Published online by Cambridge University Press: 26 September 2008
Previous work has suggested that infants may segment continuous speech by a BRACKETING STRATEGY that segregates portions of the speech stream based on prosodic cues to their endpoints. The two present studies were designed to assess whether infants also can deploy a CLUSTERING STRATEGY that exploits asymmetries in transitional probabilities between successive elements, aggregating elements with high transitional probabilities and identifying points of low transitional probabilities as boundaries between units. These studies examined effects of the structure and redundancy of speech context on infants' discrimination of two target syllables using an operant head-turning procedure. After discrimination training on the target syllables in isolation, discrimination maintenance was tested when the target syllables were embedded in one of three contexts. Invariant Order contexts were structured to promote clustering, whereas the Redundant and Variable Order contexts were not. Thirty-six seven-month-olds were tested in Experiment I, in which stimuli were produced with varying intonation contours; 36 eight-month-olds were tested in Experiment 2, in which stimuli were produced with comparable flat pitch contours. In both experiments, performance of the three groups was equivalent in an initial 20-trial test. However, in a second 20-trial test, significant improvements in performance were shown by infants in the Invariant Order condition. No such gains were shown by infants in the other two conditions. These studies suggest that clustering may complement bracketing in infants' discovery of units of language.
Study 1 formed part of a doctoral dissertation by the first author submitted to the University of Minnesota. This work was supported in part by a Dissertation Research Grant to J. Goodsitt from the University of Minnesota, NICHD Grant T32-HD07151 to the Center for Research in Learning, Perception, and Cognition, University of Minnesota, and NIH Grant 26521 and a MacArthur Foundation Grant to Patricia Kuhl. We thank Casey Haake and Robert Ling for assistance in developing computer software, and Caroline Abdala, James White and Karen Wolak for assistance in testing subjects. We also thank Jeanne Miller and two reviewers for helpful comments on previous drafts.