Generalizing across speaker and gender during early word learning: Evidence from a statistical learning paradigm

Before children can speak, they can track the likelihood that two syllables co-occur to pull words out of a continuous stream of speech. Previous research with 17-month-olds has suggested that words that have high co-occurrence statistics (i.e., high transitional probability, HTP) make better object labels than words with low transitional probability (LTP). Here we test whether infants can generalize the patterns tracked in a continuous stream of speech to a novel speaker and gender. Infants are familiarized with an Italian corpus produced by a female speaker, that contains both HTP and LTP words. Following familiarization, infants are trained to pair HTP words with novel objects. The HTP words used during test where either produced by the same female speaker as during familiarization or by a novel male speaker. If infants recognize the HTP words when produced by a novel speaker, they should readily learn to map these words onto novel objects. The ability to generalize across speakers and genders is an important ability during early language acquisition.