Word Learning in Quiet and in Noise: A Preliminary Study

Word learning involves finding words in continuous speech and mapping them onto novel objects. Previous research has demonstrated that infants can track the transitional probability (TP) between syllables (i.e., the likelihood two syllables will co-occur) in continuous speech to discover word boundaries. Here we ask whether infants can map sound sequences they have extracted from fluent speech onto novel objects. We used a naturally produced Italian corpus in which the TP between syllables was manipulated in 4 target words: two high TP (HTP; TP=1.0) words with component syllables only occurring within those words, and two low TP (LTP; TP=.3) words with component syllables occurring in other words throughout the corpus. After familiarization with the corpus, 20- to 24-month-olds were trained to pair HTP and LTP words with novel objects. Following training, accuracy and reaction time to find the labeled object was tested. Preliminary results suggest that infants mapped both HTP and LTP words onto novel objects, although the time course for looking at the target object may be somewhat slower for LTP words. In follow-up studies we are investigating how the presence of background nose affects infants’ ability to track and use statistical regularities during speech segmentation and word learning.