International Workshop on Spoken Language Translation, Honolulu, HI,
Wade Shen, Tim Anderson, Ray Slyh
This paper describes the MIT-LL/AFRL statistical MT system and the improvements that were developed during the IWSLT 2008 evaluation campaign. As part of these efforts, we experimented with a number of extensions to the standard phrase-based model that improve performance for both text and speech-based translation on Chinese and Arabic transla¬tion tasks.
We discuss the architecture of the MIT-LL/AFRL MT system, improvements over our 2007 system, and experi-ments we ran during the IWSLT-2008 evaluation. Specifi¬cally, we focus on 1) novel segmentation models for phrase-based MT, 2) improved lattice and confusion network decod¬ing of speech input, 3) improved Arabic morphology for MT preprocessing, and 4) system combination methods for ma¬chine translation.