Date:
Friday, March 7, 2014, 4:30pm to 6:00pm
Location:
Boylston 103
Modeling the Acquisition of Phonological Structure
Gaja Jarosz, Yale University
A rapidly growing research area in phonology investigates the grammatical underpinnings of gradient acceptability and the kinds of computational models that can explain human learning of such knowledge from language input. An important and consistent finding of this work is that knowledge of gradient and statistical patterns references the same phonological representations and principles that are relevant in the categorical domain. However, this body of work has largely ignored the question of how knowledge of static restrictions typically assumed to be involved in such tasks relates to the more traditional purview of phonological theory, namely morpho-phonological alternations and processes. Specifically, how does knowledge of gradient phonotactics interact with speakers' phonological grammars and how are these types of knowledge acquired? In this talk I develop an answer to these questions that provides a unified perspective on (gradient) phonotactics and phonology and a formal theory of their acquisition. I begin by discussing formal learnability results from my recent work that make more successful and realistic modeling of phonological acquisition possible. The computational models endow learners with learning strategies for coping with hidden structure, such as the hidden structure that arises when learners face the joint learning of phonotactics and alternations. Then, utilizing such enhanced learning strategies, I propose a computational model that casts phonological learning as a statistical inference problem in which learning of stochastic phonotactics serves as a statistical prior. This formalizes the developmental findings indicating that phonotactic learning precedes and informs the learning of phonological alternations. I show that this integrative approach establishes a mutually informing link between linguistic theory and language acquisition. I demonstrate via computational simulation that the proposed interaction between phonotactics and phonology provides a simple explanation for experimental findings on the acquisition of phonological alternations and that it leads to new, testable predictions for language acquisition.
Registration Closed