Proceedings of the Annual Meeting of the Cognitive Science Society

Theoretical and empirical work in the field of classificationlearning is centered on a ‘reference point’ view, where learn-ers are thought to represent categories in terms of stored pointsin psychological space (e.g., prototypes, exemplars, clusters).Reference point representations fully specify how regions ofpsychological space are associated with class labels, but theydo not contain information about how features relate to oneanother (within- class or otherwise). We present a novel exper-iment suggesting human learners acquire knowledge of within-class feature correlations and use this knowledge during gen-eralization. Our methods conform strictly to the traditional ar-tificial classification learning paradigm, and our results can-not be explained by any prominent reference point model (i.e.,GCM, ALCOVE). An alternative to the reference point frame-work (DIVA) provides a strong account of the observed perfor-mance. We additionally describe preliminary work on a noveldiscriminative clustering model that also explains our results.