Building a corpus of question frames

From HLT@INESC-ID

Ana Cristina Mendes

Date

  • 15:30, Friday, October 9th, 2009
  • Room 4

Speaker

Abstract

In this presentation we address the issue of automatically building a corpus of question frames by making use of already existing knowledge resources. The goal is to express questions originally formulated in Natural Language in the form of disambiguated frames, that structure some of their important information. For this purpose, a modular approach that integrates and reuses the information contained in several external repositories was developed. It is composed by three main steps: match frames contained in a catalogue, expand the frames’ predicate, disambiguate and choose the final frame. Experiments conducted to evaluate this approach used VerbNet as the catalogue of frames, WordNet as the database of synonyms, and Google as the disambiguating engine.