move overtly to check features with the appropriate heads. Third, languages can
vary with respect to whether obliques, such as a goal or source complement, are
realized as a PP, as in English, or as an NP (possibly inherently case-marked). The
traditional subject and object grammatical relations are ephiphonemal. The
theory yields a set of natural classes or categories of NP types, based on structural
positions and feature-checking relationships. Syntactic phenomena often work
diVerentially across and within languages in terms of these notions. For the sake of
theory comparison, another way of looking at the situation is that there are various
potential deWnitions of the traditional subject and objectnotions, as shownin (5.5).
(5.5) Potential deWnitions of subject and object
a. Logical subject ¼ base speciWer of vP
b. Subject ¼ checker of nominal features (case, agreement, etc.) of I
c. Subject ¼ speciWer of IP
d. Logical object ¼ NP in the highest base position in VP10
e. Object ¼ checker of nominal features of v

5.2 Typological variation

5.2.1 Ergativity
Largely in response to fundamental typological distinctions among languages
with respect to grammatical relations, TG (especially in its Minimalist
182 Transformational Grammar

IP

I vP

NOM, etc.
agent
NP
v VP

patient
NP
ACC, etc.
V (PP)

P NP location

Figure 5.4. Minimalist schema for transitive clause in accusative languages

manifestation) makes available a range of options with respect to possible

manipulations of what can be conceived of as a basic universal clause struc-
ture. Given the framework outlined above, an ergative language can be
characterized as one with only a couple of deviations from the accusative
schema, from which most of the other diVerences essentially follow. In a
syntactically ergative language such as Dyirbal (see Section 2.1.2), the S/O
category (i.e. the object/patient or subject of an intransitive) is the most
syntactically privileged. The patient in a transitive clause, thus outranks the
agent in terms of grammatical prominence. To the extent that grammatical
prominence means a checking relationship with I via the speciWer of IP
position, it is necessary to get the patient to move out of the vP. This can be
accomplished by attributing to ergative languages the ability to case-check the
agent in its base position. In essence, the only structural case of v is called
ergative and is necessarily checked with the base speciWer, unlike in the default
scenario in accusative languages. The agent still moves (or can move) to the
speciWer position of IP, where it can check the EPP feature of I. However, the
structural case feature of I (¼ absolutive) cannot be checked by the agent,
which has an ergative case that has been checked with v. Consequently, the
patient (which cannot check a case feature with v anyway) must move to a
second, higher speciWer of IP position to check the absolutive case feature and
any remaining nominal features of I. In a nutshell, the basic schema for an
ergative transitive clause is as shown in Figure 5.5 (abstracting away from V
movement, overt vs. covert movement, and language-speciWc linear order
diVerences).11
 Transformational Grammar 183

IP

I vP

NP agent
EPP
v VP
ABS

ERG NP patient
V (PP)

location
P NP

Figure 5.5. Minimalist schema for transitive clauses in ergative languages

In order to explain case marking with intransitive verbs in ergative lan-

guages, it is necessary to take into account the two main types of intransitive
verbs, for which a form of the Unaccusative Hypothesis (see Section 3.3.4) is
adopted. With unergative verbs, the vP part of the VP shell is present and the
agent dependent originates in the speciWer position of the vP. Although the
inner VP may contain an oblique argument, it does not have a patient
argument. With unaccusative verbs, since there is no agent there is no v and
therefore no vP. The patient argument originates within the VP, in the same
position as the patient in a transitive clause. In the case of unaccusatives, the
VP-internal NP must check the absolutive case of I, as that is the only way for
it to have its case feature checked and the only way for the nominal features of
I to be checked. In the case of unergatives, there are only two possible
scenarios: the agent NP checks an ergative case feature of v and I has no
case feature (that needs to be checked) or the v has no case feature and the
agent NP checks the absolutive case feature of I. In either case, the agent NP
may still move to the speciWer of IP position (to check the EPP feature of I, if
it is strong). Only in the latter case, however, does the language have a true
ergative case marking system. In the former, the language would have a split-
intransitive case marking system, since there would be a split among intransi-
tive verbs. Thus, the relative rarity of the ergative pattern can be attributed, in
part, to the unlikelihood of a language both allowing the v to enter into a
case-checking relation with its base speciWer (which might be characterized as
a marked option) and allowing the v to not have an ergative case in intransi-
tive clauses. The feature-checking and NP movment scenario for intransitive
clauses in an ergative language are illustrated in Figure 5.6.
184 Transformational Grammar

IP
IP

I vP
I VP
agent
NP
EPP, ABS, etc. patient
v VP EPP, ABS, etc. NP
[–ERG]

unergative clause unaccusative clause

Figure 5.6. Minimalist schema for intransitive clauses in ergative languages

For languages such as Dyirbal (Section 2.1.2), in which the absolutive NP

is the syntactically most privileged dependent (controller of conjunction
reduction, relativization pivot, controller of purposive clauses, etc.), the
highest-ranking grammatical relation can be deWned either as the absolutive
case-checker or as the highest speciWer of IP. Whatever subject-like (or ‘quasi-
subject’) properties the ergative NP may have can be attributed either to its
status as the EPP-feature checker or as a speciWer of IP.12 Inuit (Eskimo-Aleut)
is generally considered to be another example of a syntactically ergative
language (Woodbury 1977, Bittner 1987, Bittner and Hale 1996a, 1996b),
because the absolutive NP, for example, is the agreement controller in inWni-
tival clauses (in which only one NP triggers agreement), the controller of the
missing NP in subordinate adjunct clause constructions, and the relativiza-
tion pivot, as illustrated by (5.6) (from Bittner and Hale 1996b).

(5.6) a. [arna-p [_ ani-sima-su-p]] angut

woman-Ergi Abs pivoti go.out-Perf-Relat.Intr-Erg man
taku-v-a-a
see-Ind-Tr-3SgA.3SgO
‘The woman who had gone out saw the man.’
(5.6) b. [arna-t [_ miiqqa-p isiginnaa-ga-i]]
woman.Abs-Plj Abs pivotj child-Erg watch-Relat.Tr-3Sgi.Pl
mirsur-p-u-t
sew-Ind-Intr-3PlS
‘The women the child is watching are sewing.’
Nevertheless, the reXexivization controller is the S/A dependent (i.e. the
absolutive NP of an intransitive clause but the ergative NP of a transitive
clause) as indicated by (5.7) (from Ura 2000: 190).
 Transformational Grammar 185

(5.7) a. Junna immi-nut tatigi-v-u-q

Junaa.Abs self.Dat trust-Ind-Intr-3SgS
‘Junna trusts in self.’
b. Junna-p Kaali immi-nik uqaluttuup-p-a-a
Junna-Erg Kali.Abs self-Instr tell-Ind-Tr-3SgA3SgO
‘Junnai told Kalij about selfi= j .’
Thus, the reXexive controller must be the lowest speciWer of IP, whereas for
other subject phenomena the highest speciWer of IP is implicated.13
Morphologically ergative languages whose syntax is primarily accusative,
such as Warlpiri and Basque (see Section 2.1.2), are accounted for by assuming
that, in such languages, the patient NP does not move overtly to the speciWer
of IP position (because the absolutive case feature of I is ‘weak’). Thus, a
variant of the schema shown in Figure 5.5 has only the ergative NP in a
speciWer of IP position. The absolutive NP checks the absolutive case of I by
moving (its features) only covertly to I. With respect to ‘subject’-sensitive
phenomena, the ergative NP wins out over the absolutive in such languages
(yielding, in eVect, syntactic accusativity) because an NP in a speciWer of IP
position outranks an NP that only checks a nominal-feature of I at LF.
Naturally, the kinds of split-ergative coding properties (case vs. agreement,
for example) and split-ergative syntactic behaviour that many languages show
create a myriad of complexities that can be diYcult to explain. However, by
providing a myriad of possible syntactic-function deWnitions (e.g. LF
I-feature checker vs. overt I-feature checker, higher speciWer of IP vs. lower
speciWer of IP, case checker vs. person/number checker vs. EPP checker, etc.),
the Minimalist approach provides a means of navigating through at least
some of this complexity, while maintaining the view that there is a cross-
linguistically invariant conWgurational manifestation of basic generalized
semantic roles (agent in speciWer of vP position, for example) and some
kind of more or less invariant hierarchy of syntactic function privilege (higher
speciWer of IP > lower speciWer of IP > LF checker of I features, etc.).

5.2.2 Split intransitivity

A split-intransitive head-marking or dependent-marking language is one in
which the agent (or most agent-like) dependent is marked in the same way
and the patient (or most patient-like) dependent is marked in a diVerent way,
independently of the transitivity of the verb. From a TG/Minimalist perspec-
tive, this can occur if the case-marking and agreement properties of either the
agent-referring NP or the patient-referring NP (or both) are determined
on the basis of base position, given the kind of unaccusative/unergative
186 Transformational Grammar

distinction among intransitive clauses shown in Figure 5.6. Thus, under one
scenario, the ergative case of the v in an unergative clause is not suppressed
and is checked by the agent NP which then moves to the speciWer position of
IP to check the EPP feature of I (and any other nominal features of I, other
than absolutive, which must either be suppressed or, as suggested by Ura,
allowed to go unchecked). This would technically be a split-intransitive
ergative language, such as Georgian, for example, is claimed to be in Bittner
and Hale (1996a).
Although it is unclear that this basic analysis could not be extended in some
form to all split-intransitive languages, since a distinction between ergative
and accusative split-intransitive languages may be largely terminological, as
noted by Dixon (1994: 78), there is also a way of getting accusative split
intransitivity. It might be assumed that the agent always moves to the speciWer
of IP position to check all nominal features of I (including a nominative case
feature). The patient, on the other hand, is case-licensed internal to the VP,
perhaps because V can have an accusative case feature, analogous to the
ergative case feature of v in ergative languages. This kind of analysis (certain
details aside) is posited for Acehnese and Eastern Pomo (a Hokan language of
Northern California) in Bittner and Hale (1996a). Eastern Pomo provides a
nice example, inasmuch as it is a dependent-marking language in which the
case of the patient is overtly marked and the agentive case is zero (which is,
crosslinguistically, typical of nominative rather than ergative), as shown by
the following examples from Bittner and Hale.
(5.8) a. mı́ip mı́ip-al sáaka transitive
he.Nom him-Acc killed
‘He killed him.’
b. mı́ip-al xáa baakúma unaccusative
him-Acc in.the.water fell
‘He fell in the water (accidentally).’
c. mı́ip káluhuya unergative
he.Nom went.home
‘He went home.’
It is preferable to analyse the nominative NPs as subjects and the accusative
NPs as non-subjects (even in intransitive clasues like (5.8b)), insofar as the
nominative NPs of either transitive or unergative clauses count for purposes
of same-subject marking in subordination structures, whereas the accusative
NP of an unaccusative clause cannot yield same-subject marking with respect
to a nominative NP in a related clause, as shown by the following examples
(SS ¼ same-subject marking; DS ¼ diVerent-subject marking):14
 Transformational Grammar 187

(5.9) a. [háa káluhu-y], siimáa _ mérqakiihi

I.Nomi went.home-SS Ø(Nom)i went.to.bed
‘I went home, and then I went to bed.’
b. [háa qákki-qan] wi qaalál táala
I.Nomi took.a.bath-DS me.Acci sick became
‘I took a bath, so I got sick.’
Thus, the most straightforward analysis of transitive and intransitive clauses in
Eastern Pomo is as schematized in Figure 5.7, ignoring V movement and any
feature checking other than for case and EPP. It is, of course, necessary to
assume that I can have its nominative case suppressed or unchecked in unac-
cusative clauses, as must be assumed for absolutive case in unergative clauses on
the ergative split-intransitive analysis discussed above.15 For languages (such as
Kamaiurá and Choctaw, discussed in Section 2.1.3) in which the patient NP of
unaccusative clauses has clear subject-like properties, including (preferred)
clause-initial positioning, it could be assumed that this NP moves overtly to
the speciWer of IP position. Needless to say, since this approach to split
intransitivity is built around the Unaccusative Hypothesis and the problematic
assumption that universal grammar somehow yields the unaccusative vs. uner-
gative distinction automatically, it faces the same challenges as the Relational
Grammar approach discussed in Section 3.3.4.

IP

NPi
míip vP
I

agent
NOM, EPP NP
VP v
transitive clause ti

NP patient V
IP míip-al sáaka
IP
unaccusative clause
NPi ACC
I VP
míip I vP
NP patient
agent míip-al
NOM, EPP NP PP V
ti VP v xáa baakúma
unergative clause káluhuya EPP at LF
ACC

Figure 5.7. Three clause types in Eastern Pomo

188 Transformational Grammar

5.3 Some case studies

5.3.1 Multiple subjects and quasi-subject phenomena
Given that there can be multiple speciWers of a phrase, including either vP or
IP, and multiple (overt or covert) checkers of nominal features of I, certain
kinds of quasi-subject phenomena in languages can be attributed to the
secondary subjecthood of an NP. This is routinely the case for syntactically
ergative languages, as noted above, since these can have the ergative case-
marked NP in a lower speciWer of IP position and the absolutive case-marked
NP in a higher speciWer of IP position, for example. There is no reason why
this basic kind of structure could not occur in syntactically accusative lan-
guages as well, albeit only as a consequence of voice.
One obvious candidate for a multiple-subject analysis is the Seri passive
construction, discussed in Section 2.2.1. The interesting fact about Seri is that
although the patient in the passive construction has most of the ‘subject’
coding and behavioural properties, for the purposes of subject-sensitive
switch-reference marking only the implicit agent in the passive construction
counts as a subject. The following sentence, for example, shows that even
though the patient subjects of the main and subordinate clauses are diVerent,
the diVerent-subject marker cannot be used, since the implicit agents of the
two clauses are the same.
(5.10) [[?a t
t
s ki? p-a ?-kat
s
t
s (*ta)-x] ?e poL ki? mos
t
s

limberbush the Irr-Pass-seek DS-Aux ratany the also

si- a ?-ka
t
s
t
s ?a¼?a]
Irr-Pass-seek Aux¼Decl
‘If limberbush is looked for, white ratany should be looked for also.’
Under the analysis of the embedded passive clause shown in Figure 5.8,16 the
implicit agent, a null pronoun with an unspeciWed referent (proARB), occupies
the base speciWer of vP position, just as any agent-referring NP would in an
active clause.
What makes the passive construction diVerent from the active construction
is that the passive morphology on the verb causes the accusative case of v to be
transformed (eVectively) into a kind of ergative case, that is a case that must
be checked with the base speciWer of the vP. Although the agent NP moves to
the speciWer of IP position where it can check the EPP feature of I, it cannot
check the nominative case feature of I. The patient NP moves to the higher
speciWer of IP position to check the nominative case feature (and the other
nominal features of I). As in the case of reXexive control in Inuit (Section
 Transformational Grammar 189

IP

NOM, etc.
NPj
a:t ki: NP EPP
i
proARB
vP I

ERG
NP agent
ti VP v
PASS

patient V
NP p-a: -ka:
tj

Figure 5.8. Passive clause in Seri

5.2.1), the controller of switch reference in Seri must be the NP in the lowest
speciWer position of IP (i.e. the Wrst subject). This is a natural result insofar as
switch reference seems to be incorporable into a more general theory of
anaphora (Finer 1985, Enç 1989) and the same kind of condition apparently
must be an available option for reXexive control in some languages.
The Jarawara reversal construction and inverse voice phenomena more gener-
ally (see Section 2.2.3) are also amenable to a kind of multiple-subject analysis,
insofar as ‘base speciWer of vP’ is a kind of subject. Following Ura’s (2000) analysis
of inverse voice in Kinyarwanda, the following clauses, which exemplify the direct
and inverse constructions of Jarawara, can be analysed as shown.
(5.11) a. [IP Miotoi [vP Watatij ti [VP tj wa-ka] v] I]
Mioto(Masc) Watati(Fem) see-DeclMasc
‘Mioto saw Watati.’
b. [IP Watatii [vP ti Mioto [VP ti hi-wa] v] hi-ke]
Watati(Fem) Mioto(Masc) Inv-see Inv-DeclFem
‘Mioto saw Watati.’ (or, ‘Watati was seen by Mioto.’)
In the direct voice, the agent originates in the speciWer of vP position but
cannot check its case with the v and therefore moves to the speciWer of IP
position, where it checks the nominative feature of I, as well as any other
nominal features of I (including the EPP feature and the gender feature that is
manifested morphologically in the gender agreement suYx). The patient
moves, presumably overtly, to the speciWer position of vP to check the
nominal features of v, including its accusative case feature. In the inverse
voice, the accusative case of the v (which could be called ‘ergative’) is assigned
190 Transformational Grammar

to the base speciWer of vP due to a voice feature on the v, which is sometimes

morphologically manifested on the amalgamated verb (assumed to move to v
overtly, in a way not shown here) as well as on the auxiliary (assumed to be
overtly in the I position). Since the patient NP, which may still move to the
higher speciWer position of vP cannot check its case with the v, whose only
case is assigned to the agent NP, it moves higher up to the speciWer position of
IP, where it checks all the nominal features of I, including the nominative case
feature. Thus, the apparent syntactic-function reversal is due to the fact that
in each voice one NP moves to the speciWer position of IP and one stays in a
speciWer of vP position. The former has the subject properties and privileges,
the latter has the O properties. However, the underlying subjecthood of the
agent (¼ NP in the base speciWer of vP) and the underlying objecthood of the
patient (¼ highest base argument of the VP) can be used to account for the
constraints on the pronominal forms that show up in cases where the features
of the NPs in question are other than 3rd singular.
As indicated by the following examples (¼ (2.64)), preverbal pronouns
indexing the agent and patient show up in a Wxed order (patient < agent),
independently of voice. Moreover, an accusative suYx can appear on the
patient-referring pronoun only in the direct voice and a copy pronoun of
the nominative NP shows up in a pre-auxiliary, or postverbal position.

