Seminar in Computational Linguistics

  • Datum: –14.30
  • Plats:
  • Föreläsare: Ahmed Ruby
  • Kontaktperson: Artur Kulmizev
  • Föreläsning

Towards modeling implicit and underspecified language

Exploring aspects of sentential meaning that are implicit or underspecified in context is important for sentence understanding. Despite recent developments in NLP, which have led to high performance on downstream NLP tasks, they still reflect only explicitly stated aspects of meaning. Implicit/Underspecified phenomena are currently under-explored in NLP, but it is very important for a concrete understanding of the situation. In this talk, I will present our work of the shared task at the First Workshop on Understanding Implicit and Underspecified Language. We propose a novel architecture based on mentions to predict whether a sentence involves underspecification information and thus requires clarification or revision. I will also present my initial work with some implicit phenomena such as ellipsis, fused constructions, and discuss some ideas for future work.