Preferential Encoding of Features Distinctive for Multiple Categories

Understanding how features are encoded during category acq uisition is a fundamental challenge in the study of human learning. The current work pr oposes that in order to maintain accurate generalization in large scale categorization sys tems, features that are useful in discriminating multiple categories must be actively prefe red. This multi-class hypothesis stands in contrast with theories in which features are encoded for e ach categorization task individually, as well as theories that focus solely on encoding coincidenc es in input patterns. The current paper provides a methodology for empirically testing the propose d hypothesis under controlled conditions. It is shown that in the process of acquiring a seq uence of new categories, features that are informative for recognizing several categories are preferably encoded. Moreover, evidence is provided that after acquiring these features, representat io s of categories learned in the past are actively reconstructed by the newly encoded features. Fina lly, it is demonstrated that encoded features play a role in facilitating future category acquis ition. These results are observed using both perceptual and semantic stimuli. It is suggested that p referring features that provide maximum information on multiple categories is a general cha racteristic of the human categorization systems.