Computational Linguistics Colloquium

Thursday, 1 December 2011, 16:15
Conference Room, Building C7 4

Towards Discourse-enhanced Statistical Machine Translation (SMT)

Bonnie Webber
ILCC and School of Informatics, University of Edinburgh

Although Machine Translation is a well-established field, the majority of current systems translate sentences in isolation, losing valuable contextual information from previously translated sentences in the same document and valuable structural information from documents of a similar type.

In this talk, I will first lay out my take on discourse, and then describe recent work (at Edinburgh and elsewhere) that is attempting to incorporate various aspects of discourse into SMT, with the goal of improving both the naturalness and the interpretability of texts produced in the target language. I will conclude with a discussion of some problems that may hold up our ability to fully exploit our growing understanding of discourse and its incorporation into tools for monolingual discourse processing.

All this work is in its early stages, so the results presented are more suggestive than definitive.

If you would like to meet with the speaker, please contact Dietrich Klakow.