Learning correlations in categorization tasks using large, ill-defined categories

The experiments revealed whether individual participants are sensitive to exemplar information in the form of within-category correlations between stimulus dimensions after training on large overlapping categories. Participants were trained in 1 of 2 categorization conditions. The sign of the correlation between dimensions differed across conditions, but the categorization rules that best separated the categories were identical. An unannounced attribute-prediction task followed categorization training. Several participants produced predictions consistent with the correct correlation between the dimensions. For other participants, the predictions reflected the correlation only within the region they had associated with the given category, even though the categories overlapped, suggesting that the decision boundary was explicitly represented in memory. Finally, for other participants, no correlational information appeared to be accessible for the prediction task.