Figure 8. Rule Plus Exception Results.

Note. (A) Human Behavior: Recognition (“Rec.”) and categorization (“Cat.”) accuracy for rule-following (“Rule”) versus exception (“Ex.”) stimuli (Davis, Love, & Preston, 2012a). Although categorization accuracy was greater for rule-following items, recognition accuracy was greater for exception items. Error bars reflect 95% confidence intervals. (B) Model Behavior. Mirroring human behavior, the model displayed greater categorization accuracy for rule-following items than for exception items, but greater recognition strength (Equation 13) for the exception items than the rule-following items. (Note differences in the vertical-axis scale for categorization accuracy and recognition strength). (C) Model Categorization Behavior: Categorization accuracy for rule and exception items for the standard and yoked (“Y”) models. Mirroring human behavior, both models displayed better categorization accuracy for rule-following items than for the exception items. Accuracy for the yoked model was lower than that of the standard model. Model-type and stimulus-type interacted such that the yoked model displayed a greater difference in accuracy between rule-following and exception stimuli than the standard model. (D) Model Sampling Behavior. Although the standard model sampled a greater number of features for exception items than for rule-following items, the yoked model did not.