Direct voice; feminine suYxes because both pronominal slots are Wlled
(5.12) a. aba mee(-ra) otaa kaba-haro otaa ama-ke
Wsh(Masc) 3Nsg(-O) 1NsgExcl eat-TnsFem 1NsgExcl extent-DeclFem
‘We were eating Wsh.’
Inverse voice; feminine suYxes because both pronominal slots are Wlled
b. aba mee(*-ra) otaa kaba-haro mee ama-ke
Wsh(Masc) 3Nsg(-O) 1NsgExcl eat-TnsFem 3Nsg extent-DeclFem
‘We were eating Wsh.’ (or, ‘Fish was being eaten by us.’)

The proposed structures of the two clause types yield a natural account of
both the positions and the form of the pronominal elements, if they are
analysed as non-moving adjuncts that are coindexed with the Wrst speciWer
of the verbal/auxiliary heads to which they attach, as illustrated in the analysis
shown in Figure 5.9.
Dative subject phenomena in languages such as Icelandic as well as dative
quasi-subject phenomena in other languages (see Section 2.3.1), also appear to
be amenable to some version of a multiple-subject analysis (e.g. Belletti and
Rizzi 1988, Ura 1999). Consider, for example, the following schematic analyses
of Icelandic dative passive clauses (¼ (4.27)).
 Transformational Grammar 191

IP IP

NPi NOM, etc. NPj NOM, etc.

pro aba
vP I vP I

otaai I meej I
NPj ama-ke NP ama-ke
aba agent ACC
NP tj NPi agent ACC/ERG
pro
VP v VP v
ti
otaai v otaai v
patient V DIRVoice patient V INV Voice
NP NP
mee-raj V meej V
tj kaba-haro tj kaba-haro

Direct voice clause Inverse voice clause

Figure 5.9. Direct vs. inverse voice in Jarawara

(5.13) a. [IP Biilarniri voru [vP sýndir [VP ti tv henni]]]

cars.Nom were shown 3SgFemDat
‘The cars were shown to her.’
b. [IP Hennii voru [vP sýndir [VP bı́larnir tv ti ]]]
3SgFemDat were shown cars.Nom
‘She was shown the cars.’
It is assumed here that the passive ‘be’ auxiliary is overtly in I and the main
verb moves to the v position overtly and that, due to the passive morphology,
the implicit agent is suppressed and the accusative case of the v is taken away,
although it is unclear how crucial these assumptions are. The recipient NP has
an inherent dative case by virtue of its semantic role and base position.
As such, it cannot check the nominative case feature of I. The patient NP
must check the nominative case feature and person/number (agreement)
features of I. Since this is possible without overt movement, as shown by
(5.13b), the case and person/number features of I must be able to be ‘weak’,
although the EPP feature must be strong, since some NP must move overtly
to the speciWer of IP position. There are two possibilities, the dative NP moves
overtly to the speciWer of IP position, as in (5.13b), in which case it has all
the typical ‘subject’ privileges associated with this position. The patient
NP covertly checks the case and person/number features of I, for which reason
it is the ‘subject’ only with respect to case and agreement morphology.
The other possibility is that the patient NP simply moves overtly to the
speciWer of IP position and checks all the nominal features of I, as in
(5.13a), and has both the coding and behavioural properties of the ‘subject’.
192 Transformational Grammar

Although why the two possibilities should exist in Icelandic is not entirely
clear, the grammatical properties of the two constructions are accounted for
straightforwardly.
The psych verb alternation discussed in Sections 1.2, 3.3.5, and 4.3.3 can
presumably be handled in essentially the same way (modulo whatever diVer-
ences may be attributable to the base positions of the arguments of psych
verbs and the lack of an agent role). Other languages may allow the dative NP
in certain dative constructions to move overtly to IP, without, however,
gaining all the subject privileges of the language by virtue of not checking
the case and person/number features of I. For languages like French, which do
not allow the ‘dative’ NP in psych verb constructions to move overtly to the
speciWer of IP position at all and restrict its subject properties to limited kinds
of control of adverbial phrases and the like, it may be that the dative NP is
subject-like only by virtue of covertly checking some feature of I (although
why such should occur at all is not entirely clear).
An interesting question that arises is whether the passive construction in all
languages might be analysed in a way similar to the proposed analyses of
either the Seri passive or the Jarawara inverse, or whether such a possibility is
at least more widely utilized. It is commonly assumed, following the inXuen-
tial analyses of Chomsky (1981), Jaeggli (1986), and Baker et al. (1989), that
passive morphology has the eVect of not only suppressing (or itself taking) the
accusative case otherwise destined for the patient NP, but also causing the
agent NP to not be introduced in its canonical position (i.e. a speciWer of vP
position in the Minimalist framework). If the agent semantic role is implicit,
it is associated with the passive morphology rather than with an NP; if the
agent is expressed as an oblique, it is an adjunct that shares the agent semantic
role by virtue of being necessarily coindexed with the supposedly nominal
passive morphology. This sort of analysis is problematic from the point of
view of the Uniform Thematic Alignment Hypothesis (Baker 1988a, 1997),
insofar as the agent semantic role is not uniformly associated with the base
speciWer of vP position. An associated problem is that it is not clear why
passive agents have quasi-subject properties in many languages. Consider, for
example, the fact that the pivot in the raising-to-object construction in
Halkomelem is restricted to the S/A (subject) or the otherwise oblique
agent of the embedded clause (see Section 2.2.1). What this means is that
either the absolutive NP of a passive clause or the agent can show up as
the (absolutive) O of the main clause, as illustrated by the following examples
(¼ (2.42)):
 Transformational Grammar 193

S of embedded passive clause = O of xec-

(5.14) a. ?i c n xe?xcí-t
e kwθ Bob
e
Aux 1Subj wonder.Cont-Tr Det Bob
[?u ?i-? se ?l m-? t- m? ? -y ’
e e e e e John ∅]
Lnk Aux-3SgSubj look-Tr-Intr Obl-Det John
‘I’m wondering if Bob is being watched b John.’

Agent of embedded passive clause = O of xec-

b. ?i c n xe?xcí-t
e kwθ John
e
Aux 1Subj wonder.Cont-Tr Det John
[?u ?i-? s e l ?l m-? t- m? ∅ kwθ Bob]
e e e e e
Lnk Aux-3SgSubj look-Tr-Intr Det Bob
‘I’m wondering if Bob is being watched by John.’

Given that raising from oblique does not otherwise occur in Halkomelem and
seems to be unprecedented from a crosslinguistic perspective, it is worth
wondering whether the agent NP in the passive construction has something
structurally in common with absolutive subjects. For example, it might be that
the agent NP is invariably introduced in the speciWer of vP position and can
perhaps move to a speciWer of IP position, which may be a prerequisite for
raising into a higher clause. Unfortunately, exactly how such an analysis could
be made to work with respect to all details remains unclear. For example, it is
not clear how it might be guaranteed that the passive agent is necessarily
marked like an oblique when the patient raises to check the absolutive case
of the higher I (in a simple main clause, for example) and yet nevertheless need
not be oblique in principle, in that it too can check the absolutive case of the I in
a higher clause. In a theory such as Relational Grammar in which the necessity
of checking case features determines neither syntactic function nor position in
a clause, the analysis is more straightforward. Agents are always initial subjects.
In passive clauses they are simply necessarily demoted in some way. Although
there may be a universal constraint on raising constructions limiting the pivot
to subjects, neither position nor case is relevant to the precise condition in
Halkomelem. Rather, the pivot must simply be a subject in some stratum.
5.3.2 Applicative constructions and the indirect object notion
There have been a number of diVerent TG approaches to applicative and
double-object constructions. If the traditional indirect object notion has a
conWgurational characterization, it may only be manifested in these construc-
tions. Consider, for example, the following examples from Halkomelem
(¼ (2.70a–b)).
194 Transformational Grammar

(5.15) a. ni ?ám-@s-t-@s kwu@ sqw@m@
y? ?@ kwu@ su’ám?

Aux give-Recip-Tr-3A Det dog Obl Det bone
‘He gave the dog the bone.’
b. ni q’w@
l-@Lc-t-@s L@ sLéni ?@ kwu@ s@plı́l
Aux bake-Ben-Tr-3A Det woman Obl Det bread
‘He baked the bread for the woman.’
These are transitive clauses, as there is a transitive suYx on the verb and there
is an agreement marker for the 3Sg A (expressed as a null pronoun in these
cases), which only occurs in transitive clauses. However, the patient is not an
O, as evidenced by the fact that it occurs in a PP with the all-purpose oblique
preposition and has none of the behavioural properties of an O. The recipient
NP in (5.15a) and the beneWciary NP in (5.15b), on the other hand, behave like
the O of an ordinary transitive clause, not only in lacking a preposition and
controlling O agreement (when not third person), but also by their behaviour
with respect to passivization, quantiWer Xoat, possessor extraction, and vari-
ous A-/S-/O-restricted syntactic phenomena, as discussed in Section 2.2.4.
Being an ergative language, the agent only needs to check its case with the v
and need not move overtly. Since the preferred word order for the NPs, when
overt, is A < O and the verb immediately follows the clause-initial auxiliary
(which may be assumed to occupy an initial complementizer position), the
simplest analysis is that the verb overtly raises (through all verbal head
positions) to I and the NPs remain within the vP, which is the result of all the
nominal features of I being weak. The higher grammatical prominence of the
applicative O can be attributed to its being introduced in a slot between the v
and the VP. In the case of (5.15b), for example, the benefactive applicative aYx
can be analysed as a verb with a beneWciary argument, as shown in Figure 5.10.
Following Ura (2000), the intermediate verbal category is simply labelled Vmid
here.17
The motivation for introducing the applicative O in a higher slot within
the extended verb phrase is that it has primary O properties crosslinguisti-
cally, independently of whether the patient is treated like an oblique or
not. The primary O properties are necessarily associated with the applicative
O because, by virtue of its higher base position, it must necessarily be the
NP that moves overtly or covertly to check the absolutive case feature of I
(in an ergative language) or the accusative case feature of v (in an
accusative language). In any case, the traditional indirect object grammatical
relation can be deWned as the NP which originates in the speciWer position
of the VmidP, although ‘primary object’ would presumably be a better desig-
nation.
 Transformational Grammar 195

CP
IP
C
ni
I vP

NP agent
EPP pro
v VmidP

NP ben
ERG e s éni VP
Vmid
e- c
ABS PP
V
q’w l
e patient
P NP
kwθ s plíl
e e

Figure 5.10. Benefactive applicative construction in Halkomelem

With respect to benefactive and recipient applicative constructions, English

diVers from Halkomelem (independently of ergativity) primarily in that the
patient is (at least) a quasi-O (see Sections 1.3.2 and 2.3) and the Vmid is not
associated with phonological content in the form of an applicative aYx. The
double-object construction might be analysed as shown in Figure 5.11 (based
on Ura 2000). The applicative O is the primary O since it checks the
accusative case of v and is in a higher position in the tree. The Vmid has a
weak accusative case feature which the patient NP can check covertly,
accounting for its quasi-O properties. Languages in which both the applica-
tive O and the patient manifest primary O properties (i.e. symmetric appli-
cative languages), including perhaps dialects of English that allow promotion
of either the patient or the applicative O in the passive construction, might be
characterized as allowing overt movement of the patient to a speciWer position
of the VmidP, which would give it conWgurationally equivalent status with the
applicative O.
One potential drawback of such an analysis is that it is unclear how the
relationship between the double-object construction and the alternative con-
struction that has the recipient in a PP with to might be accounted for. If it is
assumed that the recipient PP originates in a lower position within the VP
than the patient and the patient checks the accusative case feature of v, it is
necessary to either abandon the default assumption that semantic roles are
associated with initial positions in a uniform way (e.g. Baker 1988a, 1997, Hale
and Keyser 1993) or to entertain the problematic hypothesis that give the boy a
bike and give a bike to the boy have a suYciently diVerent meaning as to
196 Transformational Grammar

vP

NP agent
v VmidP

NP recip
the boy VP
Vmid
patient
ACC NP
V
a bike
give

ACC

Figure 5.11. Double-object construction in English

warrant claiming that the semantic roles are diVerent. On the other hand, if it
is assumed that the recipient semantic role is always associated with the
speciWer position of the VmidP and what diVers is only whether the prepos-
ition to is present (to case-license the recipient NP or perhaps as a manifest-
ation of an inherent dative case), some ad hoc stipulation is required to
guarantee overt movement of the patient NP to a higher position than the
recipient just in case the latter is a PP.
Another alternative would be to have the recipent always originate in a PP
in a lower position than the patient within the VP and have it move to a
higher position in the tree, perhaps a higher speciWer of VP position, just in
case the preposition is null, essentially following the GB-style analysis shown
in Figure 5.2, which is based on the analysis of applicative constructions
proposed by Baker (1988a, 1988b). Particularly for constructions such as the
goal applicative construction in Halkomelem, illustrated by the following
examples (¼ (2.73c–d)), the P incorporation approach seems to be the most
straightforward.
(5.16) a. ni ném ?@-l’ John


Aux go Obl-Det John

‘He went to John.’
b. ni ném-n-@s-@s kwu@ John
Aux go-Goal-Tr-3A Det John
‘He went to John.’
Assuming that semantic roles have a uniform conWgurational instantiation
crosslinguistically, the goal argument must be a PP complement of the V. One
need only say that if the P that indicates the goal semantic role takes the form
 Transformational Grammar 197

of a verbal aYx, the NP must enter into an absolutive case-checking relation

with I (in an ergative language like Halkomelem) and therefore has the
properties and privileges of an O.
Needless to say, the combination of diVerent potential ways of character-
izing PP/NP alternations, various potential base and target postions for NPs
within an extended VP with uncertain limits, and the possibility of both overt
and covert movement, leads to the possibility of numerous alternative ana-
lyses of applicative phenomena.

