Exemplar-Based Approach to Relating Categorization, Identification, and Recognition
Robert M. Nosofsky Indiana University
This chapter provides an overview of an exemplar-based approach to relating performance in tasks of multidimensional perceptual categorization, identification, and recognition. By categorization, I mean a choice experiment in which people classify distinct items into groups, whereas in identification each item is assigned a unique response. Recognition refers to a memory paradigm in which people judge whether items are "old" (previously experienced) or "new."
At the heart of the present approach is the assumption that people represent categories by storing individual exemplars in memory, and make classification decisions on the basis of similarity comparisons with the stored exemplars. This exemplar view of category representation strongly motivates the study of relations among categorization, identification, and recognition. Presumably, when people learn to identify stimuli, a unique representation of each item is stored in memory. Furthermore, the extent to which individual items are confused is determined by similarity relations among the items (e.g., Lockhead, 1970; Luce, 1963a; Shepard, 1958b). If categorization decisions are also based on similarity relations among individually stored items, then one might be able to predict categorization performance given knowledge of performance in an identification paradigm involving the same set of stimuli. Likewise, if individual exemplars are stored in memory during category learning, then this fact should be corroborated by postacquisition recognition memory tests.
A conceptual difficulty associated with using an exemplar-similarity model to relate categorization, identification, and recognition, however, is that similarities among exemplars may not be invariant across the paradigms. The present theory adopts a Multidimensional Scaling (MDS) approach to representing context- dependent changes in similarity. Exemplars are represented as points in a multi-