Symbolic, conceptual and subconceptual representations

Peter Gärdenfors

Printed as pp. 255-270 in Human and Machine Perception: Information Fusion, ed. by V. Cantoni, V. di Gesť, A. Setti and D. Tegolo, Plenum Press, New York.

Abstract

Within logic, AI and early cognitive science, almost all research concerning knowledge representation has been based on symbolic (propositional) models. In the 80's connectionist models became popular as an alternative way of modelling knowledge.

I propose that also a third type of knowledge representation based on conceptual models should be studied. In particular, I will discuss conceptual models based on geometrical and topological notions and show their relevance for concept revision and for non-monotonic aspects of concepts. I will argue that concpetual models of knowledge can be seen as a bridge between symbolic and connectionist representations. The three kinds of knowledge representation models should thus not be seen as competitors, but rather as complementary models suitable for different modelling tasks.


The paper is available in PostScript format (get) or in rtf (get).


To LUCS Home Page