Computational models of early language acquisition

How do children acquire the sounds, words, and structures of their native language? Awealth of recent evidence suggests that probabilistic learning mechanisms play a role inlanguage acquisition. Nevertheless, the structure of these mechanisms is controversial andit is still unknown how broadly they apply to the tasks faced by language learners.Computational models can serve as formal theories of probabilistic learning byinstantiating proposals about the learning mechanisms available in early languageacquisition. However, fulfilling this promise requires that models be evaluated on twogrounds: their sufficiency—whether they are able to learn aspects of language givenappropriate input—and their fidelity—whether they fit the patterns of success and failureshown by human learners. I review experimental and computational evidence for theapplication of probabilistic learning across a range of acquisition tasks and argue thatmodels of probabilistic learning succeed when they use expressive representations tocapture complex regularities in the input and when they implement a parsimony bias.