Learning phonetic categories from infant-directed speech
Infants begin to learn their native language by discovering the phonetic categories that define consonants and vowels. It has been widely assumed that categories are acquired using a statistical learning mechanism that groups together tokens that are similar along acoustic dimensions. However, recent studies have found that infant-directed speech is highly variable, and it is unlikely that the problem of category learning can be solved by relying exclusively on statistical clustering. In this talk I will present a corpus of infant-directed speech which provides further evidence on the variability of vowel distributions. I will then present computational modeling studies that explore how infants might navigate through highly variable input. The models focus on two sources of contextual information which might guide the learning of vowel categories: prosody and consonantal context. The findings support a view in which the infant’s learning mechanism is anchored in context, in order to cope with variability in the linguistic environment.