Specificity of Infant Statistical Learning

AbstractSensitivity to transitional probabilities (TP) in continuous speech has been extensively documented, yet little is known about how infants represent sequences that are the output of statistical learning. Across 3 experiments we test 8-month-old English-learning infants’ indexical, segmental, and suprasegmental representations of newly-encountered statistically-defined words. Following familiarization with a naturally-produced Italian corpus that contained two trochaic (strong-weak) high TP (HTP) words produced by a female speaker, infants were tested on their ability to discriminate modified HTP words (Experiment 1=male voice; Experiment 2=onset consonant change); Experiment 3=iambic stress pattern), from foils. Infants demonstrated a significant familiarity preference for modified HTP words in Experiments 1 and 3, but failed to recognize consonant modified HTP words in Experiment 2. Findings demonstrate infants can generalize representations of statistically-defined words across a range of acoustic forms less relevant to word meaning in English, but not across phonemic characteristics that are core to word meaning.

Return to previous page