Hidden Structure and Ambiguity in Phonological Learning
Jarosz, Gaja. January 2019. Hidden Structure and Ambiguity in Phonological Learning . Second Annual Meeting of the Society for Computation in LinguisticsNew York, New York.
Language acquisition proceeds on the basis of incomplete, ambiguous linguistic input, one source of which is hidden phonological structure. I argue that addressing this learning challenge requires sensitivity to quantitative properties of the data. Numerous approaches now exist for learning hidden phonological structure, many derived from methods developed in NLP/ML. Such quantitative models automatically make strong, testable predictions about how learners generalize from incomplete, ambiguous data and how statistical properties of the language input affect learning outcomes. Investigating both of these pressures experimentally and typologically is essential to understanding the language acquisition device and its influence on variation and change.