Research Group
in Analytic Philosophy

Abstract

Theories of concepts in psychology have traditionally been divided in two groups: prototype and exemplar theories (Rosch 1999; Hampton 1995; Medin & Schaffer 1978), where concepts are identified with typical members of the class the concept is about, or some other body of statistical knowledge about this class; and the various theory theories (???; Murphy & Medin 1985) according to which concepts are theories about the class they target. Choosing the correct theory of content among these two families has proven hard: each of them is better suited to explaining different aspects of our competence in the use of concepts, and does a better job at explaining different experimental results. This has lead some theorists to defend that concepts are composites of prototypes and/or exemplars, and theories (Osherson & Smith 1981; Nosofsky, Palmeri & McKinley 1994), or even that there are no such things as concepts (Machery 2009).

In this paper I present an information-theoretic account of concepts out of which these two perspectives on concepts fall out naturally, as part of the same, simple theoretical construct. The main idea is that concepts do lossy compression of the state of the world: they say a lot about the world that is relevant to the thinking subject, in an economic way. I propose that we see concepts as signals that attempt to solve a coding problem: given the available rate in a certain channel (say, from perception to higher cognition), how should signals (concepts, in the case that interests us) encode the world so as to minimize errors in the picture that the thinking subject forms of it?