Categorizing Fragments of Exemplars: Experimental and Computational Results

Analysis of the relative categorization accuracy of whole and partial items may provide deeper insight into the nature of cat-egorization processes. I compared several logical, mathemati-cal and computational models of categorization (including the FLMP, KRES, and variants of the GCM), and ran an experi-ment to get robust empirical data. The experiment showed a mixture of categorization strategies used by different subjects. When known rule-based strategies were excluded, the remain-ing subjects responded more accurately than theoretically op-timal to single feature stimuli. This result is inconsistent with standard exemplar and connectionist models, and with theoret-ical models of perceptual integration, but is consistent with an interactive model of categorization that allows top-down feed-back to affect representations of partial stimuli