Projects

Learning phonetic categories from statistics

Each language has a unique set of sounds that it uses to form words. For example, the sounds [r] and [l] are used distinctively in English, as shown in a pair of words like rake and lake. In contrast, the Japanese language does not differentiate between these two sounds, and thus there is no pair of Japanese words in which the difference between [r] and [l] can result in a difference in meaning. An infant learning English or Japanese must learn whether the [r]/[l] difference is a relevant distinction in their language, but they must do this before they have learned word meanings. Our research indicates that during the first year of life, infants solve this sort of problem by keeping track of the statistics with which various sounds occur in the speech that they hear spoken around them. Current studies are examining how much an infant can learn about speech sounds on the basis of short exposure to non-meaningful speech that exemplifies various statistical distributions.

The effect of phonology on infants' perception of speech

The phonological context in which a sound occurs also affects speech perception. For example, in Cantonese the sounds [g] and [k] are contrastive in word-initial, but not word-final position (only [k] occurs at the ends of words). This results in Cantonese speakers showing better discrimination of the [g]-[k] contrast in word-initial than in word-final position (Flege, 1989). We are currently investigating the age at which context effects such as this arise, and the mechanisms underlying them.