In this meeting I'll present new computational work investigating how young learners identify movement dependencies in their language. I pursue the hypothesis, motivated by empirical data on syntactic development in infancy (Gagliardi, Mease, & Lidz 2016; Perkins & Lidz, in revision; Perkins & Lidz, in prep), that this learning process is "gap-driven" for languages like English. Learners may use the signal from argument gaps-- arguments that are predicted but unexpectedly missing in their canonical positions-- in order to identify when sentences contain non-local predicate-argument dependencies, and to infer what those dependencies are. Here, I'll present some new work that tests the feasibility of this hypothesis by computationally modelling the first steps of learning under this account. I'll then ask what further learning mechanisms are needed for learners to reach the target state.