UC Santa Barbara

While there are high-performing measures of categorization difficulty for RB categorization tasks, in which the optimal strategy is a simple explicit rule, predicting human performance in II categorization tasks, in which category membership is determined by similarity, has been historically difficult. This dissertation proposes a novel biologically motivated difficulty measure that can be generalized across stimulus types and category structures: the Striatal Difficulty Measure (SDM). The SDM is compared to 12 previously proposed measures on an extensive data set that includes conditions with continuous- and binary-valued stimulus dimensions, a variety of different stimulus types, and linearly- and nonlinearly-separable categories. Across this diverse dataset, the SDM was the most successful at predicting the numerical values of the mean error rates in each condition as well as predicting the observed rank ordering of conditions by difficulty.

This dissertation also investigates the effects of three different factors on category learning difficulty: linear separability, variability on stimulus dimensions that are irrelevant to the categorization decision, and instruction on the optimal categorization strategy. The results clarify a long-standing uncertainty in the field by showing that linear separability plays no role in II learning difficulty. They also establish two novel dissociations between RB and II category learning: variability on irrelevant stimulus dimensions impacts II learning but not RB learning while instructions on the optimal strategy impact RB learning but not II learning. Finally, the theoretical and neurobiological implications of these results are discussed.