A probabilistic model of phonetic cue restructuring

Research demonstrating that both infants and adults track statistical distributions of acoustic-phonetic cues and use this information when making phonetic category judgements (Maye, Werker, & Gerken, 2002; Clayards, Tanenhaus, Aslin, & Jacobs, 2008) has led to interest in computational models of phonetic category acquisition, which can shed light on the requirements and limitations of statistical learning. The results of a number of studies (Vallabha, McClelland, Pons, Werker, & Amano, 2007; Toscano & McMurray, 2008) have yielded encouraging results, indicating that Gaussian mixture (GMs) may be an appropriate means of representing phonetic category structure. However, these structures are not static; they can and do change over a speaker’s lifetime, albeit in ways which are not yet fully understood. This work builds on previous research by embedding the GM approach in an agent-based framework to explore the ways in which phonetic category structure changes over time.