The acquisition of native language phonotactics: Integrating insights from machine learning, and adult and infant experiments.

Languages differ in their phonotactics - the restrictions on positions and sequencing of segments. Many phonotactic restrictions are specific to a given language, and therefore must be learned from the input infants receive. First, we used data from adult experiments to identify maximally contrastive segmental regularities in English. Then, in three infant experiments, we showed that sensitivity to segmental regularities is induced by 5 months of age in English-learning infants. Finally, using methods from machine learning we evaluated two competing hypotheses about how infants learn phonotactics against the findings from the infant experiments. Our results show that 5-month-olds could not learn segmental restrictions from either unsegmented utterances or from words they associate with referents. We discuss the implication of these results for the mechanisms involved in phonotactic acquisition.