We propose a new model of human concept learning, the Rational Rules model, that provides a rational analysis for learning of rule-based concepts. This model is built upon Bayesian inference for a grammatically structured hypothesis space---a "concept language" of logical rules. We compare predictions of the model to human generalization judgments in several well-known category learning experiments, and find good agreement for both average and individual participants' generalizations. Several important concept learning effects emerge naturally in this framework. Prototype and typicality effects arise from uncertainty over the inferred definition of a concept; selective attention comes from uncertainty over production parameters of the probabilistic concept grammar. We conclude by discussing a natural extension of the model to relational features, and we describe learning of role-governed concepts---concepts defined by their role in a relational system. |