Using Semantics to help learn Phonetic Categories

Computational models of language acquisition seek to replicate human linguistic learning capabilities, such as an infant's ability to identify the relevant sound categories in a language. A key question is which aspects of the input may be relevant for a given task: is it more efficient to focus on only the most relevant cues, or can integrating cues from other domains be helpful? In this talk I will present an extension of a Bayesian model of phonetic categorisation (Feldman et al., 2013). The original model learns a lexicon as well as phonetic vowel categories, incorporating the constraint that phonemes appear in word contexts. However, it has trouble separating minimal pairs (such as 'cat'/'caught'/'kite'). Our extension adds further information via situational context information, a form of weak semantics or world knowledge, to disambiguate potential minimal pairs. This information leads to better phonetic categorisation, especially when the word contexts are degraded.