International Workshop on Spoken Language Translation (IWSLT) 2005 Pittsburgh, PA, USA October 24-25, 2005

IBM Statistical Machine Translation for Spoken Languages

Young-Suk Lee

IBM T. J. Watson Research Center, Yorktown Heights, NY, USA

We discuss performance enhancing techniques we have developed for the IWSLT 2005 Evaluation Campaign: (i) a phrase acquisition technique which expands the phrase boundaries to include target words aligned to null source words in a principled manner, and (ii) a system combination technique which selects the minimum cost translation output out of many translation outputs of the same input segment produced by various systems using different phrase translation lexicons. We also discuss IBM system performances in the Arabic to English and Chinese to English translation evaluations of the IWSLT 2005 evaluation campaign.

Full Paper    Presentation

Bibliographic reference.  Lee, Young-Suk (2005): "IBM statistical machine translation for spoken languages", In IWSLT-2005, 76-83.