Computational modeling in phonological acquisition

Computational modeling is playing an increasingly important role in the language sciences. The goal of this talk is to survey computational models in phonology, with a focus on what they can and can't tell us about language acquisition. The talk will open with an empirical survey of phonological acquisition, from the rudiments of rhythm to the production of morphological alternations. The central section of the talk will focus on the problem of phonotactic acquisition, and how it can be modeled computationally using a formalism known as maximum entropy harmonic grammar. The ideas will be illustrated with a case study on sonority sequencing -- the typological generalization that "good" word onsets rise in sonority, e.g. [ba] >> [bla] >> [bda] >> [lba]. In the final segment of the talk, I will discuss a problem that is well beyond the current generation of models -- the apparently concurrent development of phonetic categories, phonology, and the lexicon. I will survey the computational work that has been done on this issue, including the work I have done since I arrived at the LSCP.