Infant-directed vowels are easier to learn for a computer model
Bart de Boer, Patricia K. Kuhl · The Journal of the Acoustical Society of America · 2001
When addressing infants, caretakers use a special register, called ‘‘motherese’’ or ‘‘parentese.’’ Kuhl et al. (Science, 1997) demonstrated that the formant patterns of the centers of vowels in this register are farther apart than those in adult-directed speech, and postulated that this helps children learn the categories of speech in their language. In this study, this hypothesis was tested with a computer model. Three words were used: sock, shoe, and sheep. Recordings of these words were available from ten mothers in both infant-directed and adult-directed registers. The first two formants were automatically extracted from the voiced parts of each word. Automatic formant extraction often results in formant patterns with errors, but no corrections were made. After formant extraction, their distributions were compared and a mixture of three Gaussians was fitted to the data points. The results demonstrated that infant-directed tokens not only used a larger part of the acoustic space, but also resulted in a better distribution of the centers of the Gaussians, indicating better patterning of the input. Better patterning of the input makes it easier to learn the categories of speech sounds. [Research supported by the NIH and the Talaris Research Institute.]