Proceedings of the Annual Meeting of the Cognitive Science Society

Practice improves learning, yet it remains unclear which learning mechanism drives memory change across practice and testing. Competing theories attribute learning to repeated exposure, error correction, or successful retrieval, but these mechanisms are rarely evaluated in direct competition with one another. We propose a modeling approach that enables learning mechanisms to be evaluated during practice and during tests within a dynamic representation of memory change over time. Learning mechanisms are formalized as competing update rules and evaluated using nested models fit to spaced learning data with delayed tests. Results show that learning during practice is best explained by a combination of exposure-based learning and effortful successful retrieval. During posttests without feedback, learning, when present, is driven primarily by successful retrieval, with no additional benefit of retrieval effort. These findings suggest that practice supports learning through multiple mechanisms, but that the mechanisms depend on the constraints of the learning context.