QuEst MODIST - Modelling Discourse in Translation

Funding

EPRSC, First grant

Summary

This project aims at explicitly modelling discourse level relationships across sentences in Statistical Machine Translation without significantly compromising the scalability of existing approaches. The hypothesis put forward is that the relationships among elements in different sentences follow some patterns and that these patterns that can be captured by examples of translations and then used to constrain the translation process. The key research challenges we seek to address are: (i) how to model discourse level relationships in a bilingual setting; and (ii) how to use these relationships to guide the translation process while keeping it tractable.

Duration

December 2013-June 2015

Team

PI Affiliation
Lucia Specia University of Sheffield
Staff
Wilker Aziz University of Sheffield
PhD students
David Steele University of Sheffield
Karin Sim Smith University of Sheffield
Collaborators
Chris Quirk Microsoft Research Redmond
SMT Group University of Edinburgh
GALA Globalization and Localization Association
TAUS Translation Automation User Society

Software and resources