5.4 Related theories

The connections between TG and Relational Grammar and Lexical-Func-
tional Grammar, both of which diVer from TG in that they divorce syntactic
functions from structural positions, are discussed in Section 3.4. One key idea
of the Mimimalist approach is that the amalgamated verb (i.e. V plus v and
possibly Vmid) combines with its dependents one at a time, starting with the
most oblique and ending with the subject/agent, in such a way as to build a
complex structure that reXects degrees of obliqueness. This idea has its roots
in Categorial Grammar (e.g. Dowty 1982, Oehrle et al. 1988), which attempts
to have syntactic phrase structure representations mirror Montague-style
semantic composition (Dowty et al. 1981). The Minimalist device of feature
checking as a way of accounting for case and agreement, among other things,
is analogous to the device of uniWcation in a family of formal theories of
which Head-Driven Phrase Structure Grammar (Pollard and Sag 1994, Sag et
al. 2003) is the most well known. In this theory, a verb, for example, has a
bundle of semantic and grammatical features, including possibly features
concerning the person, number, and case of the items with which it can
combine. The features of all words are represented in tree structures. In a
well-formed tree structure the grammatical features of the arguments of a
verb must match the feature speciWcations of the verb, for example. Technic-
ally, the feature content of the constituents of phrases can ‘unify’ at the
phrasal level only if there are no incompatibilities. In terms of grammatical
relations, Head-Driven Phrase Structure Grammar, by and large, shares with
TG the view that position in phrase structure trees determines (or correlates
with) syntactic functions. Although verbs bear a speciWc ‘subject’ feature that
licenses a constituent in the phrase structure with a special function (unlike in
TG), there is also a special subject position that is superordinate to the phrase
consisting of the verb and its other arguments (i.e. a position analogous to the
clausal speciWer position in TG).18 Other syntactic functions are not, however,
198 Transformational Grammar

distinguished by speciWc positions in a hierarchically complex tree structure.

Rather, it is assumed that the non-subject arguments of a verb are more
oblique than the subject and are ranked according to degree of obliqueness
vis-à-vis each other, which correlates simply with linear order (at least in
English and many other languages).
 Notes
1 Introduction
1 An analysis of roles according to which the subject of steal-type verbs is a recipient
and the subject of give-type verbs is a source is presented in JackendoV (1990)
where these NP are considered to also have the roles agent and subject.
2 This is a tack that is sometimes taken. Most textbook characterizations of the agent
role, such as Radford’s (1997: 326) deWnition of agent as the ‘instigator of some
action’ are based directly or indirectly on such inXuential deWnitions as Fillmore’s
(1968) ‘typically animate perceived instigator of the action’. Other theorists have
deWned the concept in such a way as to encompass even participants about which
it is diYcult to say that they do anything—perhaps something as schematic and
potentially all-encompassing as ‘Wrst cause’ or ‘autonomous cause’ (Delancey 1991,
Croft 1991) or even Pānini’s ‘independent participant’ (Kiparsky 2002)—which do
not entail the common _ conception of an agent as either typically animate or
volitionally instigative.
3 More precisely, one might say that case relations in this theory are considered to be
grammatical relations that can be roughly characterized semantically. In fact, the
distinction between syntactic functions and semantic roles is simply not system-
atically drawn in Lexicase Grammar. Case relations are fundamentally conceived
of as a set of basic roles around which the grammars of languages revolve.
4 In an active transitive sentence, the actor macrorole is borne by the (nominative)
agent. In a passive sentence the actor macrorole is borne by the (nominative)
patient.
5 It is perhaps worth noting that Old English like had a dative experiencer and a
nominative stimulus, apparently with the possibility of alternating frames like
those of Icelandic henta (Allen 1986, 1995). Over time the dative subject frame
prevailed and the case marking pattern evolved into the nominative–accusative
one of modern English (She likes him). See Section 4.4.2 for further discussion.
6 A key diVerence with the Role and Reference Grammar actor notion is that it is
assigned in a uniform way across paraphrases of a proposition, such that the
patient subject of a passive sentence, for example, is not the actor, and the actor
role would also not vary across the two paraphrases of an Icelandic henta sentence
(see Section 4.3.3).
7 This kind of statement, here and elsewhere, is of course subject to the qualiWcation
that what is meant by subject may possibly vary across languages and theories.
8 An even rarer type of language allows agreement with subject, direct object,
indirect object, and beneWciary (e.g. Choctaw). An example from Kinyarwanda
of four verbal indexes (for subject and three objects including beneWciary), which
200 Notes

could be considered to be a kind of agreement marking (example (2.78)) is

discussed in Section 2.2.4. The idea that indirect object necessarily occupies a
slot on a hierarchy of agreement possibilities is challenged by the Hokan language
Seri, which distinguishes for purposes of agreement subjects, direct objects, and
obliques in general (see example (2.43a) in Section 2.2.1).
9 For the sake of simplicity, I describe the situation as one in which pronominal
clitics are proclitics (i.e. verb-initial). Actually, they can sometimes be enclitics,
mainly with inWnitival verb forms. In this case, the orthographic system represents
them with hyphens (e.g. amá-lo ‘love.Inf-3SgAcc’). One of the ways Brazilian and
European Portuguese diVer is with respect to the proclitic vs. enclitic distinction,
the latter using the enclitic option much more. Being phonologically dependent
on a verb, the dative and accusative pronominal forms are generally considered
to be clitics, although they might actually be better analysed as pronominal aYxes,
as has been argued to be the case for the cognate elements in French (Miller and
Sag 1997).
10 Although lhe and the accusative pronominal clitics are glossed ‘3Sg’, they (especially
lhe) can be given a 2nd person interpretation, by virtue of the fact that the category
‘2nd person’ is routinely realized with pronominal forms that are grammatically 3rd
person. For example, the 2nd person informal full pronoun você triggers 3rd person
agreement on the verb when it is used as a subject. Although widely prescribed
against, lhe is often used for direct objects, but only with the 2nd person interpret-
ation, such that in addition to Eu te amo ‘I love you’ (with 2nd person dative/
accusative clitic te) and Eu amo você ‘I love you’, both of the following are possible:
Eu lhe amo ‘I love you’ and Eu o amo ‘I love you/him’.
11 This kind of analysis is sketched in Napoli (1993: ch. 3), where the label ‘Wrst
object’ is used informally for O1s, although the possibility that these are really also
indirect objects is left open.
12 A deWnition in terms of ‘all P-marked NPs that have the same kind of semantic
role as O1s’ makes it possible to include, for example, the addressee of the of-
marked NP in I asked a question of the teacher, which can alternatively be an O1 (I
asked the teacher a question) or the beneWciary for-marked NP in I baked a cake for
my mother (cf. I baked my mother a cake). For a sober discussion and unresolved
analysis of this and other diYculties that arise in an attempt to identify an
indirect object category in English, see Kaplan (1995: ch. 7).
13 See, for example, the strict adjacency principle of Government-Binding Theory
(Chomsky 1981, Stowell 1981), according to which a verb can only assign its
structural accusative case to an NP that is right-adjacent to it.
14 Some speakers (apparently mainly of British English) allow an O2 to be the
subject of a passive clause (e.g. Czepluch 1982, Postal 1986: 58, Siewierska 1991:
97, Ura 2000: 244). Even for American speakers, it is also at least easier to get the
O2 to be the subject of a passive clause if the O1 is a pronoun (Oehrle 1976: 177).
15 In some cases, intransitive verbs with an oblique complement allow their only
non-subject dependent to be a subject in the passive construction (e.g. This bed
 Notes 201

has not been slept in). There are various analyses of this latter phenomenon,
sometimes called pseudopassive, including one according to which slept in, for
example, is a compound verb whose direct object dependent is expressed as its
subject (for a detailed overview see Postal 1986: ch. 6). By expressing the gener-
alization concerning object preference in the passive construction with the qua-
liWcation ‘given a choice between two non-subject dependents’, the pseudo-
passive case is taken out of the equation, whatever analysis is adopted.
16 Tough movement, for example, is not restricted to objects (People like these are
hard to talk to).
17 Role and Reference Grammar (see Section 4.1.3) and Functional Grammar (see
Section 4.4.1) also do not recognize the indirect object relation.
18 For example, at least in the version of the theory articulated in Falk (2001), the
analysis of semantic roles is in terms of a hierarchy with agent at the top and a slot
occupied by beneWciary or patient, followed by instrument, theme, and locational
roles (see Section 3.4.1). BeneWciary, however is in fact used as the term for a
generalized role that I am calling ‘recipient’ and ‘theme’ is used (with verbs such
as give) for what I am calling ‘patient.’
19 This is also the approach taken in Radford’s (1997: 519) introduction to syntactic
theory and English grammar. A blend of this kind of analysis and the standard
Relational Grammar analysis is also proposed in Farrell (1994a). The NPs in to-
phrases are analysed as obliques at all levels. All and only O1s are claimed to be
initial indirect objects. As in such languages as Tzotzil (Aissen 1983) and Halk-
omelem (Gerdts 1988a) (see Section 2.2.4), however, indirect objects are assumed
to be obligatorily promoted to direct object, such that the main idea of the
Relational Grammar analysis is maintained. That is to say, O1s behave in key
ways like prototypical direct objects because they are Wnal direct objects.
20 The verb-adjacency constraint might in fact be attributed to a universal principle,
given that the more speciWc roles encompassed by ‘recipient’ are inherently roles
of animate participants and there appears to be a general preference in languages
for NPs designating more animate NPs to precede NPs designating less animate
NPs, as expressed in the animated-Wrst principle of Tomlin (1986).
21 The term oblique has also been used to designate cases other than nominative, in
languages with morphological case-marking systems of the kind found in Latin,
Greek, and Sanskrit (see Blake 1994: 31).
22 There are other languages that collapse lots of semantic roles with prepositions or
case markers but do make some distinctions. Austronesian languages, in particu-
lar, seem to operate with few dependent-marking categories. In Chamorro, for
example, A/S/O are unmarked and other dependents of verbs are either marked
with a generalized oblique preposition or a semantically speciWc preposition for
locative dependents (Gibson 1980).
23 As Fillmore pointed out, one can analyse the subject of This key opened the door
as having the semantic role of instrument. However, as has often been noted,
the possibility of being a subject doesn’t extend to most constituents with an
202 Notes

instrument role (e.g. *This funnel Wlled the glasses with wine, *This bullhorn
addressed the audience). Thus, there is no reason to assume that so-called
instrumental subjects are not simply conceived of as agents, by routine meta-
phorical extension from the protoype (Delancey 1991). Constituents that behave
like adjuncts in one paraphrase can routinely be the subject or the object in
another, both in English (e.g. Labor Day saw the Democrats on the brink of losing
power vs. One saw the Democrats on the brink of Losing power on Labor Day—see
Perlmutter and Postal 1984) and in languages such as Kinyarwanda (Kimenyi
1980) with robust ‘promotion to object’ strategies that can aVect adjuncts of
various kinds. In Kinyarwanda, dependents interpreted as having the instrument
role can clearly have the direct object syntactic function and are presumably,
therefore, arguments rather than adjuncts (see the Kinyarwanda examples (2.78)
in Section 2.2.4).
24 Although Kroeger (1993), for example, analyses the so-called nominative case as
indicating the subject, which implies that sentences such as (1.39b–d), for example,
are essentially passive clauses, this analysis is not uncontroversial (see Section 2.2.5).
25 Whaley claims that passive by phrases behave like adjuncts with respect to VP
pseudoclefting based on examples such as What was done by the man was to give
Mary a book. Not only is this sentence a rather odd paraphrase of What the man
did was give Mary a book, it also is not a pseudocleft version of a passive clause, as
the object is not promoted to subject, the verb is not in a past-participial form,
and there is no be auxiliary.
26 See also Baker (1997) for a similar conception of how semantic roles can be
reduced to three basic types.

2 Grammatical relations across languages

1 Although relatively rare, there are languages that utilize neither head marking nor
dependent marking to indicate grammatical relations. It has been argued for one
of the better known cases of this type of language, that is Chinese, that the
grammar is organized around topic-comment structure and grammatical rela-
tions such as subject and direct object are not distinguished at all (La Polla 1993).
Tripartite systems, in which A, S, and O are all morphologically distinguished, are
also rare. There are languages that distinguish all three for certain grammatical
phenomena, such as relative clause formation in Kamaiurá (discussed in Section
2.1.3). The complex case-marking systems of Australian languages often display
tripartite properties with respect to classes of NP types, although only a few, such
as Wangkumara and Galali, have been reported to consistently use a tripartite
marking system across all NP constituents (Dixon 1994: 41).
2 A tacit assumption here is that a se clause with O pivot is transitive and therefore
has an A, in spite of the absence of an overt O. In some Romance languages, such
as Italian (e.g. Rosen 1988, La Fauci 1988), there are good reasons to claim that se
clauses are intransitive. The same kind of evidence doesn’t exist for Brazilian
 Notes 203

Portuguese. In any case, there is a type of se clause that has an overt direct object
and an indirect object pivot (see (1.15a)). Thus, the point that the controller can
be either an S or an A is valid independently of the transitivity of sentences such
as (2.2a).
3 It does not appear to be possible to reduce the postverbal placement of the S of
intransitives to a neat class of verbs, such as telic verbs, for example, or inanimate Ss,
although animacy and telicity are among the inXuencing factors. See Dutra (1987)
for a detailed study of the phenomenon and Perlmutter (1976) for relevant
discussion.
4 However, ‘nominative’ is sometimes used instead of ‘absolutive’, as in Bittner and
Hale (1996a), in such a way as to highlight the similarity in terms of syntactic
behaviour between S/O in certain ergative languages and S/A in accusative
languages. There are also languages in which the absolutive marker in the ergative
domain of the grammar counts as the nominative marker in the accusative
domain.
5 Consistent with the choices made for other languages, here and throughout I
substitute A and O in glosses for Gerdts’ Erg and Obj, respectively. I use her Subj,
with the understanding that this means S/A, as it does, in eVect, for her.
6 Nouns in this language almost always end in an -a suYx that Seki analyses as a case
marker, which she labels ‘nuclear’. Since this is used for all NPs in the S, O, and
A categories as well as objects of postpositions, possessors within NPs, and nom-
inalized verbs in various functions, it does very little, if anything, to discriminate
grammatical relations. In all the examples discussed here and elsewhere, this suYx
is not indicated as such or glossed.
7 This is a bit of a simpliWcation. More precisely, when the A is 1st person and the O
is 2nd person, a number hierarchy also comes into play. In this situation, only if
the A is also higher on the number hierarchy (plural > singular) does it take
precedence with respect to the verbal morphology. If the two dependents are of
equivalent rank with respect to number, both the A and the O are indexed with a
portmanteau preWx (Seki 2000: 160), as in example (2.21b).
8 There are, however, languages in which the apparent semantic basis for unac-
cusativity is obscured by unexpected classiWcations of various kinds, as noted in
Dixon (1994: 74).
9 Here and elsewhere, I use A and O in the glosses for Choctaw agreement
morphology, whereas Davies uses Nom and Acc respectively. Although partici-
pant control appears to be the main factor determining the distinction between
the Sa and So categories, as is often the case in split-intransitive systems (e.g.
Dixon 1994: ch. 4), there are some verbs that seem not to follow the general
pattern. In Choctaw, for example, ‘die’, numeral predicates, and verbs of exist-
ence take A marking, even though it seems diYcult to conceive of them as
ascribing control to a participant. Dixon considers Choctaw to fall into a special
category of languages, because it also has semantically-sensitive dative agreement
morphology that increases the complexity of the grammar. In essence, it is
204 Notes

fundamentally a split-intransitive language with a dative quasi-subject and dative

quasi-object overlay that is amenable to varying interpretations.
10 Note that a pormanteau preWx such as oro- in this example indexes both the A
and O, just in case neither is 3rd person and they are equivalent in terms of
number (see note 7 in Chapter 2).
11 What diVers is only that I follow Durie (1985) in indicating clitics with the ‘¼’
notation, and don’t explicitly indicate pronoun optionality in some of the examples.
12 The O can be unincorporated only if two of three conditions are met: (i) it is
human-referring, (ii) it is animate singular, or (iii) it has a demonstrative or
numeral modiWer (Allen et al. 1984: 293–97, Rosen 1990: 683).
13 Glosses are changed from the original to conform to the general system employed
in this book. Moreover, certain derivational aYxes of no signiWcance in the
present context are not glossed.
14 See also Bittner (1987) on some of the precise semantic correlates of the anti-
passive vs. active distinction.
15 Certain glosses are changed from the source to conform to the general system
employed in this book or to avoid implicitly endorsing a particular analysis.
Moreover, certain irrelevant details are not explicitly indicated. Tense mor-
phemes, for example, can indicate several tense categories (recent past, immediate
past, etc.) as well as evidentiality information and gender agreement. I only
indicate the gender category that these encode, since that is what plays a relevant
role in the discussion here.
16 There are a number of agreement and pronoun restrictions that make the
grammar of Jarawara rather complicated. For example, the form of the pronom-
inal element on or preceding a mood suYx in an inverse clause (i.e. the second o-
in (2.63b)) is actually determined on the basis of person (1st/2nd > 3rd) rather
than syntactic function.
17 This deWnition abstracts away from the eVects of other derivational operations
such as passivization or reciprocalization, which might detransitivize an applica-
tive clause. It also implicitly excludes what would have to be analysed as an anti-
applicative construction in Inuit, as described in Bittner and Hale (1996a). Either
the recipient or the patient can be the O (with characteristic absolutive marking,
i.e. zero case). However, the verb bears the so-called applicative aYx when the
patient is the O and the recipient is in the dative case form, rather than when the
recipient is O and the patient is in the instrumental case form. One could include
this construction in the applicative class by deWning the applicative construction
as one with a ‘non-canonical’ mapping of O with semantic roles. It would also be
necessary, of course, to assume that the recipient is the canonical O in Inuit, that
is that Inuit is a kind of primary-object language in the sense of Dryer 1986.
18 Gerdts glosses the applicative morphemes with AdvA, AdvB, etc., for type A/B
‘advancements’. I use, instead, theory-neutral descriptive names for what I char-
acterize as a phenomenon that closely parallels the so-called voice phenomenon
in Philippine languages. What I am calling the recipient applicative morpheme—
 Notes 205

