Proceedings of the Annual Meeting of the Cognitive Science Society

Research on semi-supervised category learning has beensparse despite its representativeness of naturalistic categorylearning and potential applications. Most of the semi-supervised literature to date has focused on establishing thephenomenon. These efforts have produced mixed results andhave explored a relatively limited set of learningcircumstances. In the current work, we contribute a novelinvestigation of semi-supervised learning by extending theparadigm to relational category learning and evaluating therole that item similarity plays in the effectiveness ofunsupervised learning opportunities. Our results show first-ever evidence of semi-supervised learning in the induction ofrelational categories and, further, that the similarity betweensupervised and unsupervised examples critically dictateswhether benefits of unsupervised exposures accrue. Weconclude with implications and future directions.