Robinson Hall A, #447
May 18, 2015, 10:00 AM to 07:00 AM
A rich body of research has shown that language learners can track and use distributional information in the input to acquire multiple levels of linguistic structure (see Krogh et al., 2013 for a review). There is reason to believe, however, that learning about the statistics of the input is not confined to language acquisition, but is part of ongoing language experience. In particular, language processing appears to be influenced by expectations—e.g., about probable sounds, words, structures—which are dynamic and can be rapidly updated based on the current linguistic environment (e.g., Fine et al., 2013). If a general mechanism like statistical learning underlies both acquisition and later processing and use, a clear prediction is made: performance on an independent measure of statistical learning should correlate with ability to adapt native language expectations based on novel information.
This dissertation reports three sets of experiments designed to address this prediction. The first set of experiments serves as proof of concept, demonstrating that response time data can be collected accurately and efficiently over the web using crowd-sourcing services like Amazon Mechanical Turk. The second set of experiments explores the relationship between statistical learning ability and syntactic adaptation, showing that the two processes co-vary within an individual. The third set of experiments tests whether statistical learning also underlies phonetic adaptation, addressing the novel prediction that syntactic adaptation and phonetic adaptation are driven by the same underlying cognitive mechanism.