Stimulus generalization is a fundamental component process of any cognitive task, because stimuli and contexts are never precisely repeated. While generalization has been well studied in conditioning paradigms, and has been proposed to depend directly on perceptual similarity, the phenomenon is less well understood in tasks of category learning due to the inherent difficulty of empirically disentangling the effects of multiple prior stimuli. Nevertheless, nearly every extant model of human category learning assumes that learning generalizes across stimuli, according to the similarity of their representations. The present project will refine and make use of a novel method for directly measuring stimulus generalization in category learning, based on isolating the marginal learning effect of the previous trial upon the current response. This method will be applied at different stages of learning in order to discern how patterns of generalization change in reponse to a (probabilistic) category structure. A theory of perceptual reorganization and adaptation of generalization is proposed which predicts that strength of generalization will initially depend on physical similarity among stimuli but after sufficient learning will depend more on prediction of common outcomes. Such a change would indicate reorganization of internal perceptual representations to reflect abstract (vs. superficial) commonalities among stimuli. In the long run it is expected that this process of second-order learning will be seen to underlie concept formation, categorical perception, and perceptual grounding of symbolic processing, thus providing a crucial link between low- and high-level cognition. ? ?
| Sakamoto, Yasuaki; Jones, Mattr; Love, Bradley C (2008) Putting the psychology back into psychological models: mechanistic versus rational approaches. Mem Cognit 36:1057-65 |
| Jones, Matt; Love, Bradley C (2007) Beyond common features: the role of roles in determining similarity. Cogn Psychol 55:196-231 |
