... 72–80,Columbus, Ohio, USA, June 2008.c2008 Association for Computational LinguisticsCohesive Phrase-based Decoding for Statistical Machine TranslationColin Cherry∗Microsoft ResearchOne ... al., 2005). What may have been forgotten duringthis transition is that there is a reason it was once be-lieved that a cohesive translation model would work: for some language pairs, cohesion explains ... weakness isin movement modeling; its default distortion model applies a flat penalty to any deviation from source72order, forcing the decoder to rely heavily on its lan-guage model. Recently, a...