following the idea, from Section 1.3.2, that recipient is the prototype for a family
of more speciWc roles—is used at least for addressees as well as transfer-of-
possession recipients.
19 It is not the case that oblique dependents can’t be relativized or focused on in
clefting or wh-question constructions. However, a diVerent construction with a
nominalized verb must be used.
20 Of course, as with passive and inverse voices, lines could be drawn in such a way
as to exclude one or more construction types from an ‘applicative’ classiWcation.
For example, one might take overt morphological marking of the construction to
be criterial, in which case the ‘double-object’ constructions of English and
Korean, for example, would be excluded. As there are clear commonalities as
well as diVerences across languages and constructions, various kinds of line
drawing could be done for labelling purposes.
21 Although the applicative analysis of Philippine voices has been an integral part of
the ergative approach, it could conceivably be incorporated into other ap-
proaches as well. For example, under the accusative approach, one might analyse
locative, benefactive, and instrumental voices as involving promotion of obliques
to O (¼ applicative) with obligatory promotion of applicative Os to S (¼
passive).
22 This terminological distinction roughly parallels that of oblique/dative subject vs.
I-nominal adopted in Moore and Perlmutter (2000) and Sigurðsson (2002).
Quasi-subject is more general than I-nominal, which is meant to cover only the
dative dependents in what has been called an inversion construction in Relational
Grammar. The dative dependent in an inversion construction is analysed as an
initial subject and Wnal indirect object. Its restricted subject-like properties are
attributed to its initial subjecthood.
23 At least for many speakers the a-marked direct object does not appear to be truly
dative-marked, insofar as the dative clitic that typically (often obligatorily)
appears with the verb in a sentence with an indirect object, as in the case of
(2.105a) does not occur with the a-marked O. For some speakers an accusative
clitic could be used instead, especially if the O is pronominal (Marta la vio a ella
‘Marta 3SgFemAcc-saw her’). There are, however, dialects/idiolects in which the
dative clitic is routinely used for both Os and IOs.
24 As in other Romance languages, the se morpheme can be used to indicate an
inchoative or accomplishment interpretation of a verb. Thus se cansar can mean
‘to get tired’.
25 There are certain verbs, such as lembrar ‘remember’ and esquecer ‘forget’ whose
stimulus dependent can either be de-marked or not. The stimulus is a straight-
forward O in one of the alternative constructions. The status of the de-marked
NP is harder to establish because in most potential O-behaving situations, the
construction would be indistinguishable from the alternant with the O. The de-
marked NP with these verbs does, however, behave like a quasi-object in that it
can be the pivot in the otherwise A/S/O-restricted o que pseudocleft construction
206 Notes

(O que lembro é da alegria da minha mãe ‘What I remember is of-the my mother’s

joy’).
26 In conversational discourse 3rd person accusative clitic forms are generally
avoided in favour of what could be analysed as a null pronoun (e.g. Farrell
1990, Kato 1993). This null pronun interpretation is possible for Minha mãe
gosta ‘My mother likes Ø (¼ it/him/her).’ However, it is not clear that this is
an O-deWning property of dependents, since the omission of obliques can
sometimes yield a similar interpretation (e.g. A gaveta não está cheia. Então vou
colocar mais algumas coisas ‘The drawer isn’t full. So, I’m going to put a few more
things Ø (¼ in it).’). Athough there is a clear diVerence between the possible
(albeit formal) use of a 3rd person pronominal clitic with a verb such as amar
‘love’ and the ungrammaticality of such a clitic with gostar ‘like’, the situation is
more complex with 1st and 2nd person pronominal clitics, which are routinely
used in all modes of discourse and for which there is no phonological dative vs.
accusative contrast. Although some speakers consider it odd to use 1st and 2nd
person clitic pronouns with gostar, it is not hard to Wnd examples in actual
discourse, such as Me faz um bem . . . o jeito que ele me gosta ‘It does me good . . .
the way he 1Sg-likes’ (from a Brazilian internet blog) or Eu te gosto, você me gosta
desde tempos imemoriais ‘I 2Sg-like, you 1Sg-like since time immemorial’ (from a
poem by Carlos Drummond de Andrade).

3 Relational Grammar
1 The RG characterization of the deWning property of passive clauses diVers from
the characterization given in (2.35). In RG, it is the promotion of a 2 to 1 from a
transtitive stratum that is criterial. Nevertheless, essentially the same range of
construction types are identiWed by the RG deWnition and (2.35), since any passive
clause with the agent demoted from 1 would be analysed as having an advance-
ment to 1, even if it is a null ‘dummy’ that advances. Thus, impersonal passive
constructions of the form Was danced by the children (which do not occur in
English but do in other languages) are analysed as having a null dummy subject
that advances from 2 to 1. Whether passive advancements from 3 or oblique
(rather than 2) to 1 need to be recognized is an unsettled issue (see Postal 1986).
2 There are various potential ways of looking at the so-called ‘pro-drop’ situation
that one Wnds in Kamaiurá and many other languages. One could say, from an RG
perspective, that the arguments of the verb are manifested as independent
pronouns that bear the grammatical relations represented in the relational
network and determine the morphology on the verb. By rule these pronouns
are ‘dropped’ or deleted. One could also say that the inXectional morphemes
themselves bear the grammatical relations or that abstract (phonologically null)
entities with person-number features bear the grammatical relations and deter-
mine the forms of the morphology on the verb. Whichever analysis one adopts,
the relational networks themselves would not diVer substantially. For the sake of
 Notes 207

simplicity, I just put the grammatical relations in the tabular stratal diagrams of
the relational networks above the associated inXectional morphemes.
3 It is not actually necessary to make any assumptions that have not been appealed
to elsewhere with good reasons. The idea that recipients, for example, can be
initial obliques in languges is adopted in Rosen’s (1990) analysis of Southern
Tiwa. As for instruments, one could assume that they are necessarily initial
obliques in all languages, following the usual RG assumption; but that they can
and sometimes must advance to 3 in Tagalog.
4 This analysis, which allows the promotion of a dependent to 1 to induce demotion
to 3 for the ‘overrun’ 1, also entails a version of RG without the Chômeur Law (also
known as the Overrun Law), which requires demotion to chômeur in such cases.
Independent reasons for abandoning this law are discussed in Perlmutter and Postal
(1983). The RG 1–2 reversal analysis for inverse voice languages (Rhodes 1976,
Perlmutter and Rhodes 1988, Section 3.3.3 below) would be banned by the Chômeur
Law. The simpler of the two analyses of Icelandic psych verbs given in Section 3.3.5
also assumes that there is no Chômeur Law. The virtue of having the 1 demote to 3 in
Tagalog is (i) that it allows a uniWed account of the case marking for Wnal 2s,
instrumental NPs, and demoted 1s (i.e. they are all Wnal objects), and (ii) that it
accounts for the term-like behaviour of demoted 1s with respect to adjunct-front-
ing. An alternative would be to have an overrun 1 always demote to 2, inducing
demotion of the initial 2 to 3 in certain cases. It is diYcult to see how this more
complex alternative might be empirically distinguished from the 1–3 demotion
analysis.
5 Also known as the ‘ergative hypothesis’, due to the terminological innovation
introduced in Burzio’s (1986) translation of the RG analysis of Italian unaccusa-
tivity phenomena into the Government-Binding framework of Transformational
Grammar. The terms unaccusative and unergative are rather unfortunate to begin
with, inasmuch as the subject of what would generally be considered unaccusative
verbs is often treated like an accusative dependent. Calling this class of verbs
‘ergative’, while appropriately naming the class of verbs that contrasts with
unergative verbs, is at least as confusing, inasmuch as the phenomenon of split-
intransitivity is unrelated to the phenomenon of ergativity, in the sense of most
descriptive and typological work.
6 Whether ‘controller of action’ is the right semantic notion here is immaterial.
More nuanced accounts could presumably be given by appealing to the Role and
Reference Grammar distinction ‘actor’ vs. ‘undergorer’ (see Section 4.1.3) or
some kind of account in terms of the agent vs. patient semantic entailments
(Dowty 1991, Ackerman and Moore 2001).
7 For the sake of brevity, I focus attention here on two of the potential diagnostics of
unaccusativity in English. Others include possible construal with a resultative
adjective (the lake froze solid ) (see especially Levin and Rappaport Hovav 1995: ch.
2); the possibility of locative inversion (Coopmans 1989, Levin and Rappaport
Hovav 1995: ch. 5), which is constrained much like expletive there insertion;
208 Notes

inability to be suYxed with ‘agentive’ -er (freezer ¼ agent of transitive freeze; not
patient of intransitive freeze) (Rappaport Hovav and Levin 1992, Farrell 1994a);
and inabilitity to undergo ‘pseudopassivization’ (*This tree was fallen from vs. This
bed was slept in) (Perlmutter and Postal 1984). A purely semantic constraint in
terms of the patient semantic role is proposed for resultatives in Goldberg (1991).
That the constraint on agentive -er suYxation is semantically constrained in a
diVerent way than the other phenomena is made clear in Ryder (1999) and Farrell
(2001), one kind of evidence being that the class of intransitive verbs that do
undergo -er aYxation (supposedly unergatives) overlaps partially with the class of
verbs that form participial adjectives (supposedly unaccusatives) (e.g. the escaped
prisoners vs. the escapers). Although pseudopassivization does, by and large, fail
with the class of verbs that can form participial adjectives (a fallen leaf vs. *This
tree was fallen from by a leaf ), it appears that a semantic constraint on pseudo-
passivization is needed independently of the verb classes deWned by participial
adjective formation. The verb jog, for example, would be unergative rather than
unaccusative by the participial adjective criterion (*the jogged person); yet, this
verb, like putative unaccusatives, does not allow pseudopassivization (*Central
Park was jogged in this morning). Again, there is a mismatch between the verb
classes deWned by these two phenomena and there is a semantic factor at play: the
pseudopassive subject must have as a referent something that is construed as
being clearly aVected (hence, This bed appears to have been slept in vs. *This hotel
appears to have been slept in).
8 While acknowledging unaccusative mismatches and the relevance of lexical
semantics in ‘unaccusative’ phenomena in English, Levin and Rappaport Hovav
(1995) argue for the view that split intransitivity is both semantically motivated
and syntactically manifested in the way that the unaccusative hypothesis suggests.
This stance is also taken, for example, for Italian in Farrell (1994b) and for French
in Legendre (1989b). That this stance can be taken is not being questioned here.
Rather, it is being suggested that building a solid case for the necessity and
explanatory utility of such a stance remains a clear challenge for RG or any
other theory that adopts the unaccusative hypothesis. With this said, it must be
acknowledged that Italian, taken on its own, presents a compelling case for the
unaccusative hypothesis, inasmuch as the unaccusative diagnostics generally pick
out a uniWed class of dependents in which (underlying) direct objects of transitive
clauses are generally included (Rosen 1988, Perlmutter 1989).
9 It would be fruitless to try to engage the posited initial 1-hood of the experiencer
in accounting for the ‘subject’ properties of the dative experiencer in the imper-
sonal construction. One might say, for example, that the notion of ‘subject’ that
most phenomena in Icelandic are sensitive to is ‘working non-expletive 1’, that is:
a dependent that is a 1 in some stratum, a Wnal term, and not an expletive. This
would account for the subject properties of the experiencer in the impersonal
construction (3.16b). The problem is that the personal construction, illustrated by
 Notes 209

(3.16a), would have two working non-expletive 1s, that is both the stimulus Wnal 1
and the experiencer, which is an initial 1 and Wnal 3.
10 The null expletive pronoun is required, due to the Motivated Chômage Law
(3.2d)—to get the stimulus 1 to retire to chômeur—and the Final 1 Law (3.2c),
which requires that a clause have a Wnal 1.
11 For dialects that allow the O2 to be a passive-clause subject (see note 14 in Chapter
1), either object can presumably be intrinsically [
r], as in symmetrical applica-
tive languages (Bresnan and Moshi 1990).
12 An arc with no Wnal stratum coordinate technically gets ‘erased’, for example,
either by itself or, more commonly, by some other arc. An arc whose presence is
entailed by some other arc A in a network is said to be sponsored by A. Thus, for
example, in the case of, say, a 2–3 demotion construction, a 2-arc in stratum s
sponors and is erased by a 3-arc in stratum sþ1.

4 Role and Reference Grammar

1 Because the Natural Semantic Metalanguage of Wierzbicka dispenses with the
predicate-calculus formalism and allows meanings of verbs to be stated in a form
of ordinary English that is built out of semantic primitives (e.g. put ¼ x does
something to y ; because of this y goes to someplace (z)), verb meanings elsewhere
in this book are expressed using a simpliWed version of this metalanguage (see, for
example, Figure 1.2).
2 Aktionsart (or ‘aspectual’) distinctions among verb classes are motivated empir-
ically, in English and other languages, on distributional grounds (e.g. Vendler
1967, Dowty 1979, Foley and Van Valin 1984, Filip 1997). For example, stative verbs
in English diVer from all others in that they do not generally occur in the present
progressive construction (She is singing a song [activity] vs. *She is liking the pizza
[state]) and can denote a present time state of aVairs, rather than a habitual state
of aVairs, in their present tense form (She likes the pizza). Activity verbs do not
occur with prepositional adjuncts that designate temporal endpoints (She jogged
for an hour/*in an hour), unlike accomplishment verbs (She learned the song in an
hour/*for an hour). The key point is that there is an evidentiary basis for the LSs
posited for verbs in the RRG framework.
3 Since the variable x occurs in two places in the LS for run, the subject of run has
two semantic roles: mover and theme. Although theme is the default choice for
undergoer, the fact that such verbs have only one macrorole puts the principle in
(4.4) in play. Since there is an activity predicate in the LS of run the single
macrorole is necessarily actor. The least-marked actor assignment is to the mover.
4 Thus, the default for Brazilian Portuguese is that the most patient-like argument
on the actor–undergoer hierarchy is the undergoer, whereas in Halkomelem, the
second-most patient-like argument is the default undergoer.
5 RRG has formal mechanisms to distinguish robust head-marking languages, in
which the verbal aYxes themselves are characterized as the primary manifestation
210 Notes

