Studies of similarity

Any event in the history of the organism is, in a sense, unique. Consequently, recognition, learning, and judgment presuppose an ability to categorize stimuli and classify situations by similarity . As Quine (1969) puts it: "There is nothing more basic to thought and language than our sense of similarity ; our sorting of things into kinds [p . 1161 ." Indeed, the notion of similarity that appears under such different names as proximity, resemblance, communality, representativeness, and psychological distance is fundamental to theories of perception, learning, and judgment . This chapter outlines a new theoretical analysis of similarity and investigates some of its empirical consequences . The theoretical analysis of similarity relations has been dominated by geometric models. Such models represent each object as a point in some coordinate space so that the metric distances between the points reflect the observed similarities between the respective objects . In general, the space is assumed to be Euclidean, and the purpose of the analysis is to embed the objects in a space of minimum dimensionality on the basis of the observed similarities, see Shepard (1974) . In a recent paper (Tversky, 1977), the first author challenged the dimensionalmetric assumptions that underlie the geometric approach to similarity and developed an alternative feature-theoretical approach to the analysis of similarity relations. In this approach, each object a is characterized by a set of features, denoted A, and the observed similarity of a to b, denoted s(a, b), is expressed as a function of their common and distinctive features (see Fig . 4.1) . That is, the observed similarity s(a, b) is expressed as a function of three arguments : A f1B, the features shared by a and b ;A B, the features of a that are not shared by b ; B A, the features of b that are not shared by a . Thus the similarity between