Aletheia Cui will be defending her dissertation proposal on Tuesday, May 1st at 10:00 a.m.
Title: The Emergence of Phonological Categories
Advisor: Charles Yang
Proposal Committee: Don Ringe, Gene Buckley, Jianjing Kuang
This dissertation investigates how phonological features are learned in first language acquisition. While phonological features are often assumed to be innate and universal (Chomsky and Halle, 1968), recent work favors the idea that phonological features are emergent and that they are acquired from linguistic input (Dresher, 2004; Mielke, 2008; Clements and Ridouane, 2011). Research in first language acquisition reveal that infants are excellent phonetic learners, and these results led to proposals of statistical learning as a means of sound category acquisition. However, statistical cues alone cannot adequately account for the acquisition of phonological contrasts since there is often significant overlap between categories. Additionally, many studies do not make the distinction between phonetic and phonological categories, a crucial difference in phonology. In this dissertation, I propose that lexical contrast is a sufficient cue for learning the necessary phonological features.
To support this proposal, I implement a computational model of phonological feature acquisition based on lexical distinctions. In this model, when the learner identifies two words as distinct in meaning, a phonologically contrastive dimension is learned, and the most prominent acoustic distinction between these two words is associated with this learned phonological contrast. Additional phonological contrasts are learned when the current contrasts are insufficient for representing the lexical distinctions the learner has identified. I implemented this model and tested the model on a small, simulated lexicon. The results indicate that this model is successful in identifying the relevant number of contrasts for the given lexicon. This baseline model was extended to learn co-varying acoustic cues for each phonological contrast and homophonous representations. In addition to the model, I used the Providence Corpus to identify what lexical information is available in the parental input. Overall, I demonstrate that lexical contrast cues are a central part in the acquisition of phonology.