ISSN : 1226-9654
This study was planned to survey assumptions and problems of existing categorization models and to test the relative plausibility of an hybrid categorization approach based on mixed representation. Specially, we explored whether we could regard empirically defined prototypes as proper summary representations of a category and two-way interaction between exemplar frequencies and learning trials as a reasonable evidence for an hybrid categorization model. In experiment 1 and experiment 2, we distorted exemplars from category protptype and investigated change patterns of category judgment, accuracy with category size. Results showed that category learning strategies weren't fixed but adjusted by learning conditions ceaselessly, In experiment 3, we distorted exemplars from category prototypes and examined an existing hybrid model's assumption that, categorization depends on exemplar information primarily and summary information later, In consequence, we found same results as Homa, Dunbar, & Nohre (1991)'s study in the highly distorted category but couldn't find out two-way interaction between exemplar frequencies and leaning trials in low distorted category In conclusion, these result implicated that categorization processes are not prototype abstraction processes nor exemplar retrieval processes but dynamic learning processes optimizing performance according to learning conditions.