ter Schure, S.
ACLC, University of Amsterdam
In their first year, infants' perceptual sensitivity to non-native speech sounds decreases, while sensitivity to native speech sounds remains. One mechanism that could explain this development is distributional learning (Maye, Werker & Gerken, 2002): Infants learn to categorize sounds into phonemes on the basis of their relative frequencies in the input. So far, this hypothesis has been investigated only in the context of auditory exposure to the phoneme distribution, and then mostly for consonants. In the current study, we tested whether infants learn a vowel contrast from a multimodal distribution of speech. Visual and auditory instances of a woman saying /fEp/ and /faep/ were manipulated to create an audiovisual continuum of this English vowel contrast. In an eye-tracking experiment we presented Dutch 8-month-old infants with a one-peaked version (midpoints most frequent) or a two-peaked version (endpoints most frequent) of this continuum. After 2.5 minutes, all infants were habituated to one of the training videos. Next, a video from the other side of the continuum was played. Infants in the two-peaked training group looked longer at this switch item compared to a repetition of the habituation item (M 5.2s at switch - 4.6s at same) than infants in the one-peaked group (M 4.6s at switch - 5.5s at same). The effect of training condition on the difference between looking times at the two items was marginally significant (F (1,36) = 3.849, p = 0.058). This finding shows that infants can use the distribution of multimodal information to build non-native vowel categories.