Gaja Jarosz (Yale) will be presenting a talk on Friday, May 1st at 2:30 in the Linguistics Department Library.
Title: Learning of Phonology: Integrating Developmental and Computational Perspectives
Abstract:
Work in language learnability focuses on defining and finding solutions to language learning problems and understanding the conditions under which successful learning can take place. Empirical and experimental research in phonological acquisition investigates the mechanisms by which children (actually) acquire language and the properties of the input and assumptions about their innate endowment or biases that can explain language development. Although the former line of work (often) deals with formal systems and the latter with the real data, many of the questions are fundamentally the same. While some work in computational modeling has been informed by developmental and experimental findings, in this talk I advocate an even tighter interaction, arguing that these lines of research are mutually informing.
Focusing on the domain of phonological learning, I discuss two sets of experiments comparing the results of computational simulations with developmental and experimental findings. Recent work examining the cross-linguistic acquisition of initial vs. final consonant clusters has raised questions about the relative roles of frequency, markedness, and other factors in determining acquisition order (Kirk and Demuth, 2005; Demuth and McCullough, to appear). I show that a particular formulation of frequency, markedness, and their interaction correctly predicts the acquisition order cross-linguistically and is in fact the exact formulation embodied in several existing Optimality Theoretic learning models. In the second part of the talk I discuss recent experimental findings addressing the interaction of phonotactic learning and the learning of voicing alternations in Dutch (Zamuner, Kerkhoff, & Fikkert, in prep). I show that a computational model of phonological grammar and lexicon learning predicts the same effects observed in the experiment and that examination of the model can provide a potential explanation of the complex experimental results. Finally, for both sets of experiments I discuss additional testable predictions made by the computational models that could inform further experimental work.