Proceedings of the Annual Meeting of the Cognitive Science Society

Our ability to learn statistically regular patterns present in our environment is central to many cognitive processes. Thereare many competing theories about what kind of mechanisms could explain this ability. While different theories makeslightly different predictions about the kinds of patterns that can be learned, they often make very different predictionsabout the process of learning. One way to constrain the set of possible theories is to measure the shape of learning curvesas people learn new patterns. To do this, we gathered response time data as people learned new patterns. We fit probabilisticmodels to individual-level data using a hierarchical Bayesian nonlinear regression. Our results suggest the learning curvesat the level of individual items tend to have strong inflection points, which is inconsistent with cognitive models that arebased purely on associative and error-driven learning.