of LS arguments, from languages with agreement, in which NPs, which are the
primary manifestation of LS arguments, may simply ‘control’ agreement mark-
ing. For the sake of simplicity, no attempt is made here to systematically draw this
distinction.
6 Since dative case-marked NP arguments are considered ‘direct’, the correspond-
ence between ‘direct non-macrorole core argument’ and ‘2nd object’ in the sense
of Dryer (1985) or Bresnan (1982a) is also imperfect. A further distinction that can
be drawn among PPs is argument-adjunct vs. argument (Van Valin and La Polla
1997: 160–3). The idea is that the goal argument of a verb such as put, for example
in the box in Sam put the rabbit in the box, is an argument-adjunct since the LS of
the verb is underspeciWed with respect to the speciWcs of the locative relation
involved: on the box, in the box, and under the box, for example, are all possible.
Thus, the box in such cases is an argument of a predicative P, in a way that the boy
isn’t in give the book to the boy.
7 In languages such as Icelandic, discussed in Section 4.3.3, the traditional indirect
object category is overtly indicated by dative case marking on NPs. The category
can be deWned as ‘direct core non-macrorole argument’. In Brazilian Portuguese,
on the other hand, the traditional indirect object category consists of a subset of
NPs that occur in a PP with the preposition para ‘for’ or a ‘to’ as head, that is
those which can be realized as a dative pronominal clitic and can be a reXexive/
reciprocal se pivot (see Section 1.3.2). This class of NPs can be deWned neither as
direct core non-macrorole arguments (because they are not ‘direct’) nor as
oblique non-macrorole arguments, because they constitute only a subset thereof.
Although the semantic role ‘recipient’ may be the prototype, it is insuYcient as a
deWning criterion, as non-recipient possessors can be dative clitics (Aquilo não
lhe-pertence ‘That doesn’t 3SgDat-belong’) as can perceivers (Aquilo lhe parece
doido ‘That 3SgDat-seems crazy’). The notion animate oblique argument comes
close, but fails to exclude, for example, animate source oblique arguments (Eu
recibi uma carta de sua mãe ‘I reveiced a letter from your mother’ vs. *Eu lhe recibi
uma carta ‘I 3Sdat-received a letter’). Of course, the problem that RRG faces in
specifying the category of arguments that can be dative clitics in Brazilian
Portuguese is not solved in an interesting way in theories that simply claim that
‘indirect object’ is a primitive category.
8 In Van Valin and La Polla (1997) the term grammatical relation is used to signify
essentially only ‘syntactic function’ in the sense used in this book, whence the
claim that Acehnese has no grammatical relations. Another way of looking at the
matter, of course, is that the boundary between syntactic functions and semantic
roles is unclear and perhaps largely an analytical artefact. Clearly, actor and
undergoer are at most quasi-semantic roles, given that such matters as whether
a verb or verb class has one or two macroroles and whether the location-type or
theme-type semantic role is chosen as undergoer with 3-argument verbs are not
semantically determined. The fact that the actor (‘semantic role’) almost invari-
ably corresponds to the initial subject (‘syntactic function’) of Relational Gram-
 Notes 211

mar shows how much the terminological distinction between semantic role and
syntactic function may be drawn in diVerent ways for the same state of aVairs.
9 The glosses for the inXectional morphology are changed somewhat from the
original to conform to the general schema used here and to be consistent with the
glossing schema used with other Inuit examples discussed elsewhere.
10 Note that argument-modulation passive clauses, unlike argument-modulation
antipassive clauses, are characterized as M-transtitive. This brings out a diVerence
between the RRG notion of M-transitivity and the common notion of syntactic
transitivity which is appealed to in the description of passive voice in Section
2.2.1, where it is claimed that passive is, in eVect, a detransitivization phenom-
enon (as in Dixon 1994, for example). In the RRG analysis of the English passive
construction, a non-actor argument (typically the undergoer) is the PSA. The
actor macrorole is not taken away, however, from the highest-ranking argument
on the actor-preference hierarchy; it is simply associated with an unspeciWed
argument position in the LS or with an adjunct PP. Thus, the passive construc-
tion is M-transitive (i.e. it has both actor and undergoer macroroles), although it
is generally syntactically intransitive inasmuch as it has only one direct core
argument. In antipassive constructions with argument modulation as in Dyirbal
or Inuit, on the other hand, the undergoer macrorole is suppressed, yielding an
M-intransitive clause. However, an antipassive clause may apparently be syntac-
tically transitive, since it may have two direct core arguments, as in Dyirbal.
11 This kind of a PSA-modulation-only analysis is presented for the Nilo-Saharan
language Lango (Noonan 1992) in Van Valin and La Polla (1997: ch. 7) and Van
Valin (to appear: ch. 4). For inverse-voice languages of the Algonquian kind (see
Section 2.2.3), RRG makes possible the traditional morphological inversion an-
alysis, according to which the interpretation of the agreement morphology of the
verb diVers between active and inverse voices and there is no PSA modulation
(Van Valin and La Polla 1997: ch. 7).
12 Antipassive has been claimed to occur, for example, in French (Postal 1977,
Herslund 1997), although whether the constructions at issue would be analysed
as antipassive in RRG is unclear.
13 In the Relational Grammar analysis of Marlett (1984) and Farrell et al. (1991), the
constraint is stated in terms of the notion ‘Wrst subject’. In a passive clause, the
initial subject is the agent/actor and the Wnal subject is the patient/undergoer. The
initial subject is the Wrst subject (i.e. subject in the earliest stratum). In all other
clauses in Seri, the Wnal subject would be the Wrst and only subject.
14 According to Barðdal, henta and most of the verbs in the same class are stative
and the alternative syntactic frames are semantically equivalent in essentially the
same way that active and passive paraphrases are. For this reason, it is assumed
that the LS of henta contains only the state predicate like’ and is the same whether
the dative or the nominative NP is the PSA. The be pleasing to and be pleased with
English glosses are used simply to give idiomatic English translations with the
212 Notes

PSA varying in the way that it does in Icelandic. Icelandic also has psych verbs
that are more like English please ; they do not alternate and have a nominative
actor PSA and an accusative undergoer.
15 Another principle—not active, or barely active, in Icelandic and English—yields
ergative phenomena, that is the lowest-ranking argument with respect to the
actor end of the actor–undergoer hierarchy is the preferred PSA.
16 The category ‘DCA’ includes the actor in an M-transitive clause as well as the S of
an M-intransitive clause. In fact, as a general rule, all members of this category
can control a Xoated quantiWer or be a relativization pivot.
17 An alternative analysis might be that cansar is a one-argument verb and the de-
headed PP that can occur with it is an adjunct, similar to por causa do sol ‘because
of the sun’ in Ele cansou por causa do sol ‘he got tired because of the sun’. Although
such an analysis is semantically reasonable, the de-headed PP with cansar does
not display the syntactic properties of a typical adjunct, as illustrated by the
following examples:
(i) a. O que aconteceu nesse lugar foi que eu conheci ela.
‘What happened in this place was that I met her.’
b. O que aconteceu por causa do sol foi que ele cansou.
‘What happened because of the sun was that he got tired.’
c. O que aconteceu foi que todo mundo gostou do Wlme.
‘What happened was that everyone liked the movie.’
d. *O que aconteceu desse Wlme foi que todo mundo gostou.
‘What happened of this movie was that everyone liked (it).’
e. *O que aconteceu nesse lugar foi que eu pensei.
‘What happened about this place was that I thought.’
f. *O que aconteceu dessa música foi que eu cansei.
‘What happened of this song was that I got tired.’
As in English, it is possible in Brazilian Portuguese to focus on a verb phrase
constituent in a pseudocleft construction only if all complements of the verb are
included with it, as in the case of the italicized focused phrases in (ia–c). It is
possible to leave out an adjunct, as shown by (ib–c). It is not, however, possible to
leave out a complement, as shown by (id–e). The ungrammaticality of (if)
suggests that the stimulus PP of psychological cansar is a complement. Most
importantly, for present purposes, however, even if the de-marked stimulus PP of
cansar were analysed as an adjunct, the problem of the quasi-object status of the
stimulus of gostar would not be solved. One would presumably want to say that
the stimulus of gostar is more object-like in its behaviour than the stimulus of
cansar because it is a complement and it is nevertheless oblique-like due to its
non-macrorole status. Its special properties would simply follow from its status as
a non-macrorole complement. But the problem is that there are other non-
macrorole complements of verbs of cognition and emotion, such as the PP
complement of pensar ‘think’ (Eu pensava em você ‘I was thinking about you,’
 Notes 213

literally ‘in you’), which do not have any of the object-like properties of the
stimulus of gostar (e.g. *Eles só se pensam ‘They only think about each other’,
*Essas pessoas são difı́ceis de não pensar ‘These people are hard not to think
(about)’).
18 This is the simplest version of the algorithm for assigning semantic functions to
the A1, A2, and A3 categories and is one that works for English. In order to
account for complications that arise in certain languages, the possibility of
allowing the A2 to have variable semantic functions even in clauses with three
arguments is entertained (Dik 1997: ch. 11).

5 Transformational Grammar
1 In addition to the syntactic transformations, Figure 5.1 shows in the surface
structure representation the eVects of the transformational rules of the phono-
logical component of the grammar, which technically operate on the surface
structure representations. These phonological rules, among other things, change
the abstract representations of may PST (i.e. may plus a past tense morpheme)
into might and put-en (i.e. put plus a past-participial suYx) into put. Certain
other aspects of the derivation, such as placement of word boundaries, are also
not made explicit.
2 In Figure 5.2, the PP and the direct object NP are shown as sisters of the verb, as
they are in classical TG and in at least early versions of GB, including Chomsky
(1986). By the end of the GB period, other representational systems for multiple
complements had been developed (notably the so-called VP shell of Larson 1988),
such that a binary-branching constraint on phrase structure, which would be-
come de rigueur, could be maintained. The Minimalist approach to multiple-
complement and double-object constructions, which respects the binary-branch-
ing constraint, is discussed in Section 5.3.2.
3 In the Aspects model, as well as in early versions of GB, the clause was given the
category label S (for sentence) and was assumed to have inXectional and gram-
matical elements under a node usually labelled Aux (for auxiliary), as shown in
Figure 5.1. Partly to allow a uniform template for all phrasal categories, the IP
analysis of the clause was introduced in Chomsky (1986). Following Pollock
(1989), the individual inXectional elements (tense, agreement, aspect, etc.) have
been commonly analysed as heading individual phrases of their own category,
which are embedded within each other in an elaborate way. These distinctions,
which are largely orthogonal to grammatical relations and which have been
implemented in numerous ways, are abstracted away from here.
4 The idea of LF is not new to the Minimalist programme, as it has its roots in GB
accounts of the semantic scope properties of quantiWers and other similar
elements (May 1985), and ultimately in the Generative Semantics tradition within
TG (e.g. LakoV 1971). What is new is its extended use to account for certain
214 Notes

grammatical-relation phenomena and other routine matters of word order

variation.
5 The guiding principle is known as the Uniform Thematic Alignment Hypothesis
(UTAH) (Baker 1988a), which is a TG analogue of the Uniform Alignment
Hypothesis of RG.
6 Realizing the vision of reducing all semantic roles to structural positions is
contingent, in part, on being able to pack much of the equivalent of lexical
conceptual structure in the sense of, say, JackendoV (1990) into the syntactic
representation of phrase structure, as was attempted early in the history of TG in
the movement that came to be known as Generative Semantics (e.g. LakoV 1971,
McCawley 1971, 1973). Some empirical diYculties with this vision as implemented
in Hale and Keyser (1993) are noted in Farrell (1998). See Baker (1997) and Gruber
(2001) for more comprehensive accounts of what might be required to fully
implement the vision. Baker, for example, suggests an analysis in the Generative
Semantics spirit wherein semantic primitives such as CAUSE and GO are placed
under v and V, for example, and the verb send replaces the amalgam CAUSE-GO
(created by incorporation of V into v) at some point in the derivation. As Baker
correctly notes, such an endeavour leads inexorably to a variant of a theory such
as Cognitive Grammar (see Section 4.4.2), in which no distinction is drawn
between grammatical and conceptual structure.
7 I follow Baker (1997) in the assumption that the patient (or ‘theme’) originates in
a speciWer position of the VP. Hale and Keyser (1993) adopt the opposite analysis,
with the patient in the lowest position. What I am calling NPs are generally
analysed as determiner phrases (DPs) in this theory, that is as phrases headed by
an article or other determiner (possibly null) that has an NP as its complement.
Since the NP/DP distinction is not relevant to the issues of concern here, it is
suppressed merely for the sake of consistency with terminology employed else-
where.
8 It is also often assumed that heads can have a feature specifying the lexical
category of the speciWer it must have. This feature is systematically ignored here.
9 In one version of the theory (e.g. Chomsky 1993, Watanabe 1993, Radford 1997),
an AgrO (object agreement) head that projects an AgrOP (either above vP or
between vP and VP) is postulated. The direct object then moves (overtly or
covertly) to the speciWer position of AgrOP to check its nominal features with
AgrO.
10 In a transitive VP, the patient, assumed to be in the speciWer position, is higher
than the other argument. In an intransitive VP, there is at most one NP.
11 The ergative vs. absolutive distinction has been a much-considered matter that has
given rise to a number of diVerent analyses, often correlated with diVerent basic
assumptions about the overall theory of movement, case, and grammatical rela-
tions, which vary somewhat from author to author and across time periods. The
 Notes 215

account sketched here, based on Ura (2000), is chosen for its balance of relative
simplicity, relative eVectiveness, and up-to-dateness. For some alternatives, see
Levin (1983), Marantz (1984), Murasugi (1992), and Bittner and Hale (1996a,
1996b). Although Bittner and Hale’s account of ergativity vs. accusativity and related
typological distinctions is probably the most comprehensive, it doesn’t Wt neatly
into either the GB or the Minimalist paradigm and is too complex to be given justice
here, although some of its ideas are incorporated in places in the analysis presented
here.
12 Ura (2000: 190) notes that the controller of reXexivization in Dyirbal can be either
the A of a transitive verb or the S of an intransitive verb, suggesting that the agent
of a transitive construction has a subject property. However, as Dixon (1994: 138)
makes clear, reXexivization involves adding a reXexive morpheme which indicates
agent/patient coreference and involves absolutive marking of the single expressed
NP. Thus, reXexive clauses are intransitive. In the Minimalist analysis the reXexive
morpheme must be characterized as suppressing the ergative case of the v and one
of the arguments (either agent or patient). The remaining argument is morpho-
logically and syntactically an S that is indistinguishable from the S of any
intransitive clause (i.e. the only NP in a speciWer of IP position and the checker
of absolutive case). Thus, whether there is compelling empirical motivation for
the posited movement of the ergative case checker (agent of transitive) to a
speciWer of IP position in Dyirbal is unclear. In the analysis of Bittner and Hale
(1996a), although the patient (absolutive NP) moves to the speciWer of IP
position, the agent (ergative NP) is subject-like by virtue of its base position as
the speciWer of vP, from which it does not move.
13 It is, of course, curious that the ergative NP, hypothesized to be the lower speciWer
of IP, actually precedes the absolutive NP in Inuit. Ura provides no explanation.
Bittner and Hale (1996a), whose analysis diVers primarily in that it assumes that
the ergative NP remains in the speciWer position of vP, merely suggest that the
linear order is changed at the level of ‘phonetic form’. In other words, they assume
that there is some kind of exceptional linearization rule that Wxes the order that
the phrase structure speciWes.
14 Taking into account the person/number features of the diVerent- and same-
subject morphemes, ignored here for the sake of simplicity, and analysing switch
reference in terms of obviation, Bittner and Hale use the glosses 3Sg.Proximate
and 3Sg.Obviative for SS and DS, respectively.
15 Alternatively, a null expletive pronoun could be posited as a nominative Wller of
the speciWer of IP position, which is what Bittner and Hale propose.
16 Note that the basic phrase structure of Seri is diVerent from that of English and
the other languages considered above because as an SOV language the heads of
phrases follow both speciWers and complements.
216 Notes

