CATEGORIZATION USING CHAINS OF EXAMPLES

People can infer unknown features of a stimulus by retrieving memories of similar examples. It is proposed that we can reason from chains of examples. For example, stimulus A may remind us of B, which reminds us of C. Information about C may then affect reasoning about A. A mathematical model for categorization (extended from the context model of Medin & Schaffer, 1978), using multiple-step chains of reasoning and memory for examples, is presented. In five experiments, subjects memorized feature descriptions of fictional people, then made predictions from incomplete descriptions. Various predictions could be made using one-, two-, or three-step chains of reasoning. These experiments varied in terms of stimulus structure, complexity of test questions, and response method (probability estimate or forced choice). The multiple-step context model, with the assumption that people performed one- and two-step chains of inference, successfully accounted for the results of all five experiments. © 1992.