Statistical learning in natural language acquisition
Dr. Casey Lew-Williams
Tuesday, February 23, 2016, 01:00pm - 02:00pm
Princeton University, Department of Psychology
Infants and toddlers have a prodigious ability to find structure (such as words) in patterned input (such as language). Learning regularities between sounds and words often occurs seamlessly in early development, leading many to conclude that 'statistical learning' launches and enables language in the first place. This might be true. Alternatively, it might be an irrelevant artifact of distilled, simplified experiments. Here I ask: Can statistical learning scale up to explain natural language acquisition? I will address this question by presenting a series of studies that exploit various dimensions of variability inherent in real learning environments, such as the presence of multiple talkers and social/communicative cues. I will also address the question of scalability by turning to an important product of early statistical learning: the ability to process language efficiently in real time. Collectively, this work shows that we can understand the beginnings of language learning by scrutinizing the intersection of domain-general cognitive mechanisms and specific features of the language environment.