Proceedings of the Annual Meeting of the Cognitive Science Society

One of the challenges facing a child learning language is when to generalize over their input and infer productive rules. Twomathematically precise models of this problem have been proposed recently: Fragment Grammars (ODonnell, 2015) andthe Tolerance Principle (Yang, 2016). Both are based on the learner optimizing computation costs: Fragment Grammarsbalance the costs of storing forms whole and decomposing them into parts, while the Tolerance Principle reflects a trade-offbetween the processing time of serial search over all forms or only irregular forms. We implement versions of these modelsthat are directly comparable and perform a series of analyses that show that the models make systematically differingpredictions in some domains and parameter regimes. We then compare these predictions to the empirical literature on theemergence of productivity over development and evaluate which model under what assumptions provides a more accurateaccount of childrens learning.