17 Other possibilities include calling the intermediate verb phrase an ‘aspect phrase’
(e.g. Baker 1997), or an indirect object agreement phrase or AgrIOP (e.g. Radford
1997). Larson (1988), whose analysis focuses speciWcally on English, simply places
the ‘indirect object’ in the speciWer position of the lower VP and the patient in a
lower position.
18 See, for example, Pollard and Sag (1994: 362). In early versions of the theory,
subjects were assumed to be simply the least oblique of the arguments of a verb.
 References
Ackerman, F. and J. Moore (1999) ‘ ‘‘Telic entity’’ as a proto-property of lexical
predicates’, in M. Butt and T. H. King (eds), Proceedings of the LFG 99 Conference.
Stanford, CA: CSLI Publications.
—— —— (2001) Proto-properties and Grammatical Encoding. Stanford, CA: CSLI
Publications.
Aissen, J. (1983) ‘Indirect object advancement in Tzotzil’, in D. M. Perlmutter (ed.),
Studies in Relational Grammar 1. Chicago: University of Chicago Press, pp. 272–302.
—— (1987) Tzotzil Clause Structure. Dordrecht: D. Reidel.
—— (1988) ‘Extensions of brother-in-law agreement’, in M. Barlow and C. A. Fergu-
son (eds), Agreement in Natural Language: Approaches, theories, descriptions. Stan-
ford, CA: CSLI Publications, pp. 219–35.
—— (1997) ‘On the syntax of obviation’, Language 73: 705–50.
—— (1999a) ‘Agent focus and inverse in Tzotzil’, Language 75: 451–85.
—— (1999b) ‘Markedness and subject choice in optimality theory’, Natural Language
and Linguistic Theory 17: 673–711.
—— (2003) ‘DiVerential object marking: Iconicity vs. Economy’, Natural Language
and Linguistic Theory 21: 435–83.
Allen, B. J. and D. G. Frantz (1983) ‘Advancements and verb agreement in Southern
Tiwa’, in D. M. Perlmutter (ed.), Studies in Relational Grammar 1. Chicago: Uni-
versity of Chicago Press.
——, —— and D. B. Gardiner (1984) ‘Noun incorporation in Southern Tiwa’,
International Journal of American Linguistics 50: 292–311.
Allen, C. L. (1986) ‘Reconsidering the history of like’, Journal of Linguistics 22:
375–409.
—— (1995) Case Marking and Reanalysis: Grammatical relations from Old to Early
Modern English. Oxford: Oxford University Press.
Alsina, A. and S. A. Mchombo (1990) ‘The syntax of applicatives in Chichewa—
problems for a theta theoretic asymmetry’, Natural Language and Linguistic Theory
8: 493–506.
Anderson, J. M. (1971) The Grammar of Case: Towards a localist theory. Cambridge:
Cambridge University Press.
—— (1977) On Case Grammar: Prolegomena to a theory of grammatical relations.
London: Croom Helm.
Anderson, S. (1976) ‘On the notion of subject in ergative languages’, in C. Li (ed.),
Subject and Topic. New York: Academic Press, pp. 1–23.
Aranovich, R. (2000) ‘Split intransitivity and reXexives in Spanish’, Probus 12:
165–86.
218 References

Arnold, J. E. (1997) ‘The inverse system in Mapudungun and other languages’,

Revista de Linguı́stica Teórica y Aplicada 34: 9–48.
—— (1998) Reference form and discourse patterns, Stanford: Ph.D. Dissertation.
Baker, M. C. (1988a) Incorporation: A theory of grammatical function changing.
Chicago: University of Chicago Press.
—— (1988b) ‘Theta theory and the syntax of applicatives in Chichewa’, Natural
Language and Linguistic Theory 6: 353–89.
—— (1996) The Polysynthesis Parameter. Oxford: Oxford University Press.
—— (1997) ‘Thematic roles and syntactic structure’, in L. Haegeman (ed.), Elements
of Grammar: Handbook in generative syntax. Dordrecht: Kluwer, pp. 73–137.
—— K. Johnson and I. Roberts (1989) ‘Passive arguments raised’, Linguistic Inquiry
20: 219–51.
Barry, A. K. (1998) English Grammar: Language as human behavior. Englewood CliVs,
NJ: Prentice Hall.
Barðdal, J. (2001) ‘The perplexity of dat-nom verbs in Icelandic’, Nordic Journal of
Linguistics 24: 47–70.
Bell, S. J. (1976) Cebuano subjects in two frameworks, MIT: Ph.D. Dissertation.
—— (1983) ‘Advancements and ascensions in Cebuano’, in D. M. Perlmutter (ed.),
Studies in Relational Grammar 1. Chicago: University of Chicago Press, pp. 143–218.
Belletti, A. and L. Rizzi (1988) ‘Psych-verbs and u-theory’, Natural Language and
Linguistic Theory 6: 291–352.
Berinstein, A. (1990) ‘On distinguishing surface datives in K’ekchi’, in P. M. Postal
and B. D. Joseph (eds), Studies in Relational Grammar 3. Chicago: University of
Chicago Press, pp. 3–48.
Bickel, B. and Y. P. YAdava (2000) ‘A fresh look at grammatical relations in Indo-
aryan’, Lingua 110: 343–73.
Bittner, M. (1987) ‘On the semantics of the Greenlandic antipassive and related
constructions’, International Journal of American Linguistics 53: 194–231.
—— and K. Hale (1996a) ‘The structural determination of case and agreement’,
Linguistic Inquiry 27: 1–68.
—— and —— (1996b) ‘Ergativity: Towards a theory of a heterogeneous class’, Lin-
guistic Inquiry 27: 531–604.
Blake, B. J. (1990) Relational Grammar. London: Routledge.
—— (1994) Case. Cambridge: Cambridge University Press.
Bloomeld, L. (1917) Tagalog Texts with Grammatical Analysis. Urbana-Champaign,
IL: University of Illinois Studies in Langauge and Literature.
Bolinger, D. L. (1971) The Phrasal Verb in English. Cambridge, MA: Harvard Uni-
versity Press.
Bossong, G. (1991) ‘DiVerential object marking in Romance and beyond’, in
D. Wanner and D. A. Kibbee (eds), New Analyses in Romance Linguistics: Selected
papers from the 18th linguistic symposium on Romance languages. Amsterdam: John
Benjamins, pp. 143–70.
 References 219

Bresnan, J. (1982a) The Mental Representation of Grammatical Relations. Cambridge,

MA: MIT Press.
—— (1982b) ‘The passive in lexical theory’, in J. Bresnan (ed.), The Mental Represen-
tation of Grammatical Relations. Cambridge, MA: MIT Press, pp. 3–86.
—— (1994) ‘Locative inversion and the architecture of universal grammar’, Language
70: 72–131.
—— (2000) Lexical-functional Syntax. Malden, MA: Blackwell Publishing.
—— and J. M. Kanerva (1989) ‘Locative inversion in Chichewa: A case study of
factorization in grammar’, Linguistic Inquiry 20: 1–50.
—— and L. Moshi (1990) ‘Object asymmetries in comparative Bantu syntax’, Lin-
guistic Inquiry 21: 147–85.
Burzio, L. (1986) Italian Syntax: A government-binding approach. Dordrecht:
D. Reidel.
Carnie, A. (2002) Syntax: A generative introduction. Malden, MA: Blackwell Publish-
ing.
Centineo, G. (1986) ‘A lexical theory of auxiliary selection in Italian’, Davis Working
Papers in Linguistics 1: 1–35.
Chomsky, N. (1957) Syntactic Structures. The Hague: Mouton.
—— (1965) Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
—— (1981) Lectures on Government and Binding. Dordrecht: Foris.
—— (1986) Barriers. Cambridge, MA: MIT Press.
—— (1993) ‘A minimalist program for linguistic theory’, in S. J. Keyser and K. Hale
(eds), The View from Building 20. Cambridge, MA: MIT Press, pp. 1–52.
—— (1995) The Minimalist Program. Cambridge, MA: MIT Press.
Comrie, B. (1978) ‘Ergativity’, in W. P. Lehman (ed.), Syntactic Typology. Austin, TX:
University of Texas Press, pp. 329–94.
Coopmans, P. (1989) ‘Where stylistic and syntactic processes meet: Locative inversion
in English’, Language 65: 728–51.
Cooreman, A., B. Fox and T. GivO
n (1988) ‘The discourse deWnition of ergativity:
A study in Chamorro and Tagalog texts’, in R. McGinn (ed.), Studies in Austronesian
Linguistics. Athens, OH: Ohio University Press, pp. 387–425.
Croft, W. (1991) Syntactic Categories and Grammatical Relations: The cognitive
organization of information. Chicago: University of Chicago Press.
—— (1993) ‘Case marking and the semantics of mental verbs’, in J. Pustejovsky (ed.),
Semantics and the Lexicon. Dordrecht: Kluwer, pp. 55–72.
—— (2001) Radical Construction Grammar: Syntactic theory in typological perspective.
Oxford: Oxford University Press.
Czepluch, H. (1982) ‘Case theory and the dative construction’, The Linguistic Review
2: 1–38.
Dahlstrom, A. (1991) Plains Cree Morphosyntax. New York: Garland.
Davies, W. D. (1986) Choctaw Verb Agreement and Universal Grammar. Dordrecht: D.
Reidel.
220 References

De Guzman, V. P. (1988) ‘Ergative analysis for Philippine languages: An analysis’, in R.

McGinn (ed.), Studies in Austronesian Linguistics. Athens, OH: Ohio University
Press, pp. 323–45.
Delancey, S. (1981) ‘An interpretation of split ergativity’, Language 57: 626–57.
—— (1991) ‘Event construals and case role assignment’, in Proceedings of the 17th
Annual Meeting of the Berkeley Linguistics Society. Berkeley: University of California,
pp. 338–53.
Dik, S. C. (1978) Functional Grammar. Amsterdam: North-Holland Publishing Co.
—— (1980) Studies in Functional Grammar. New York: Academic Press.
—— (1997) The Theory of Functional Grammar. Berlin: Mouton de Gruyter.
Dixon, R. M. W. (1972) The Dyirbal Language of North Queensland. Cambridge:
Cambridge University Press.
—— (1979) ‘Ergativity’, Language 55: 59–138.
—— (1994) Ergativity. Cambridge: Cambridge University Press.
—— (2000) ‘A-constructions and o-constructions in Jarawara’, International Journal
of American Linguistics 66: 22–56.
—— and A. Y. Aikhenvald (2000) ‘Introduction’, in R. M. W. Dixon and A. Y.
Aikhenvald (eds), Changing Valency: Case studies in transitivity. Cambridge: Cam-
bridge University Press, pp. 1–28.
Doak, I. G. (1998) Coeur d’Alene grammatical relations, University of Texas, Austin:
Ph.D. Dissertation.
Dobrovie-Sorin , C. (1994) The Syntax of Romanian. Berlin: Mouton de Gruyter.
Dowty, D. R. (1979) Word Meaning in Montague Grammar: The semantics of verbs and
times in generative semantics and in Montague’s PTQ. Dordrecht: D. Reidel.
—— (1982) ‘Grammatical relations and Montague grammar’, in P. Jacobson and G. K.
Pullum (eds), The Nature of Syntactic Representation. Dordrecht: D. Reidel, pp.
79–130.
—— (1991) ‘Thematic proto-roles and argument selection’, Language 67: 547–619.
—— R. Wall and S. Peters (1981) Introduction to Montague Semantics. Dordrecht:
D. Reidel.
Dryer, M. S. (1978) ‘Some theoretical implications of grammatical relations in
Cebuano’. Paper presented at University of Michigan Papers in Linguistics 2, Ann
Arbor, Michigan.
—— (1983) ‘Indirect objects in Kinyarwanda revisited’, in D. M. Perlmutter (ed.),
Studies in Relational Grammar 1. Chicago: University of Chicago Press, pp. 129–40.
—— (1986) ‘Primary objects, secondary objects, and antidative’, Language 61: 808–45.
Durie, M. (1985) A Grammar of Acehnese. Dordrecht: Foris.
—— (1988) ‘Preferred argument structure in an active language: Arguments against
the category ‘‘intransitive subject’’ ’, Lingua 74: 1–25.
Dutra, R. (1987) ‘The hybrid S category in Brazilian Portuguese: Some implications
for word order’, Studies in Language 11: 163–80.
Dziwirek, K. (1994) Polish Subjects. New York: Garland.
 References 221

En
, M. (1989) ‘Pronouns, licensing, and binding’, Natural Language and Linguistic
Theory 7: 51–92.
—— (1991) ‘The semantics of speciWcity’, Linguistic Inquiry 1–26.
Falk, Y. (2001) Lexical-functional Grammar. Stanford, CA: CSLI Publications.
Farkas, D. F. (1978) ‘Direct and indirect object reduplication in Romanian’, in Papers
from the Fourteenth Regional Meeting of the Chicago Linguistic Society. Chicago:
University of Chicago, pp. 88–97.
Farrell, P. (1990) ‘Null objects in Brazilian Portuguese’, Natural Language and
Linguistic Theory 8: 325–46.
—— (1994a) Thematic Relations and Relational Grammar. New York: Garland.
—— (1994b) ‘The locative alternation and multistratalism’, Linguistics 32: 5–42.
—— (1998) ‘Comments on the paper by Lieber’, in S. G. Lapointe, D. Brentari, and P.
Farrell (eds), Morphology and its Relation to Syntax and Morphology. Stanford, CA:
CSLI Publications, pp. 34–53.
—— (2001) ‘Functional shift as category underspeciWcation’, English Language and
Linguistics 5: 109–30.
—— (2004) ‘Psych verbs with quasi-objects’, in A. S. da Silva A. Torres, and M.
Gonçalves (eds), Linguagem, cultura, e cognição: Estudos de linguı́stica cognitiva.
Coimbra: Almedina, pp. 367–84.
—— S. A. Marlett , and D. M. Perlmutter (1991) ‘Notions of subjecthood and
switch reference: Evidence from Seri’, Linguistic Inquiry 22: 431–55.
Filip, H. (1997) Aspect, Eventuality Types and Nominal Semantics. New York: Garland.
Fillmore, C. J. (1968) ‘The case for case’, in E. Bach and R. T. Harms (eds), Universals
in Linguistic Theory. New York: Holt, Rinehart, and Winston, pp. 1–88.
Finer, D. L. (1985) ‘The syntax of switch reference’, Linguistic Inquiry 16: 35–55.
Foley, W. A. and R. D. Van Valin, Jr. (1984) Functional Syntax and Universal
Grammar. Cambridge: Cambridge University Press.
Franks, S. (1995) Parameters of Slavic Morphosyntax. Oxford: Oxford University
Press.
Gerdts , D. B. (1988a) Object and Absolutive in Halkomelem Salish. New York:
Garland.
—— (1988b) ‘Antipassives and causatives in Ilokano: Evidence for an ergative analy-
sis’, in R. McGinn (ed.), Studies in Austronesian Linguistics. Athens, OH: Ohio
University Press, pp. 295–321.
Gibson , J. D. (1980) Clause union in Chamorro and in universal grammar, University
of California, San Diego: Ph.D. Dissertion.
GivO
n , T. (1978) ‘DeWniteness and referentiality’, in J. Greenberg (ed.), Universals of
Human Language, Vol. 4. Stanford, CA: Stanford University Press, pp. 291–330.
—— (ed.) (1984) Syntax: A functional-typological introduction. Amsterdam: John
Benjamins.
—— (ed.) (1997) Grammatical Relations: A functionalist perspective. Amsterdam: John
Benjamins.
Goldberg, A. E.(1991) ‘A semantic account of resultatives’, Linguistic Analysis 21: 66–96.
222 References

Goldberg , A. E. (1995) Constructions: A construction grammar approach to argument

structure. Chicago: University of Chicago Press.
Green , G. M. (1974) Semantic and Syntactic Regularity. Bloomington: Indiana Uni-
versity Press.
Grimshaw , J. (1990) Argument Structure. Cambridge, MA: MIT Press.
Gropen , J., S. Pinker, M. Hollander , R. Goldberg , and R. Wilson (1989) ‘The
learnability and acquisition of the dative alternation’, Language 65: 203–57.
Gruber , J. S. (1965) Studies in lexical relations, MIT: Ph.D. dissertation.
—— (2001) ‘Thematic relations in syntax’, in M. Baltin and C. Collins (eds), The
Handbook of Contemporary Syntactic Theory. Malden, MA: Blackwell, pp. 257–98.
Hale , K. and S. J. Keyser (1993) ‘On argument structure and the lexical expression of
syntactic relations’, in K. Hale and S. J. Keyser (eds), The View from Building 20:
Essays in honor of Sylvain Bromberger. Cambridge, MA: MIT Press, pp. 53–109.
Halliday , M. A. K. (1973) Explorations in the Functions of Language. New York:
Elsevier North-Holland.
—— (1985) An Introduction to Functional Grammar. London: Arnold.
Harris , A. (1982) ‘Georgian and the unaccusative hypothesis’, Language 58: 290–306.
—— (1990) ‘Georgian: A language with active case marking’, Lingua 80: 35–53.
Henderson , T. S. T. (1971) ‘Participant-reference in Algonkin’, Cahiers linguistiques
d’Ottawa 1: 27–49.
Herriman , J. (1995) The Indirect Object in Present-day English. Göteborg: Acta
Universitatis Gothoburgensis.
Herslund , M. (1997) ‘Passive and antipassive in a functional description of French
reXexive verbs’, Hermes, Journal of Lingusitics 19: 75–92.
Hockett , C. (1958) A Course in Modern Linguistics. New York: Macmillan.
Holisky , D. A. (1987) ‘The case of the intransitive subject in Tsova-Tush (Batsbi)’,
Lingua 71: 103–32.
Jackendoff, R. S. (1972) Semantics in Generative Grammar. Cambridge, MA: MIT
Press.
—— (1977) X-bar Syntax: A study of phrase structure. Cambridge, MA: MIT Press.
—— (1987) ‘The status of thematic relations in linguistic theory’, Linguistic Inquiry 18:
369–411.
—— (1990) Semantic Structures. Cambridge, MA: MIT Press.
Jaeggli, O. A. (1986) ‘Passive’, Linguistic Inquiry 17: 587–622.
Jespersen, O. (1927) A Modern English Grammar on Historical Principles. London:
Allen & Unwin.
Johnson, D. E. and P. M. Postal. (1980) Arc Pair Grammar. Princeton, NJ: Princeton
University Press.
Joshi, S. (1993) Selection of grammatical and logical functions in Marathi, Stanford:
Ph.D. Dissertation.
Kaplan , J. P. (1995) English Grammar: Principles and Facts. Englewood CliVs, NJ:
Prentice Hall.
 References 223

