Memory Versus Generalization During Category Learning: An Extension of the Retrieving Effectively from Memory Model

Abstract This paper introduces an extension of the Retrieving Effectively from Memory (REM) model that supports both knowledge generalization and category learning. Simulation results demonstrate that core memory principles from REM can be applied beyond recognition, providing a foundation for concept acquisition and generalization. These findings offer theoretical support for the idea that recognition memory and knowledge generalization share underlying mechanisms and suggest that the former may facilitate the latter. We then use this extended model to examine inconsistent findings regarding the learning benefits of classification and observation in category learning. The model formalizes differences between learning modes in terms of encoding and feedback processes, including whether exemplars are stored once or twice (single vs. double encoding), whether feedback prevents erroneous encoding, and how effectively learners update their knowledge from feedback. Across simulations, classification leads to poorer learning when incorrect initial judgments result in erroneous encoding, but classification matches or exceeds observation when feedback prevents such errors and supports updating of existing knowledge. These findings help reconcile prior discrepancies in the literature, which have reported advantages for classification, observation, and sometimes comparable learning between the two.