The Automatic Acquisition of Verb Subcategorisations and Their Impact on the Performance of an HPSG Parser

We describe the automatic acquisition of a lexicon of verb subcategorisations from a domain-specific corpus, and an evaluation of the impact this lexicon has on the performance of a “deep”, HPSG parser of English. We conducted two experiments to determine whether the empirically extracted verb stems would enhance the lexical coverage of the grammar and to see whether the automatically extracted verb subcategorisations would result in enhanced parser coverage. In our experiments, the empirically extracted verbs enhance lexical coverage by 8.5%. The automatically extracted verb subcategorisations enhance the parse success rate by 15% in theoretical terms and by 4.5% in practice. This is a promising approach for improving the robustness of deep pars ing.