Learners Use Word-Level Statistics in Phonetic Category Acquisition

One of the first challenges that language learners face is discovering which sounds make up their language. Evidence suggests that infants learn about the phonetic categories of their language between six and twelve months, as demonstrated by reduced discrimination of non-native contrasts and enhanced discrimination of native language phonetic contrasts (Werker & Tees, 1984; Narayan, Werker, & Beddor, 2010). The problem of how infants acquire phonetic categories is typically considered in isolation. In this paper, however, we consider the learning problem from a broader perspective, testing the hypothesis that word-level information can feed back to influence phonetic category acquisition. Distributional learning accounts (Maye, Werker, & Gerken, 2002) propose that learners obtain information about which sounds are contrastive in their native language by attending to the distributions of speech sounds in acoustic space. If learners hear a bimodal distribution of sounds along a particular acoustic dimension, they can infer that the language contains two categories along that dimension; conversely, a unimodal distribution provides evidence for a single phonetic category. Distributional learning is supported by evidence that adults and infants are sensitive to these types of speech sound distributions. Maye and Gerken (2000) tested adults’ sensitivity to distributional information in a phonetic category learning task. In their experiment, participants were told that they would be listening to a new language; the language consisted of monosyllables whose initial stop consonants were drawn from either a unimodal