Infants are capable of exploiting many cues, including acoustic information such as word stress, to locate word boundaries in fluent speech. The research proposed in this application is designed to uncover the developmental course of attention to acoustic cues, and the learning mechanisms that allow infants to discover what the acoustic events in the speech they hear signify about word boundaries. We hypothesize that infants use statistical learning - the process of discovering regularities by tracking patterns of covariance in the input - to initially identify word boundaries in fluent speech, then detect the acoustic regularities of the word onsets and offsets in their early vocabulary. If so, infants should be able to detect the acoustic regularities that correspond to word boundaries in novel languages, so long as they are able to extract words from such a language. Further, they should be able to learn which acoustic events indicate word boundaries not just from perfect associations, but from covariances that occur only a majority of the time, which is the type of covariance usually present in natural languages. This research will provide a model for infant acquisition of multiple acoustic regularities associated with word boundaries.
