Proceedings of the Annual Meeting of the Cognitive Science Society

Concept learning can be studied through training that involves either classification or observation. In classification, learners actively make a judgment about category membership of presented stimuli and receive feedback; in observation, learners study a labeled stimulus. Extant work shows contradictory findings, as some studies show a benefit of classification over observation, whereas others show the opposite pattern. Recent work posits that these findings might be explained by category confusability. We present an experiment that investigates this question by directly manipulating category confusability (within-subjects), wherein subjects either learned through classification or observation. All subjects learned three different pairs of natural categories that shared low, moderate, or high feature similarity. Subjects were presented a single stimulus on each trial; every fourth trial, they made an endorsement judgment. Our findings suggest that training both with (classification) and without (observation) error are equally beneficial for learning natural categories, challenging several prominent category learning models.