Similarity and Categorization: A Review

researchers are adopting richer and more well-developed similarity-based approaches. These second-wave similarity-based approaches address some of the shortcomings of previous approaches. The 13 chapters that form this volume are broad in scope but can be characterized as couching categorization in terms of more sophisticated and precise notions of similarity. This work shows promise in elucidatWhat is the nature of human categories? How do we form categories? What is the role of similarity in categorization? Can we formalize the answers to these questions to derive further insights and develop useful software systems? These were the questions addressed at an interdisciplinary meeting attended by psychologists, computer scientists, anthropologists, statisticians, and philosophers held at the University of Edinburgh. The edited volume Similarity and Categorization arises from this meeting. The publication of Similarity and Categorization is timely because the study of categorization is at a theoretical crossroads. In the 1970s, similarity was thought to be the basis of categorization—an object was assigned to the category to which it was most similar. In the 1980s, a new view emerged that held that similarity was too weak and vague a construct to ground human categorization. On this view, our categories cohere by virtue of being embedded in explanatory systems (that is, our knowledge of the world is organized around theories akin to scientific theories). According to the theory-based view, judgments of similarity are largely governed by theories that determine the relevant properties for evaluation. Recently, dissatisfaction has emerged with the theorybased account of categorization, and gories we have are natural partitions of the world. This position is supported by data collected by Eleanor Rosch and her colleagues in the 1970s. In other words, we have the categories we do because they preserve existing similarities among objects and are therefore informative. For example, our categories tend to be inductively powerful (for example, If I know an object is a bird, I can assume it has feathers and probably flies). Critics of this view contend that the notion of similarity is vague and ill defined and carries no explanatory force on its own. As the philosopher Nelson Goodman (1972) noted, objects can be similar in an unlimited number of ways (for example, weigh an odd number of grams, are less than 100 meters long, are less than 101 meters long). According to Goodman, one must specify in what respect two objects are similar, or the statement is empty. If two objects are similar only because they are in the same category, then similarity is a vacuous notion, and any account of categorization based on similarity is circular. A second line of attack on similarity-based accounts is that similarity is not a powerful enough construct to account for human categorization. For example, a man who jumps into the swimming pool at a party while he is wearing a tie and sports coat is classified as drunk not because he is similar (at least in a straightforward sense) to other examples of drunk people. Rather, he is classified as drunk because this behavior is in accord with our theories of the kinds of things that drunk people do. This theory-based view of categorization has been popularized by Murphy and Medin (1985) and is reflected in AI techniques such as explanation-based learning (DeJong and Mooney 1986). Although the theory-based view of categories does touch on some shortcomings of the earlier similarity-based approaches, the theory-based view has itself proven vague and has not made large headway in understanding human categorization. In contrast, the contributors to this volume have made headway by considering richer accounts of similarity. For example, Arthur Markman coning the nature of human categorization. The editors, Ulrike Hahn and Michael Ramscar, do an excellent job in the introductory and concluding chapters of explicating the historical shifts in the perceived relation between similarity and categorization. I briefly overview these shifts and how the contributed chapters fit into the latest movement. The idea that similarity is the basis for categorization is intuitive given that similar objects tend to be in the same category. On this view, the cateSimilarity and Categorization, Ulrike Hahn and Michael Ramscar, editors, Oxford University Press, New York, 279 pp., 2001, ISBN 019-850628-7.