Journal of Experimental Psychology: Learning, Memory, and Cognition 1988, Vol. 14, No. 4, 700-708 Copyright 1988 by the American Psychological Association, Inc. 0278-7393/8S/$00.75 Exemplar-Based Accounts of Relations Between Classification, Recognition, and Typicality Robert M. Nosofsky Indiana University Previously published sets of classification and old-new recognition memory data are reanalyzed within the framework of an exemplar-based generalization model. The key assumption in the model is that, whereas classification decisions are based on the similarity of a probe to exemplars of a target category relative to exemplars of contrast categories, recognition decisions are based on overall summed similarity of a probe to all exemplars. The summed-similarity decision rule is shown to be consistent with a wide variety of recognition memory data obtained in classification learning situations and may provide a unified approach to understanding relations between categorization and recognition. Recently, there has been an upsurge of interest among categorization researchers in exploring relations between clas- sification learning and old-new recognition memory. This interest has been fueled by the exemplar view of category representation, which holds that people base classification decisions on similarity comparisons with stored exemplars (Hintzman, 1986b; Medin & Schaffer, 1978; Nosofsky, 1986). Recognition data provide a source of converging evidence bearing on the nature of people's category representations. Presumably, if individual exemplars are being stored in mem- ory, the fact ought to be revealed by postacquisition recogni- tion tests. Indeed, a number of researchers have taken exemplar models to task on grounds of certain dissociations between classification learning and recognition memory, or patterns of recognition data deemed to be inconsistent with the pre- dictions of exemplar-only memory models. In virtually all cases, however, there has been a failure to specify and test an explicit decision rule by which exemplar memories are used to make recognition judgments. A natural decision rule is the one embodied in the memory models of Gillund and Shiffrm (1984) and Hintzman (1986a), namely, that recognition judgments are based on the summed similarity (or activation) of a probe to all stored items. This summed similarity gives a measure of overall familiarity, with higher familiarity values leading to higher recognition proba- bilities. Medin (1986) recently considered the implications of a summed-similarity decision rule and suggested that it was at least roughly consistent with a set of classification/recog- nition data collected by Estes (1986b). The purpose of the present article is to follow up on Medin's suggestion and to The work reported in this article was supported by Grant BNS 85- 19573 from the National Science Foundation to Indiana University. I would like to thank Steven dark, Douglas Medin, Janet Metcalfe, and Richard Shiffrin for many helpful discussions and three anony- mous reviewers for their criticisms and suggestions regarding an earlier version of this article. Correspondence concerning this article should be addressed to Robert M. Nosofsky, Department of Psychology, Indiana University, Bloomington, Indiana 47405. illustrate more formally, by way of application to other ex- amples in the literature, that a summed-similarity decision rule within the framework of an exemplar storage model may account well for recognition data obtained in classification learning situations and may help explain relations between classification learning and recognition memory. The use of a summed-similarity rule for interpreting typicality judgments is also explored. General Modeling Approach The analyses of the categorization and recognition data are conducted within the framework of the context model of classification proposed by Medin and Schaffer (1978) and generalized for application to continuous integral- and sepa- rable-dimension stimuli by Nosofsky (1986, 1987). According to the context model, the probability that Stimulus i is clas- sified in Category J, P(RjlSi), is found by summing the similarity of Stimulus i to all Exemplars j belonging to Cate- gory J and then dividing by the summed similarity of Stimulus i to all exemplars of all categories, 2 ( E ^kSik) ' K k«CK (1) where N f represents the relative frequency with which Exem- plar j was presented during training and where s^ represents the similarity between Exemplars i and j (Estes, 1986a; No- sofsky, 1988). Recognition judgments are assumed to be based on the overall familiarity of a stimulus, F» measured by summing the similarity of the stimulus to all exemplars of all categories, F i = (2) (Presumably, the subject sets some criterion c such that values ofFi greater than c lead to old responses.) Note that, whereas classification is assumed to be related to relative degree of target-category to contrast-category similarity (Equation 1), recognition is assumed to be related to overall summed sim- 700