In order to learn a native language, one of the first things that infants must do is assign the incoming acoustic signal to phonetic categories (i.e. consonants and vowels). However, these categories are different for each language. For example, English listeners must distinguish the sounds /l/ and /r/, but Japanese listeners must not. The paradox for the infant is how to learn which categories are relevant for the language being learned, in the absence of other knowledge about the language.

The hypothesis explored here is that infants hone in on native language phonetic categories through a sensitivity to the statistical properties of speech sounds in their environments. I will present data showing that infants are sensitive to phonetic distributions, and that this sensitivity results in changes in speech sound discrimination which mirror those seen in developing infants during the course of natural language acquisition.