The MIT-LL/AFRL IWSLT-2011 MT System
Abstract
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.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 27, 2011
- Accession Number
- ADA576644
Entities
People
- A. R. Aminzadeh
- Brian Ore
- Eric C Hansen
- Jennifer Drexler
- Ray Slyh
- Terry Gleason
- Tim Anderson
- Wade Shen
Organizations
- Massachusetts Institute of Technology