Dr. David Howcroft

Machine learning (Bayesian non-parametric methods) for natural language generation. My system learns sentence planning rules from a corpus, representing them as a synchronous tree-adjoining grammar. This work is informed by my collaborations with psycholinguistic researchers.

Thesis: "Ranking Sentences by Complexity" Extracted psycholinguistic features correlated with sentence processing difficulty and trained statistical models to predict the relative difficulty of pairs of sentences using the resulting features.

Double Bachelor's degrees in Mathematics (BS) and Linguistics (BA). Also studied Japanese (3 years), Chinese (2 years), Korean (1 year), Sanskrit (1 year), French (1 semester), and Arabic (1 semester).