Better learning of partially diagnostic features leads to less unidimensionalcategorization in supervised category learning
Abstract
Previous studies of supervised category learning show that par-ticipants often prefer a unidimensional categorization strat-egy. Studies also report that the perfectly diagnostic featureis learned better compared to the partially diagnostic features.We replicate these results, and we show that better learning ofpartially diagnostic features leads to less preference for uni-dimensional categorization. When participants have perfectknowledge about all the diagnostic features, then it becomesequivalent to memorizing the prototypes of the categories. Wecompare our results with the match-to-standards procedure,where category prototypes are shown during categorizationand unidimensional strategy is seldom preferred. We interpretour results to suggest that the preference for unidimensionalcategorization in supervised category learning, shown in ear-lier studies, could be due to poor learning of the partially diag-nostic features.