Abstract

Many theories of conceptual organization assume the existence of some form of mental similarity metric In the domain of categorization, such theories have been called "similarity-based" . Criticism of similarity-based theories has led to a call for "theory-based" models of categorization . Theory-based views remain somewhat vague, however. ;I argue that many similarity-based models of categorization fail for three reasons. First, they posit an overly stable and unitary representation of categories. Second, they treat local constraints on behavior as general principles of categorization. Finally, they rely too heavily on intuitively appealing featural decompositions as a de facto mental language of atomic properties. ;I provide an empirical example the local character of one constraint on conceptual organization by examining the effects of linear separability in classification learning. ;I describe a schema-based theory of conceptual organization and propose a computational model that embodies the principles of the theory. The model depends on the notion of a mental similarity metric but makes use of connectionist learning principles to develop a representation of concepts that avoids problems faced by some similarity-based models of categorization. I discuss the relationship of this theory to similarity-based and theory-based accounts