Computational Linguistics & Phonetics Computational Linguistics & Phonetics Fachrichtung 4.7 Universität des Saarlandes

Computational Linguistics Colloquium

Thursday July 5, 16:15, Seminar Room, Building 17

Processing Levels in Discourse Understanding and Generation

Michael Strube
European Media Laboratory
Heidelberg

Since the late 70's focus-based approaches were proposed as a theoretical alternative to heuristic approaches in discourse processing, e.g., anaphora resolution. Though there was some effort to apply focus-based approaches to naturally occurring discourse, it could not be shown that focus-based algorithms produce better results than heuristics with respect to anaphora resolution. In this talk, I show that the attempts to apply focus-based approaches were not successful because of the lack of the most crucial definitions: the definitions of the basic units of analysis. Based on linguistic evidence I provide definitions for the update and the clean-up unit which replace the (undefined) notion of utterance used, e.g., in the centering model. These definitions are the foundation of a new model of attentional state which narrows the gap between focus-based models and heuristics by combining the positive aspects of both kinds of approaches.

Literature:
Michael Strube (1998). Never Look Back: An Alternative to Centering. In COLING-ACL '98 (pp.1251-1257).

If you would like to meet with the speaker, please contact Kristina Striegnitz.