A Computational Account of Basic Level and Typicality Effects

Cognitive psychology has uncovered two effects that have altered traditional views of human classification. Basic level effects suggest that humans prefer concepts at a particular level of generality, while typicality effects indicate that some instances of a class are more readily recognized as such than others. This paper describes a model of memory that accounts for basic level effects, typicality effects, and interactions between them. More generally, computer experiments lay the groundwork for a formal unification of basic level and typicality phenomena. 2. Concept Models and Hierarchies Many psychological and AI studies have assumed that concepts are logical summaries of features that are common t*o concept members; instances are classified by insuring a ‘perfect’ match between a concept and the instance. This classical view [Smit81] implicitly treats each instance as ‘equal’, but typicality effects suggest that humans do not treat instances equally. In response, a number of representations have been proposed that reflect the variable importance of concept features [Rosc75]. In particular, probubilistic representations [Smit81] associate a probability or weight with each concept feature.