The role of attention shifts in the categorization of continuous dimensioned stimuli

Results of human category learning experiments, using stimulus dimensions with binary values, have implicated a rapidly acting mechanism of attention shifts. Theories of categorization desire that stimuli with binary, discrete and continuous valued dimensions should all be treated similarly. Theoretical analyses of attention shifting, however, have up to now only been developed for shifts between features, or shifts between entire dimensions, not shifts within dimensions. Here we present a model of how people learn to discriminate categories made up of stimuli with continuous-valued dimensions. The model uses rapid shifts in attention within stimulus dimensions to reduce errors during learning; the model generalizes J. K. Kruschke's (Psychological Review, 99, 22-44, 1992) ADIT model. In an experiment in category learning, subjects were trained to discriminate four bivariate normal distributions that are presented with differential base rates. The base-rate manipulation produces several qualitative effects, for which the model accounts very well. With attention shifting turned off, the model fails to account for some aspects of the data, suggesting that attentions shifts are an important mechanism in the model.