Thursday 11th August
(the day after the
European NLG workshop)
10am to 1pm (including a short refreshment break)
University of Aberdeen, Old Aberdeen
Lecture room 2 (1st floor),
Meston Building
Kevin Knight, USC/Information Sciences Institute, USA
Abstract: The statistical approach to machine translation provides a set of techniques for (1) automatically learning translation knowledge from bilingual data, and (2) applying that knowledge to translate previously-unseen sentences. When it was first introduced, statistical MT was far too slow and inaccurate to be useful -- it was an interesting lab experiment. In 2005, we see statistical MT significantly outperforming other methods in many language pairs and domains, at speeds permitting commercial applications like foreign news broadcast translation. What made this possible? This tutorial will cover the basic theory and the major technical advances of the past few years. Of course, there is a long way to go! The tutorial will also cover known limitations of current MT models and describe current research trends. We will also discuss problems in natural language generation, where the input is typically more abstract than foreign text, and describe how statistical MT research is currently exploiting linguistic categories.This tutorial is free of charge. It is hosted by the Natural Language Generation group at the University of Aberdeen. We are grateful for the support of EPSRC grant EP/C523156/1 which has made this tutorial possible.
If you are interested in attending this tutorial, please send an email to ccameron@csd.abdn.ac.uk so that you can be allocated a place and informed of any further developments. For more information, contact Chris Mellish (cmellish@csd.abdn.ac.uk).