Similarity-based and knowledge-based processes in category learning

Adult subjects were asked to rate a number of figures according to their perceived usefulness in a real-world task. An independent group of subjects then learned to classify these figures into one of two categories. One group of subjects were provided with category labels designed to facilitate the use of their pre-existing knowledge in the learning task and a second group were given nonsense labels. Predictions derived from similarity-based (prototype and exemplar) models of categorisation were compared with those based upon the rated usefulness of the exemplars. The contribution of preexisting knowledge to accurate responding and to typicality judgements was greater with meaningful rather than nonsense category labels. An examination of test phase performance in the two groups indicated that the effect of providing a meaningful theme extended beyond a redistribution of feature weights during encoding. These results were replicated and extended in a second study in which the categories to be learned were fully ill-defined. The two studies indicate that both prototype similarity and pre-existing knowledge may influence the encoding and retrieval of conceptual information. The results also suggest that when there is a high degree of correspondence between a subject's prior knowledge and the sorts of attributes that are present in category exemplars, the influence of prior knowledge extends beyond the dimensioning and weighting of feature values.