Robert Daland of U.C.L.A. will speak at the UCSD Linguistics Department Colloquium on May 17, 2010, at 2:00 pm in AP&M 4301.
The role of phonotactic grammar in word recognition models
Phonology is an important component of speech processing. In this talk, I will argue that what the phonotactic grammar should do is assign probabilities to novel forms, which represent the likelihood that a novel word of the language would take the given form. Several lines of argument converge on this position. First, open-vocabulary word recognition formally requires a component that does this. Second, there is an emerging consensus that gradient phonotactic effects require a scalar rather than a categorical/binary grammar; scalar theories have a natural probabilistic formulation. Third, such a probabilistic formulation provides a more elegant account for Possible Word Constraint (PWC) effects (Norris, McQueen, Cutler, & Butterfield, 1997) than in existing models (Norris & McQueen, 2008).
After developing this theoretical argument, I turn to a novel empirical prediction -- that novel words will follow the probability distribution of the phonotactic grammar. This prediction is tested using editions of the UCLA Slang Dictionary collected by Munro and colleagues (1989, 1993, 1997, 2001, 2005, 2009). Broadly speaking, slang words do follow the phonotactics of the language as a whole, but exhibit notable fine-grained differences, including a distributional preference for marked obstruents in certain positions (e.g. word-final /f/), and a dispreference for marked prosodic patterns (e.g. iambs). Some surprising and unsurprising frequency effects are documented as well.