Output details
11 - Computer Science and Informatics
University of Edinburgh
Incremental, Predictive Parsing with Psycholinguistically Motivated Tree-Adjoining Grammar
<22>Originality: Proposed a computational model that accounts for prediction, a key component in human parsing that has not previously been modelled.
Significance: The model unifies two main theories of human parsing: Dependency Locality Theory and Surprisal. They make contradictory assumptions and explain complementary experimental data. Our model unifies them, explaining an unprecedented range of data. The ground-breaking work garnered major prizes for Demberg: Glushko dissertation prize (Cogsci Society), distinguished dissertation runner-up (BCS). Computational Linguistics is the top journal in the eponymous field.
Rigour: Unlike previous psycholinguistic models, our parser is rigorously evaluated on coverage, precision, recall using standard Parseval metrics.