Speaker: Aleksei Nazarov (Harvard) Title: Learning to mark exceptionality in probabilistic OT Date/Time: Monday, March 13, 5:00–6:30pm Location: 32-D461 Abstract:
In this work in progress, I seek to simulate how the language-learning infant learns that certain words are exceptions to their phonological grammar. Existing learners that assign exceptionality marking to words in OT (Becker 2009, Coetzee 2009) are non-probabilistic, making them unable to represent within-word variation (Coetzee and Pater 2011; see Temkin-Martínez 2010 for the necessity of representing both variation and exceptionality). The logic of those learners – comparing (pairwise) ranking conditions between words – cannot be applied to most existing probabilistic OT learners (e.g., Boersma 1997, Goldwater and Johnson 2003). I present an extension of Jarosz’s (2015) Expectation Driven Learning approach that is able to embody this logic and induce exceptionality labels. The efficacy of this approach is tested on several mini-case studies, including the case of default and exceptional Dutch stress (Kager 1989).