Scribd
Upload
96 views

Geometry of Thought (Abstract)

The document summarizes Peter Gärdenfors' theory of conceptual spaces, which proposes a geometric framework for representing conceptual knowledge. It introduces conceptual spaces as a level of knowledge representation between symbolic and subsymbolic approaches. Concepts are represented as regions in a space defined by quality dimensions, allowing similarity to be calculated based on distance. While pioneering, the theory lacks empirical validation and formalization.

Uploaded by

Peter Rulon-Miller
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
96 views

Geometry of Thought (Abstract)

The document summarizes Peter Gärdenfors' theory of conceptual spaces, which proposes a geometric framework for representing conceptual knowledge. It introduces conceptual spaces as a level of knowledge representation between symbolic and subsymbolic approaches. Concepts are represented as regions in a space defined by quality dimensions, allowing similarity to be calculated based on distance. While pioneering, the theory lacks empirical validation and formalization.

Uploaded by

Peter Rulon-Miller
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

Conceptual Spaces-The Geometry of Thought

Peter Gärdenfors, A Bradford Book, 2000


Angela Schwering
In his book “Conceptual Spaces - The Geometry of Thought”, fors describes his theory only at a very abstract level and for-
Peter Gärdenfors [1] presents a pioneering theory for represent- bears from describing algorithms for the formalization of his
ing conceptual knowledge, the basic construct of human thinking theory. The realization of a computational model for concep-
and reasoning [4]. The conceptual level is not seen as an alter- tual spaces bears many practical problems which still have to be
native to traditional approaches of knowledge representation in solved. Moreover, no empirical evidence is given for his pioneer-
artificial intelligence, namely symbolic or subsymbolic methods. ing, sometimes revolutionary ideas. However, these shortcom-
Instead, it is meant to complement both approaches. The book ings should be considered as challenges to solve in the future.
is highly recommendable and worth reading, as it does not only The target audience of the book is highly interdisciplinary:
tackle many fundamental problems of knowledge representation since Gärdenfors tackles the problem of cognitive knowledge rep-
such as grounding [3], concept formation and similarity compar- resentation from a psychologic and computer science perspective
isons [2], but also outlines novel and enlightening ideas how to as well as from a philosophic, neuroscience and linguistic point
overcome these. of view, this book is worth reading for researchers from many
The book introduces the notion of a conceptual space as a different areas. It is required reading for researchers in cognitive
framework for representing knowledge at the conceptual level. science or artificial intelligence interested in knowledge repre-
It is motivated by contrasting it to other levels of knowledge sentation. The book has a clear structure and is very well writ-
representation: The highest level, the symbolic level, concep- ten. The convincing examples throughout the book illustrate the
tualizes the human brain as a computing device. Knowledge is findings very well and make it easy to understand. Therefore I
represented based on a language consisting of a set of symbols. would also deem Gärdenfors’ book to be suitable for students
Logical operations can be performed on these symbols to in- as introducing literature to various problem fields in cognitive
fer new knowledge. Human reasoning is modeled as a symbol science. It gives readers from related areas the chance to look
manipulation process. Classical, symbolic artificial intelligence beyond one’s own nose and get to know an interdisciplinary way
does not very well support central cognitive processes such as of thinking. The book certainly meets the expectations of the
the acquisition or formation of new concepts and similarity com- highly interdisciplinary research area cognitive science.
parisons. The lowest level, the subsymbolic knowledge represen-
tation, is oriented towards the neuro-biological structure of the
human brain. Concepts are implicitly represented via activa- References
tion patterns within the neural network. Learning is modeled by [1] Gärdenfors, P.: The Geometry of Thought. A Bradford Book,
modifying the activation of neurons. Explicit representation of 2000.
knowledge and concepts is not possible.
[2] Hahn, U.; Ramscar, M.: Similarity and categorization. Oxford
At the intersection between the symbolic and the subsym- University Press, 2001.
bolic level, Gärdenfors introduces the conceptual level. The the-
[3] Harnard, S.: The Symbol Grounding Problem. Physica D: Non-
ory of conceptual spaces is based on semantic spaces with a
linear Phenomena, 42:335-346. 1990.
geometric structure: A conceptual space is formed by a set of
quality dimensions. One or several quality dimensions model [4] Holyoak, K.; Morrison, R.: The Cambridge Handbook of Think-
ing and Reasoning. Cambridge University Press, 2005.
one domain. An important example used throughout the book
is the color domain represented by the quality dimensions hue,
saturation and brightness. Conceptual spaces have a cognitive Contact
foundation because domains can be grounded in qualities per-
Dr. Angela Schwering
ceivable by the human sensory apparatus. Concepts are repre-
Institute of Cognitive Science
sented as conceptual regions described by their properties on
University of Osnabrueck
the quality dimensions. The geometric structure of conceptual
Albrechtstr. 28, 49076 Osnabrueck
spaces makes it possible to determine distances and therefore
Email: [email protected]
provides an inherent similarity measure by taking the distance in
the conceptual space as indicator of the semantic similarity. The
notion of similarity is an important construct for modeling cate-
gorization and concept formation. Using similarity for reasoning
can also reflect well the vagueness typical for human reasoning.
The strong focus on the cognitive foundation makes the
book particularly valuable. It contains many challenging claims
which are related to various disciplines by giving evidence from
a wide range of literature. This shows the huge and highly inter-
disciplinary background of the author. Unfortunately, Gärden-

Page 1

You might also like