Kato , M. A. (1993) ‘The distribution of null pronouns and null elements in object
position in Brazilian Portuguese’, in W. J. Ashby, M. Mithun, G. Perissinotto, and E.
Raposo (eds), Linguistic Perspectives on the Romance Languages: Selected papers from
the 21st symposium on Romance languages. Amsterdam: John Benjamins, pp. 225–36.
Keenan , E. L. and B. Comrie (1977) ‘Noun phrase accessibility and universal gram-
mar’, Linguistic Inquiry 8: 63–100.
Kimenyi , A. (1980) A Relational Grammar of Kinyarwanda. Berkeley: University of
California Press.
Kiparsky , P. (1998) ‘Partitive case and aspect’, in M. Butt and W. Geuder (eds), The
Projection of Arguments: Lexical and compositional factors. Stanford, CA: CSLI
Publications, pp. 265–307.
—— (2001) ‘Structural case in Finnish’, Lingua 111: 315–76.
—— (2002) ‘On the architecture of Pānini’s grammar’. MS, Stanford University.
http://www.stanford.edu/~kiparsky/. _
—— and F. Staal (1969) ‘Syntactic and semantic relations in Pānini’, Foundations of
Language 5: 83–117. _
Kishimoto , H. (1996) ‘Split intransitivity in Japanese and the unaccusative hypo-
thesis’, Language 72: 248–86.
Kisseberth , C. W. and M. I. Abasheikh (1977) ‘The object relationship in Chi-
mwi:ni, a bantu language’, in P. Cole and J. Sadock (eds), Syntax and Semantics 8:
Grammatical relations. New York: Academic Press, pp. 179–218.
Klaiman , M. H. (1990) Grammatical Voice. Cambridge: Cambridge University Press.
—— (1993) ‘The relationship of inverse voice and head-marking in Arizona Tewa and
other Tanoan languages’, Studies in Language 17: 343–70.
Koopman , H. and D. Sportiche (1991) ‘The position of subjects’, Lingua 85: 211–58.
Kroeger , P. (1993) Phrase Structure and Grammatical Relations in Tagalog. Stanford,
CA: CSLI Publications.
Kuroda , S.-Y. (1988) ‘Whether we agree or not: A comparative syntax of English and
Japanese’, Lingvisticae Investigationes 12: 1–47.
La Fauci, N. (1988) Oggetti e soggetti nella formazione della morfosintassi romanza.
Pisa: Giardini Editori e Stampatori.
La Polla, R. J. (1993) ‘Arguments against ‘‘subject’’ and ‘‘direct object’’ as viable
concepts in Chinese’, Bulletin of the Institute of History and Philology 63: 759–813.
Lakoff , G. (1971) ‘On generative semantics’, in D. D. Steinberg and L. A. Jakobovits
(eds), Semantics: An interdisciplinary reader. Cambridge: Cambridge University
Press, pp. 232–96.
Langacker , R. W. (1987) Foundations of Cognitive Grammar, Volume 1: Theoretical
prerequisites. Stanford, CA: Stanford University Press.
—— (1991) Foundations of Cognitive Grammar, Volume 2: Descriptive application.
Stanford, CA: Stanford University Press.
—— (1999) Grammar and Conceptualization. Berlin: Mouton de Gruyter.
Larson , R. K. (1988) ‘On the double object construction’, Linguistic Inquiry 19: 335–91.
224 References

Legendre , G. (1987) Topics in French syntax, University of California, San Diego:

Ph.D. Dissertation.
—— (1989a) ‘Inversion with certain French experiencer verbs’, Language 65: 752–82.
—— (1989b) ‘Unaccusativity in French’, Lingua 79: 95–164.
—— (1990) ‘French impersonal constructions’, Natural Language and Linguistic The-
ory 8: 81–128.
Lesourd , P. (1976) ‘Verb agreement in Fox’, in J. Aissen and J. Hankamer (eds),
Harvard Studies in Syntax and Semantics 2. Cambridge, MA: Harvard University
Linguistics Department, pp. 445–528.
Levin , B. (1983) On the nature of ergativity, MIT: Ph.D. Dissertation.
—— and M. Rappaport Hovav (1995) Unaccusativity: At the syntax–lexical seman-
tics interface. Cambridge, MA: MIT Press.
Marantz , A. (1984) On the Nature of Grammatical Relations. Cambridge, MA: MIT
Press.
Marlett , S. A. (1984) ‘Switch reference and subject raising in Seri’, in E.-D. Cook and
D. B. Gerdts (eds), Syntax and Semantics 16: The syntax of Native American
languages. New York: Academic Press, pp. 247–68.
Matthews , P. (1997) The Concise Oxford Dictionary of Linguistics. Oxford: Oxford
University Press.
May , R. (1985) Logical Form: Its structure and derivation. Cambridge, MA: MIT Press.
McCawley , J. D. (1971) ‘Prelexical syntax’, in R. J. O’Brien (ed.), 22nd Annual Round
Table on Languages and Linguistics, 19–33. Washington, DC: Georgetown University.
—— (1973) ‘Syntactic and logical arguments for semantic structures’, in O. Fujimura
(ed.), Three Dimensions of Linguistic Theory, 259–376. Tokyo: Tec Co.
McClure , W. (1990) ‘A lexical semantic explanation for unaccusative mismatches’, in
K. Dziwirek, P. Farrell, and E. Mejı́as-Bikandi (eds), Grammatical Relations: A cross-
theoretical perspective. Stanford, CA: CSLI Publications, pp. 305–18.
McGinnis , M. (2002) ‘Object asymmetries in a phase theory of syntax’, in J. T. Jensen
and G. v. Herk (eds), Proceedings of the 2001 CLA Annual Conference. Department of
Linguistics: University of Ottawa, pp. 245–56.
Megerdoomian , K. (2000) ‘Aspect and partitive objects in Finnish’, in R. Billerey and
B. D. Lillehaugen (eds), Proceedings of the 19th West Coast Conference on Formal
Linguistics. Somerville, MA: Cascadilla Press, pp. 316–28.
Merlan , F. (1985) ‘Split intransitivity’, in J. Nichols and A. Woodbury (eds), Gram-
mar Inside and Outside the Clause: Some perspectives from the Weld. Cambridge:
Cambridge University Press, pp. 324–62.
Miller , P. H. and I. A. Sag (1997) ‘French clitic movement without clitics or
movement’, Natural Language and Linguistic Theory 15: 573–639.
Minger , D. L. (2002) An analysis of grammatical relations and case marking in
Icelandic, University of California, Davis: M.A. Thesis.
Mithun , M. (1991) ‘Active/agentive case marking’, Language 67: 510–46.
Mohanan , T. (1994) Argument Structure in Hindi. Stanford, CA: CSLI Publications.
 References 225

Moore , J. and D. M. Perlmutter (2000) ‘What does it take to be a dative subject?’

Natural Language and Linguistic Theory 18: 373–416.
Moravcsik , E. A. (1974) ‘Object–verb agreement’, Working Papers on Language
Universals 13: 25–140.
Morimoto , Y. (2002) ‘Prominence mismatches and diVerential object marking in
Bantu’, in M. Butt and T. H. King (eds), Proceedings of the LFG 02 Conference.
Stanford, CA: CSLI Publications.
Mulder , J. G. (1994) Ergativity in Coast Tshimian. Berkeley: University of California
Press.
Murasugi , K. (1992) Crossing and nested paths: NP movement in accusative and
ergative languages, MIT: Ph.D. Dissertation.
Musgrave , S. (2001) Non-subject arguments in Indonesian, University of Melbourne:
Ph.D. Dissertation.
—— (To appear) ‘The grammatical function OBJ in Indonesian’, in S. Musgrave and
P. K. Austin (eds), Voice and Grammatical Functions in Austronesian. CSLI Publi-
cations.
Napoli , D. J. (1993) Syntax: Theory and problems. Oxford: Oxford University Press.
Nichols , J. (1986) ‘Head-marking and dependent-marking grammar’, Language 62:
56–119.
Nikolaeva , I. (1999) ‘Object agreement, grammatical relations, and information
structure’, Studies in Language 23: 377–407.
Noonan , M. (1992) A Grammar of Lango. Berlin: Mouton de Gruyter.
Oehrle, R. T. (1976) The grammatical status of the English dative alternation, MIT:
Ph.D. Dissertation.
——, E. Bach, and D. Wheeler (eds) (1988) Categorial Grammars and Natural
Language Structures. Dordrecht: D. Reidel.
Ortiz De Urbina, J. (1989) Paramaters in the Grammar of Basque: A GB approach to
Basque syntax. Dordrecht: Foris.
Park, H.-Y. (2002) ‘Object asymmetries in Korean’, in M. Butt and T. H. King (eds),
Proceedings of the LFG 02 Conference. Stanford, CA: CSLI Publications.
Payne, T. E. (1982) ‘Role and reference related subject properties and ergativity in
Yup’ik and Tagalog’, Studies in Language 6: 75–106.
Perlmutter, D. M. (1976) ‘Subject downgrading in Portuguese’, in J. Schmidt-Rade-
feldt (ed.), Readings in Portuguese Linguistics. Amsterdam: North-Holland,
pp. 93–138.
—— (1978) ‘Impersonal passives and the unaccusative hypothesis’, in Proceedings of
the Fourth Annual Meeting of the Berkeley Linguistic Society. University of California,
Berkeley, pp. 157–89.
—— (1982) ‘Syntactic representation, syntactic levels, and the notion of subject’, in
P. Jacobson and G. Pullum (eds), The Nature of Syntactic Representation. Dordrecht:
D. Reidel, pp. 283–340.
—— (ed.) (1983) Studies in Relational Grammar 1. Chicago: University of Chicago
Press.
226 References

Perlmutter, D. M. (1984) ‘Working 1s and inversion in Italian, Japanese, and

Quechua’, in D. M. Perlmutter and C. Rosen (eds), Studies in Relational Grammar
2 . Chicago: University of Chicago Press, pp. 292–330.
—— (1989) ‘Multiattachment and the unaccusative hypothesis: The perfect auxiliary
in italian’, Probus 1: 63–119.
—— and P. M. Postal (1983) ‘Some proposed laws of basic clause structure’, in D. M.
Perlmutter (ed.), Studies in Relational Grammar 1. Chicago: University of Chicago
Press, pp. 81–128.
—— and P. M. Postal (1984) ‘The 1-advancement exclusiveness law’, in D. M.
Perlmutter and C. Rosen (eds), Studies in Relational Grammar 2. Chicago: Univer-
sity of Chicago Press, pp. 81–125.
—— and C. Rosen (eds) (1984) Studies in Relational Grammar 2. Chicago: University
of Chicago Press.
—— and R. Rhodes (1988) ‘Thematic-sytactic alignments in Ojibwa: Evidence for
subject–object reversal’. Paper presented at the Annual Meeting of the Linguistic
Society of America.
Pinker, S. (1989) Learnability and Cognition: The acquisition of argument structure.
Cambridge, MA: MIT Press.
Pollard, C. and I. A. Sag (1994) Head-driven Phrase Structure Grammar. Chicago:
CSLI & University of Chicago Press.
Pollock, J.-Y. (1989) ‘Verb movement, universal grammar, and the structure of IP’,
Linguistic Inquiry 20: 365–424.
Postal, P. M. (1977) ‘Antipassive in French’, Lingvisticae Investigationes 1: 333–74.
—— (1986) Studies of Passive Clauses. Albany: State University of New York Press.
—— (1989) Masked Inversion in French. Chicago: University of Chicago Press.
—— (1990) ‘French indirect object demotion’, in P. M. Postal and B. D. Joseph (eds),
Studies in Relational Grammar 3. Chicago: University of Chicago Press, pp. 104–200.
—— and B. D. Joseph (eds) (1990) Studies in Relational Grammar 3. Chicago:
University of Chicago Press.
Radford, A. (1988) Transformational Grammar: A Wrst course. Cambridge: Cambridge
University Press.
—— (1997) Syntactic Theory and the Structure of English: A minimalist approach.
Cambridge and New York: Cambridge University Press.
Rappaport Hovav , M. and B. Levin (1992) ‘-Er nominals: Implications for the
theory of argument structure’, in T. Stowell and E. Wehrli (eds), Syntax and
Semantics 26: Syntax and the lexicon. San Diego: Academic Press, pp. 127–53.
Rappaport , M. and B. Levin (1988) ‘What to do with u-roles’, in W. Wilkins (ed.),
Syntax and Semantics 21: Thematic relations. San Diego: Academic Press, pp. 7–36.
Rex , S. M. (2001) Unaccusativity in Spanish, University of California, Davis: Ph.D.
Dissertation.
Rhodes , R. (1976) The morphosyntax of the Central Ojibwa verb, University of
Michigan: Ph.D. Dissertation.
 References 227

Roberts , I. G. (1987) The Representation of Implicit and Dethematized Subjects.

Dordrecht: Foris Publications.
Roberts , L. (1995) ‘Pivots, voice and macroroles: From Germanic to universal
grammar’, Australian Journal of Linguistics 15: 157–214.
Roeper , T. (1987) ‘Implicit arguments and the head-complement relation’, Linguistic
Inquiry 18: 267–310.
Rosen , C. (1984) ‘The interface between semantic roles and initial grammatical
relations’. D. M. Perlmutter and C. Rosen (eds) Studies in Relational Grammar 2.
Chicago: University of Chicago Press, pp. 38–77.
—— (1988) The Relational Structure of ReXexive Clauses: Evidence from Italian. New
York: Garland.
—— (1990) ‘Rethinking Southern Tiwa: The geometry of a triple-agreement lan-
guage’, Language 66: 669–713.
—— and K. Wali (1989) ‘Twin passives, inversion, and multistratalism in Marathi’,
Natural Language and Linguistic Theory 7: 1–50.
Rounds , C. (2001) Hungarian: An essential grammar. London: Routledge.
Ruwet , N. (1972) Théorie syntaxique et syntaxe du français. Paris: Editions du Seuil.
Ryder , M. E. (1999) ‘Bankers and blue-chippers: An account of -er formations in
present-day English’, English Language and Linguistics 3: 269–97.
Sag , I., T. Wasow, and E. M. Bender (2003) Syntactic Theory: A formal introduction.
Stanford, CA: CSLI Publications.
Sapir , E. (1917) ‘Review of Uhlenbeck, C. C., ‘Het passieve Karakter van het Verbum
transitivum of van het Verbum actionis in Talen van Noord America’ (‘The passive
character of the transitive verb or of the active verb in languages of North Amer-
ica’)’, International Journal of American Linguistics 1: 82–86.
Schachter , P. (1976) ‘The subject in Philippine languages: Topic, actor, actor–topic,
or none of the above’, in C. Li (ed.), Subject and Topic. New York: Academic Press,
pp. 491–518.
—— and F. T. Otanes (1972) Tagalog Reference Grammar. Berkeley: University of
California Press.
Seki , L. (2000) Gramática do Kamaiurá, lı́ngua Tupı́-Guaranı́ do Alto Xingu. Campi-
nas: Editora da Unicamp.
Siewierska , A. (1991) Functional Grammar. London: Routledge.
Sigurðsson , H. Á. (2002) ‘To be an oblique subject: Russian vs. Icelandic’, Natural
Language and Linguistic Theory 20: 691–724.
Silverstein , M. (1976) ‘Hierarchy of features and ergativity’, in R. M. W. Dixon (ed.),
Grammatical Categories in Australian Languages. Canberra: Australian Institute of
Aboriginal Studies, pp. 112–71.
Singh , J. D. (2001) Kārake. Westerville, OH: Pānini Foundation.
_
Starosta , S. (1988) The Case for Lexicase: An outline of lexicase grammatical theory.
London: Pinter Publishers.
Stowell , T. (1981) Origins of phrase structure, MIT: Ph.D. Dissertation.
228 References

Talmy, L. (1996) ‘Fictive motion in language and ‘‘ception’’’, in P. Bloom, M. A.

