iReasearch学术科研网

摘要：Distributional learning is a proposal for how infants might learn early speech sound categories from acoustic input before they know many words. When categories in the input differ greatly in relative frequency and overlap in acoustic space, research in bilingual development suggests that this affects the course of development. In the present study we describe the nature and extent of vowel variation in nearly 900 vowel tokens in maternal speech in Gurindji Kriol, a mixed language of northern Australia, which, like bilingual input, has differences in the relative frequency of adjacent, overlapping vowel categories. In Analysis 1, we provide the first systematic account of vowel variation and phone frequency in maternal speech in Gurindji Kriol. In Analysis 2, cluster analysis was applied to the vowel formant and duration data, to see what categories might emerge from acoustic data alone. The results suggest that, were infants to base their initial vowel categories solely on the clusters emerging in acoustic space, they might likely set up relatively few vowel categories. We discuss implications for how infants may learn Gurindji Kriol and for distributional learning. Adapted from the source document