about

CogNetwork is an interdisciplinary weekly forum focused on discussing ongoing research in language and cognition; both theoretical and experimental approaches are welcomed.

It features presentations of works in progress by members of the UC Berkeley Linguistics Department and other campus departments as well as invited visitors.

A major goal of the forum is to foster cross-department and cross-disciplinary communication and contact.

when & where

Friday 12:00-13:30
1229 Dwinelle Hall

contact information

February 17, 2017

Simon Todd (Stanford)

Connecting Speech Perception and Sound Change: A Computational Model of Push-Chains

In this talk, I present an exemplar-based computational model of sound change. The model captures an interaction between two phonological categories (e.g. vowels), where one category is subjected to systematic external bias that causes it to encroach on the phonetic/acoustic space of the other category. This creates overlap between the categories and a consequent perceptual disadvantage for productions falling in the overlap due to their low discriminability, yielding a push-chain by self-organizing principles. I show that many parameter settings yield the same qualitative properties as seen in the Short Front Vowel Shift of New Zealand English: the categories maintain their width and overlap whilst moving together. Furthermore, I show that the introduction of the assumption that high-frequency words are more robust to low-discriminability disadvantages in perception leads to word-frequency effects in production. Low-frequency words come to lead the change in the retreating category, as observed in New Zealand English (Hay, Pierrehumbert, Walker, and LaShell, 2015). The model also makes predictions for future studies: it predicts high-frequency words to lead in the advancing category, and it predicts high-frequency words to yield a greater lexical bias in synchronic spoken word perception. I discuss studies planned to investigate these predictions.

[joint work with Janet Pierrehumbert (University of Oxford, UK) and Jen Hay (University of Canterbury, NZ)]