Benitez, V. L. . & Saffran, J. .
University of Wisconsin-Madison
Artificial language learning experiments demonstrate that infants and adults can track statistical regularities to find the word boundaries in speech (Saffran et al., 1996). Learners also acquire probabilistic phonotactic constraints from brief exposures to novel words (Chambers et al., 2003). In natural language learning, however, learners must combine these processes: learners must segment a speech stream to identify the regularities that are shared by those words. In this study, we asked whether infants and adults can acquire a probabilistic phonotactic pattern from fluent speech, necessitating segmentation prior to phonotactic learning. Learners heard an artificial language composed of 8 di-syllabic words with no cues to word boundaries except for the transitional probabilities between syllables. Critically, 75% of the words were structured with a /t/ onset and a /u/ offset. Adults (N = 53) successfully acquired this phonotactic pattern, generalizing it to novel lexical items. Nine-month-olds (N = 24) demonstrated marginal success at acquiring and generalizing the probabilistic pattern, but 8-month-olds (N = 22) showed no evidence of learning. The results suggest a developmental trajectory for the ability to discover phonotactic patterns in fluent speech.