Nature ’ S Distributional - Learning Experiment Infants ’ Input , Infants ’ Perception , and Computational Modeling

Infants begin the acquisition of language-specific phoneme perception before their first birthday. In laboratory settings, infants are able to acquire categories on the basis of distributions of speech sounds. The speech sounds in infant-directed speech are distributed in such a way that computationally modeled distributional-learning mechanisms can acquire phoneme categories categories from these distributions. It is tempting to conclude that also in real life infants acquire their language-specific phoneme perception through distributional learning from the speech sound distributions in their input. However, an integrated study of the input that infants hear, infants’ perception of those same speech sounds, and computational modeling to provide an explanatory link has never been conducted. This dissertation provides such an integrated study.