Search

Independent-cue models of concepts which have been popular over the last few years, such as feature-frequency or prototype models, have not been heuristic about conceptual functions beyond simple categorization and typicality effects. Yet many authors are now pointing to the flexible and adaptive quality of conceptual representations. Alternative models of concepts which stress the encoding of relational information among feature-variables might be able to accommodate such qualities, but attempts to develop such models have been hindered by inadequate means of describing and measuring relational information. The two experiments reported here utilized log-linear statistical models to manipulate and monitor the relational information in experienced exemplars of a category. These then made predictions about ‘missing parts’ old and new exemplars of the category which contrasted with some of the predictions of independent-cue models. In delayed tests in which subjects had to identify the missing parts from partial figures of exemplars, responses appeared to be based on relational information rather than on information about separate feature-variables (independent-cues). The implications for concept models and concept functions are discussed.