Edinburgh System Description for the 2005 NIST MT Evaluation

This document describes the first NIST MT Evaluation submission of the newly formed Edinburgh University Statistical Machine Translation Group. Our entry to the 2005 DARPA/NIST MT Evalua- tion was largely based on the 2004 MIT system. In a two month effort we fo- cused on adding more data and a few new features to our Arabic-English sys- tem. We also worked on preprocessing and applied some of the lessons learnt to our Chinese-English system. Our ef- forts resulted in improved translation per- formance over the previous 2004 system. Competing in the competition was also a valuable learning experience in large- scale system building. This document describes the first NIST MT Eval- uation submission from Edinburgh University's Sta- tistical Machine Translation Group. Our entry to this year's evaluation was based on the 2004 sys- tem from MIT (Koehn, 2004a). This was built by Philipp Koehn, who is now a faculty member in Ed- inburgh's School of Informatics. In a two month ef- fort, the group familiarised itself with the system and the data, and added a few new features to the Arabic- English system.