This paper describes the MIT-LL/AFRL statistical MT system and the improvements that were developed during the IWSLT 2011 evaluation campaign. As part of these efforts, we experimented with a number of extensions to the standard phrase-based model that improve performance on the Arabic to English and English to French TED-talk translation tasks. We also applied our existing ASR system to the TED-talk lecture ASR task. We discuss the architecture of the MIT-LL/AFRL MT system, improvements over our 2010 system, and experiments we ran during the IWSLT-2011 evaluation. Specifically, we focus on 1) speech recognition for lecture-like data, 2) cross-domain translation using MAP adaptation, and 3) improved Arabic morphology for MT preprocessing.
Cite as: Aminzadeh, A.R., Anderson, T., Slyh, R., Ore, B., Hansen, E., Shen, W., Drexler, J., Gleason, T. (2011) The MIT-LL/AFRL IWSLT-2011 MT system. Proc. International Workshop on Spoken Language Translation (IWSLT 2011), 34-40
@inproceedings{aminzadeh11_iwslt,
author={A. Ryan Aminzadeh and Tim Anderson and Ray Slyh and Brian Ore and Eric Hansen and Wade Shen and Jennifer Drexler and Terry Gleason},
title={{The MIT-LL/AFRL IWSLT-2011 MT system}},
year=2011,
booktitle={Proc. International Workshop on Spoken Language Translation (IWSLT 2011)},
pages={34--40}
}