A rule - plus - exception model for classifying objects in continuous - dimension spaces

The idea that people may represent categories in terms of simple logical rules dates back to the very beginnings of research on concept identification in cognitive psy-The rule hypothesis carries a good deal of intuitive appeal. An important purpose of categorization is to reduce the complexity of mental processing by organizing distinct objects into classes and then dealing with the classes as wholes rather than with each object uniquely. By forming a simple rule, an economical summary description is provided for an entire class of objects, thereby allowing for a vast reduction in the amount of information that one needs to store in memory. Furthermore, to decide category membership for any individual object, one need only decide whether or not the combination of attributes that composes the object satisfies the rule. Despite its intuitive appeal and the early dominance of this approach, models based on the formation of simple logical rules had, until recently, largely dropped from the scene in categorization research. Historically, the main impetus for this trend can be traced to the highly influential work of such researchers as Posner and Keele (1968) for example, argued convincingly that most categories in the natural world were not struc-tured according to simple rules. For one thing, it is difficult to state simple rules or definitions that perfectly partition the members of most natural-world categories. Furthermore , experimental research indicates that categories have a graded, internal structure, in which some objects are " better " or more typical members of the category than Models based solely on the idea that simple rules or definitions are used to represent categories seem unable to account for the full range of findings involving these typicality and graded-structure effects (for an extensive review and analysis, see E. E. Smith & Medin, 1982, chap. 3). In response to these challenges, experimental research began to examine the learning of ill-defined category structures in which no simple rules existed for deciding category membership. A wide variety of models have been developed to account for the learning of such ill-defined categories. According to ex-emplar models, for instance, people represent categories by storing previously experienced category exemplars in memory and classifying objects on the basis of their similarity to these exemplars. Such models, which have enjoyed enormous success at accounting for diverse cate-345 The authors propose a rule-plus-exception (RULEX) model for how observers classify stimuli residing in continuous-dimension spaces. The …