EVIDENCE FOR RELATIONAL-CODING IN CONCEPT FORMATION,AND A NOTE ON THE EFFECTS OF FEEDBACK*

The purpose of the present paper was to determine whether people encode relations among component features in experience, or whether they simply encode the occurrence of features separately as independent cues. This has become a major issue in theories of concept-formation at the present time. It is suggested in this paper that two problems have particularly obstructed a clear resolution of the issue. The first is a consistent failure to check to what extent training materials actually contain relations among features, before results are declared in favour of a particular model. The second is the use, hitherto, of only informal methods of analysing such relations, where they exist, and of predicting from them the structure of the concept. This paper describes an expression for measuring the overall relations among feature-variables in concept training materials. It also introduces a methodology, using log-linear models, for predicting concept structures on the basis of those relations. An experiment combined these methods to assess whether subjects abstracted relations among features, or whether they abstracted independent cues. A subsidiary aim was to assess how feedback, based on independent-cue information, during learning, affected the construction of the concept. The results came out strongly in favour of relational-coding and there was some evidence that feedback based on independent-cues retarded concept formation.