- The phonetics-phonology interface
- Learnability and exceptionality
For my MA, I modeled the COG of English /ʃ/ as a variable dependent on phonetic, phonological, prosodic, and lexical factors. I was interested in elucidating the interactions between these variables, and this necessitated creating a model which produces one type of phonetic detail with significant accuracy.
With an operationalization of feed-forwardness (in the Chomskyian sense), I was able to examine how well different grammatical architectures predict my data. I found that a feed-forward, planning model architecture best predicts my data and with a minimum of predictors. This was in contrast to production model architectures, which are less feed-forward and had more extraneous predictors. Somewhat surprisingly, I also found that the inclusion of non-lexical factors negated any direct explanatory power of lexical variables (e.g., lexical frequency). Rather, lexical variables were only found to modulate the strength of other variables, themselves not directly contributing to the output.
My results urge us to conduct more complex experiments where we can consider as many potentially relevant variables as possible. They also suggest that through inclusion of additional variables, we might see explanatory power shifted from main effects to interaction effects, which I argue to be a marker of feed-forward grammatical architecture.
I am working with Byron Ahn at Princeton University and Emily Gasser and Donna Jo Napoli at Swarthmore College on the phonetic correlates of sounding like a newscaster. This project is being wrapped up.
Analysis has shown some interesting results: neither the amount nor proportion of accented words seems to have an effect on the perceived newscaster-ness of a speaker, even though the general intuition is that newscasters emphasize more words. We have found that newscasters spend much less time in the lowest 25% of their range than non-newscasters, and much more time in the middle 50% — despite this, listeners only attend to the increased time in the middle 50%. Surprisingly, we have also found that, although newscasters speak slower than non-newscasters, speech rate has no effect on perceived newscaster-ness.
We argue that these phonetic differences between newscasters and non-newscasters reflects a difference in the goals that each population is attempting to achieve. The mismatch between performance and perceived newscaster-ness we argue to reflect an incorrect/imprecise understanding of what newscasters are attempting to do, conversationally.