Peterson, L. Nadel, and M. F. Garrett (eds), Language and Space. Cambridge, MA:
MIT Press, pp. 211–76.
—— (2000) Toward a Cognitive Semantics: Concept structuring systems (vol. 1); typ-
ology and process in concept structuring (vol. 2). Cambridge, MA: MIT Press.
Tesni
Ere, L. (1959) Éléments de syntaxe structurale. Paris: Klincksieck.
Tomlin, R. S. (1986) Basic Word Order: Functional principles. London: Croom Helm.
Ura, H. (1999) ‘Checking theory and dative subject constructions in Japanese and
Korean’, Journal of East Asian Linguistics 8: 223–54.
—— (2000) Checking Theory and Grammatical Functions in Universal Grammar.
Oxford: Oxford University Press.
—— (2001) ‘Case’, in M. Baltin and C. Collins (eds), The Handbook of Contemporary
Syntactic Theory. Malden, MA: Blackwell, pp. 334–73.
Van Valin, R. D., Jr. (1990) ‘Semantic parameters of split intransitivity’, Language 66:
221–60.
—— (1991) ‘Another look at Icelandic case marking and grammatical relations’,
Natural Language and Linguistic Theory 9: 145–94.
—— (1993) ‘A synopsis of role and reference grammar’, in R. D. Van Valin Jr. (ed.),
Advances in role and Reference Grammar. Amsterdam and Philadelphia: John
Benjamins, pp. 1–164.
—— (To appear) Exploring the Syntax–Semantics Interface. Cambridge: Cambridge
University Press.
—— and R. J. La Polla (1997) Syntax: Structure, meaning, and function. Cambridge:
Cambridge University Press.
Vendler, Z. (1967) Linguistics in Philosophy. Ithaca, NY: Cornell University Press.
Vestergaard, T. (1977) Prepositional Phrases and Prepositional Verbs. The Hague:
Mouton.
Watanabe, A. (1993) Agr-based case theory and its interaction with the a-bar system,
MIT: Ph.D. Dissertation.
Whaley, L. J. (1993) The status of obliques in linguistic theory, State University of New
York, BuValo: Ph.D. Dissertation.
—— (1997) Introduction to Typology: The unity and diversity of language. New York:
Sage Publications.
Wierzbicka, A. (1996) Semantics: Primes and universals. Oxford: Oxford University
Press.
Wolfart, C. H. and J. F. Carroll (1981) Meet Cree: A guide to the language. Lincoln:
University of Nebraska Press.
Woodbury, A. (1977) ‘Greenlandic eskimo, ergativity, and relational grammar’, in P.
Cole and J. M. Sadock (eds), Syntax and Semantics 8: Grammatical relations. New
York: Academic Press, pp. 307–36.
Yang, B.-S. (1996) ‘A role and reference grammar account of unaccusativity: Split
intransitivity’, Linguistics 4: 77–93.
 References 229

Zaenen , A. (1993) ‘Unaccusativity in Dutch: Integrating syntax and lexical semantics’,

in J. Pustejovsky (ed.), Semantics and the Lexicon. Dordrecht: Kluwer, pp. 129–61.
—— J. Maling , and H. Thr
Ainsson (1985) ‘Case and grammatical functions: The
Icelandic passive’, Natural Language and Linguistic Theory 3: 441–83.
Zagona , K. (2002) The Syntax of Spanish. Cambridge: Cambridge University Press.
Zubizarreta , M. L. (1987) Levels of Representation in the Lexicon and in the Syntax .
Dordrecht: Foris Publications.
This page intentionally left blank
 Index
absolutive 116 in 161–3; quasi-objects in 106–8,
accusative languages 14, 45, 46–9, 152–3, 161–5, 212–13 n17
202–3 nn2, 3
Acehnese: and Relational Grammar 210–11 Canadian native languages see Algonquian
n8; as split-intransitive language languages; Halkomelem; Inuit
59–61, 204 n11; unaccusativity in case 8–14, 199 n3
124 case marking 104, 105
active languages see split-intransitive Chamorro: applicative constructions in
languages 90
active voice 9 Chilean native languages see Mapudungun
actor macroroles 153–4 Choctaw: as head-marking language 118;
adjuncts: in English 31–3, 202 n24; in quasi-subjects in 104; relation-
Tagalog: 35–6 marking devices in 96; as split-
African languages see Bantu languages; intransitive language 56, 58–9, 117–18,
Kinyarwanda 203–4 n9
agents: preverbal noun phrases 4–6, 199 n2 chômage 115
agreement systems 17 classical Transformational Grammar 173–5,
Algonquian languages: inverse voice 213 n1
in 75–6 clause structure 113–16, 206 n1; and syn-
antipassive voice 64–5, 71–4, 152–3, 211 n12 tactic structure 143–8
applicative constructions 65, 83–90, 95, Cognitive Grammar 42, 168–72
193–7, 204 nn17,18, 205 n20, 216 n17 constituent order constraints: English 2, 23
applicative objects: in Indonesian 160–61, Cree see Plains Cree
212 n16
Arc-Pair Grammar 134 dative subjects: in Icelandic 15–16, 99–101,
Asian languages see Korean 127–30, 155–9, 190–92, 199 n5, 208–9
Australian languages see Dyirbal; Pitjant- nn9, 10, 211–12 nn14, 15
jatjara dependent-marking languages: Eastern
Austronesian languages see Acehnese; Cha- Pomo 186–7; English 2, 3;
morro; Indonesian; Philippine voice Halkomolem 52–4, 203 n5; Latin and
system; Tagalog Greek 7
direct object 19–20, 45–9; see also quasi-
Bantu languages: applicative constructions objects; subject and direct object
in 89, 90; case marking in 105; rela- double-object construction: in Functional
tion-marking devices in 97 Grammar; with indirect object 22, 24
see also Kinyarwanda Dyirbal: antipassive voice in 64–5, 72–3, 74;
Basque: as ergative language 49–50 as ergative language 50–52, 183, 184,
Brazilian native languages see Kamaiurá; 215 nn15, 16; macroroles in 145–6;
Jarawara voice constructions in 148–9
Brazilian Portuguese: as accusative
language 46–9; indirect object Eastern Pomo 186–7
in 17–19, 26–7, 45, 46, 200 nn9,10; English: applicative constructions in 89,
Lexical Functional Grammar 90, 195; Cognitive Grammar
in 132–3; macroroles in 142, 144, 210 in 170–72; as dependent-marking
n7; marking in 106; noun phrases language 1–2; as head-marking
232 Index
English (cont.): Icelandic: case in 12–13; dative subjects
language 2; indirect object in 21–7, in 15–16, 99–101, 127–30, 155–9,
200 n14; macroroles in 141, 142; mor- 190–92, 199 n5, 208–9 nn9, 10;
phological indications in 2, 3; noun 211–12 nn14, 15; noun phrases
phrases in 1–2, 3–5, 6, 21–3, 29, 31, 40, in 12, 15; relation-marking devices
63, 98, 200 n12; obliques in 29, 30; in 96;
pronouns in 2; pseudoclefts in 33–5, subject and direct object in 14–15
37, 201–2 nn23, 25; relation-marking indirect object 16–28, 199–200 nn8, 11, 14,
devices in 96; suffixes in 2; transitive 19; in Brazilian Portuguese 46–9; in
and intransitive clauses in 44–5; Transformational Grammar 180–81,
unaccusativity in 124–6; voice 190–92, 193–7
in 62–7; verbs in 1–2, 3–6; voice Indonesian: applicative objects in 160–61,
constructions in 150 212 n16; quasi-objects in 108–9
see also Old English see also Acehnese
ergative languages 45, 49–54, 116–17, 152–3, instrumental suffixes: in Sanskrit 10
203 n4, 211 n12; and Transformational intransitive clauses see transitive and
Grammar 181–5, 188, 214–15 nn10, 11, 12 intransitive clauses:
European languages see Basque; English; typological parameters
Finnish; French; Greek, classical; intransitive verbs see transitive and
Hungarian; Icelandic; Italian; Latin; intransitive verbs: unaccusativity
Old English; Romance languages; Inuit: obliques in 29; Transformational
Romanian; Spanish Grammar in 184–5;
voice constructions in 149, 211 n9
Final 1 law 115–16 inverse voice 65, 74–83, 121–2, 204 n16
Finnish: case in 13; obliques in 31 Italian: quasi-subjects in 104
French: Cognitive Grammar in 171; Lexical
Functional Grammar in 132; noun Jarawara: inverse voice in 77–83, 121–2, 204
phrases in 102; passive voice in 70; n16; multiple subjects in 189–90;
psychological verbs in 102–4; quasi- preverbal pronominal slots in 154;
subjects in 119, 159–60 pronouns in 81; relation-marking de-
function splitting 96–111 vices in 96; voice constructions
Functional Grammar 42, 166–8, 213 n18 in 150–51

Government-Binding theory: Transform- Kamaiurá: direct object in 19–20, 45, 46;

ational Grammar 176–8, 213 nn2, 3 macroroles in 143–4; noun phrases
Greek, classical 7 in 20, 57; passive voice in 68; prefixes
in 20–21; relation-marking devices
Halkomelem: antipassive voice in 73; in 96; as split-intransitive
applicative constructions in 84–8, language 54–6, 57, 203 n6, 204 n10;
89–90, 193–5, 204–5 nn18, 19; clause unaccusativity in 126
structure in 113–14; dependent-mark- Kinyarwanda: applicative constructions
ing in 52–4, 203 n5; morphological in 88–9, 90; relation-marking devices
marking in 117; obliques in 28, 30, in 96, 97
37; passive voice in 68, 69–70, 71, Korean: applicative constructions in
192–3; relation-marking devices 89, 90
in 96–7
Head-Driven Phrase Structure Gram- languages: types of 148–53
mar 197 Latin 7–8
head-marking languages 44–5, 202 n1; lexical decomposition: Role and Reference
Choctaw 118; English 2; Sanskrit 7 Grammar 42–39, 135–9, 209 n2
Hebrew: case marking in 105 Lexical-Functional Grammar 26–7, 43,
Hungarian: obliques in 28, 29 130–33, 201 nn17, 18, 20, 209 n11
 Index 233
Lexicase Grammar 10, 14–16 Pomo see Eastern Pomo
locative inversion construction 98–9 Portuguese see Brazilian Portuguese
predicates: in Role and Reference
macroroles: semantics 15–16, 139–44, Grammar 136–8
153–4, 199 n4, 209 nn3–5, 210 nn6, 7 prefixes: in Kamaiurá 20–21
Mapudungun: inverse voice in 74–5, 76–7 prepositional phrases: in applicative
markers see obliques constructions 195–7
Mexican native languages see Seri; Tzotzil prepositions: in English 2, 6
Minimalist approach: Transformational preverbal noun phrases 4–5, 6, 199 n1
Grammar 178–81, 197, 212–13 nn4–10 preverbal pronominal slots: in Jarawara 154
morphemes 1 primary-object languages 45–6
morphological indications: English 2, 3; privileged syntactic arguments
Latin 8; Sanskrit 6–7 (PSA) 146–7, 149–51, 152
morphological markers see obliques pronouns: in English 2; in Jarawara 81
multiple subjects: Transformational pseudoclefts: in English 33–5, 37, 201–2
Grammar 188–93, 215 n16 nn23, 25
psychological verbs 99–104, 107, 192, 206
North American native languages see Al- n26
gonquian languages; Choctaw; see also quasi-objects; quasi-subjects
Eastern Pomo; Halkomelem; Inuit;
Plains Cree; Southern Tiwa quasi-objects 104–111, 160–65, 212–13 n17
noun phrases: in Brazilian quasi-subjects 98–104, 118–19, 159–60, 188,
Portuguese 161–3; in Dyirbal 51; in 205 n22
English 1–2, 3–5, 6, 21–3, 29, 31, 40, 63,
98, 200 n12; in French 102; in Halk- relation-marking devices 96, 97; in
omelem 86; in Icelandic 12, 15; in Brazilian Portuguese 106; in
Jarawara 79; in Kamaiurá 20, 57; in ergative languages 49, 52; in
Philippine voice system 91–2; in Pitjantjatjara 104; in Seri 70–71; in
Tagalog 36, 94; in Transformational split-intransitive languages 56–9; and
Grammar 174–5, 176–7, 178–9, 181, voice 6
183, 184, 185–7, 188–9, 190, 191, 192, Relational Grammar 43, 112–34, 207 nn3, 4
194, 195 relative clauses: in Tagalog 93–4
Role and Reference Grammar 42, 135–72,
object see direct object; indirect object 199 n6, 209 n1
oblique law: clause structure 115 Romance languages: and indirect
oblique subjects 98, 99–104, 205 n22 object 17–19
obliques 28–31, 37, 52–3, 77, 108, 201 nn21, Romanian: case marking in 105
22
Old English: Cognitive Grammar in 171–2; Sanskrit 6–7, 8–11, 67
dative subject frames in 199 n5 semantic functions: Functional
Grammar 166–8
Pānini: and Sanskrit 6, 9, 10 semantic roles 38–41, 113, 202 n26 see also
_
passive and switch reference: Seri 154–5, macroroles: semantics
211 n13 Seri: passive construction in 68; marking
passive construction: in English 23, 200, in 70–71, 204 n13; multiple subjects
nn14, 15; in Seri 68; in 188–97, 215 n16; passive and switch
Transformational Grammar 192 reference in 154–5, 211 n13
passive voice 9, 65–71 single-object construction: indirect
Philippine voice system 91–5, 119–20, 205 n21 object 24
see also Tagalog South American languages see Brazilian
Pitjantjatjara: case marking in 104 Portuguese; Jarawara; Kamaiurá;
Plains Cree: inverse voice in 75–6 Mapudungun
234 Index
Southern Tiwa: applicative constructions transitive and intransitive verbs: unaccusa-
in 85; passive voice in 67–8, 204 n12 tivity of 124–6
Spanish: noun phrases in 105–6 Turkish: case marking in 105
split-intransitive languages 45, 54–62, typological parameters 44–9
117–18, 123, 152–3, 185–7, 203 nn7–9, Tzotzil: applicative constructions in 90
204 n10, 206–7 n2, 215 nn14, 15
stratal uniqueness law: clause unaccusativity 8–14, 122–7, 207 nn5, 6, 7,
structure 115–16 208 n8
subject see multiple subjects:
Transformational Grammar valences: verbs 10–11
subject and direct object 14–16, 199 n7 verb agreement: English 6
suffixes: in English 2, 6; in Sanskrit 9, 10 verb phrases: Transformational
syntactic functions 167, 197–8 Grammar 174–5, 176, 178, 179, 183,
syntax 112 195
verbs: case of 8–14; in English 1–2, 3–6; in
Tagalog: adjunct-fronting construction Icelandic 12–13; in Latin 7–8;
in 35–6, 202 n24; noun phrases in 36, psychological 99–104, 192, 206 n6; in
94; quasi-objects in 110; relation- Sanskrit 6–7; in split-intransitive
marking devices in 96, 97; relative languages 45; see also active voice;
clauses in 93–4; voice-alternations indirect object
in 120–21 subject and direct object; passive
thematic roles: theories of grammar 7 voice
Theta Criterion: noun phrases 41 voice 9–14, 62–95
Tiwa see Southern Tiwa voice-alternations: in Tagalog 120–121
Transformational Grammar 43, 130, 173–98 voice constructions 148–52, 211 nn9, 10, 11
transitive and intransitive clauses: typo-
logical parameters 44–5 X-bar theory 176