Hampton, James. 1993. "Prototype Models of Concept Representation," in Categories and Concepts: Theoretical Views and Inductive Data Analysis, editors Ivan Van Mechelen, James Hampton, Ryszard S. Michalski, and Peter Theuns, 67-95. Cognitive Science Series, General Editors. Barry Richards, and Keith Stenning, London, UK: Academic Press.
Five main sections deal with evidence leading to the prototype view, a description of the "standard" prototype model, connections to some relevant data-analytic methods, advantages of prototype models, and inadequacies of the prototype view. Conclusion identifies theory-based view and exemplar as chief rivals to prototype view, advocates for integration of all of these. Sees prototype theory as fitting between the abstract models of categorization of theories and frames, and the concrete approach to exemplars.

Includes in the standard prototype model the differential weighting of attribute values and of attributes themselves, as well as a similarity scale, a criterion (cutoff for category membership), and typicality, which author chooses to apply only within a category. Maintains that "Degree of category membership (as defined by... probability of a positive classification) and rated typicality are both derived directly from the underlying similarity scale, but are not identical - a frequent source of misunderstanding" (75). Acknowledges both within-subject instability (71) and between-subject variability (76). Notes that much of Rosch's conjectures about how categories are formed can be modeled using data-analytic methods, such as Latent Class Analysis. Advantages of prototype view are that it explains a wide variety of phenomena reported in linguistics (Lakoff 1987), anthropology, and psychiatry; also that it is a flexible model, capable of incorporating new information as it comes in and changing over time. "As we then learn more about which attributes are most central (in terms of their correlation with all the others), we can readjust the attribute-value weights, and introduce new attributes based on more subtle aspects of the objects in the domain, in a gradual way" (80). MOST HELPFUL AS A MODEL OF MUSICAL GENRE-LEARNING!

Three major objections to the prototype model are the problem of complex concepts ("pet fish"), over-generality, and over-emphasis on abstract types over the role of exemplars in categorization. Author's own research has shown that the complex concept problem can be partially solved by viewing concept combination at the level of intension rather than extension, and a willingness to drop some attributes from the union of the two attribute sets. Notes that there are still things left to explain, like how new attributes arise in combination that don't exist in either of the parent categories, e.g. pet birds live in cages.

The charge that prototype theory explains too much must be answered by "an independent account of how attributes are selected and how attribute-value weights are determined," the idea being to reduce eliminate a large portion of "conceptual structures, most of which never occur naturally" (85). Outlines two types of "independent accounts," the first a "bottom-up" account based on Gibson's theories of perception and on Rosch's thoughts on how basic-level categories arise. This passage (p. 86) is the clearest explanation of basic level yet, and contains more gist for A MODEL OF MUSICAL GENRE-LEARNING! The other type of account is "top-down" (frames and theories), and also describes mechanisms for genre-learning, as classification of instances gives way to reasoning based on theories that account of properties of things not directly observable: "The search for a stable conceptual system, in which useful generalizations can be made, leads to concepts based on more stable abstract properties" (87).

The challenge of the "exemplar-based models" is that prototype effects can be explained not by comparison to a central tendency, but by similarity to certain individual instances. Evidence comes from artificial classification experiments and from connectionist research. Treating a small group of exemplars as central is a tenable compromise solution. Similarity calculation to an exemplar is problematic without the analysis of attributes that lead back to the idea of a central tendency.
P128.C37.C38 1993 Ed-P.