Jennifer Culbertson (University of Edinburgh)
Abstract. In this talk I outline a general approach to investigating cognitive constraints that shape language. The approach formulates hypotheses based on findings from linguistic typological, tests predictions using laboratory experiments, and formalizes confirmed biases using probabilistic models. Using word order in the nominal domain as a case study, I show that adults and children learning miniature artificial languages exhibit soft biases which line up with asymmetries in typological frequency. These biases can be formalized as probabilistic prior constraints in a Bayesian model of grammar learning. I discuss the implications of this approach for a two central issue in the cognitive science of language: (1) the relative contributions of linguistic input and prior knowledge in learning, (2) the existence of hard constraints vs. soft biases in grammar.