Research Projects and Papers
Discovering and Modeling Vowel Harmony
Following up on Goldsmith and Xanthos and Goldsmith and Riggle, I am compare hidden Markov Models and a Minimum Description Length approach using Gibbs distributions to discover and model vowel harmony in Finnish, Turkish, Italian, and English. I show that HMMs do well at discoving vowel harmony in Finnish and does not spuriously find vowel harmony in Italian or English, but does not capture all the effects in the Turkish vowel system. The MDL approach accurately captures the facts about Turkish and Finnish vowel harmony. Further, the gains from modeling vowel-to-vowel effects in Turkish or Finnish are better than in English or Italian.
Discourse Coherence and Verb Phrase Ellipsis
I adapt Prüst, Scha, and van den Berg's (1994) account of Verb Phrase Ellipsis resolution as a side effect of discourse coherence calculations to Kehler's (2002) account of discourse coherence relations. Using Prüst et al.'s account as a model of Kehler's Parallel relation and Kehler's (2002, Chapter 6) account of pronoun resolution, I provide a model of VPE resolution as a side effect of discourse coherence that not only accounts for the simple cases, but also handles Webber's (1978) split antecedent examples.
Doubly Licensed Negative Polarity Items
Giannakidou (1999) argues that NPIs in Greek are licensed in non-veridical environments, rather than downward entailing environments, and further, that emphatic NPIs required an anti-veridical environment. This leads to the question: are NPIs (emphatic or not) licensed underneath the scope of two anti-veridical operators? My squib does little more than report Giannakidou's judgments that non-emphatic NPIs are ok and emphatic NPIs are not in such contexts, and speculates about a possible connection between this pattern and the scope taking pattern suggested by Giannakidou (2000).
Agent Based Simulations
The Chicago Language Modeling Lab has a reading group on agent-based modeling. To get the ball rolling on building actual models, Max Bane implemented an simulated version (source code) of the analytical model of language death presented by Abrams and Strogatz (2003). I altered that model by assigning agents a position and radius of influence (source code). I wrote a class paper describing the changes, my motivation for them, and some informal experiments I performed with the model.