Naomi Feldman
Naomi Feldman of Brown University (soon to be Assistant Professor at the University of Maryland) will speak at the UCSD Linguistics Department Colloquium on September 27, 2010, at 2:00 pm in AP&M 4301.
A probabilistic approach to phonetic learning and perception
Infants' apparent loss of discriminative abilities for non-native contrasts between 6 and 12 months is generally interpreted as evidence for acquisition of native phonetic categories. Studies of phonetic category learning have typically approached this process in isolation, without consideration of how it may interact with other contemporaneous learning processes, such as word segmentation. Moreover, despite extensive research on links between categories and perception, the precise manner in which perceptual patterns reflect category knowledge remains unclear. This talk examines both issues from a probabilistic modeling perspective. An initial set of simulations examines learning in a learner who acquires multiple layers of linguistic structure simultaneously, comparing this to a learner who is solely focused on learning phonetic categories. Results demonstrate that simultaneous learning of multiple domains can make the learning problem more tractable. A second set of simulations draws an explicit mathematical link between category knowledge and perceptual patterns, analyzing perceptual bias toward category centers as optimal perception of speech sounds under conditions of uncertainty. Empirical data are presented in support of both computational analyses, showing cross-domain influences in a phonetic category learning task and demonstrating that listeners increase their reliance on category information under uncertainty, such as when hearing speech sounds in noise.