Psycholinguistic research shows that key properties of the human sentence processor are incrementality, connectedness (partial structures contain no unattached nodes), and prediction (upcoming syntactic structure is anticipated). However, there is currently no broad-coverage parsing model with these properties. In this talk, I will present a probabilistic parser for "PsychoLinguistically motivated Tree-Adjoining Grammar" (PLTAG), a new version of TAG. Our framework instantiates incrementality and connectedness, from which prediction then follows naturally. We train the parser on a TAG-transformed version of the Penn Treebank and show that it achieves performance comparable to existing incremental TAG parsers. From a psycholinguistic perspective, the key innovation of our parser is an explicit mechanism for generating and verifying syntactic predictions. We show that this mechanism makes it possible to capture both locality effects and surprisal effects, and thus unify a body of experimental results that have so far been accounted separately. In addition to these experimental results, we also show that our processing theory can explain sentence processing difficulty on broad-coverage newspaper text more adequately than previous theories.

If you would like to meet with the speaker, please contact Vera Demberg.