Towards a Computational Model of Early Syntactic Acquisition

We propose a formal, computational model of early syntactic acquisition, based on stochastic finite-state automata. The model learns from adult and child utterances and generalizes them such that novel utterances can be produced. Unlike much work in computational grammar induction, our model is informed by several well-established psycholinguistic principles, which are implemented as biases in the learning process. Unlike some works in cognitive psycholinguistics, our model is mathematically well-defined and robustly evaluated. We demonstrate the ability of the model to generalize by evaluating it on several corpora of child-adult interac-