... 31–36,Uppsala, Sweden, 13 July 2010.c2010 Association for Computational LinguisticsUnsupervised Search for The Optimal Segmentation for StatisticalMachine TranslationCos¸kun Mermer1,3and ... segmentationschemes can outperform a straightforward linguis-tic morphological segmentation, e.g., (Habash andSadat, 2006), and (ii) it may result in even worseperformance than a word-based system, ... likelihood for Model 1. However, running the EM algorithmto optimization for each considered segmentationmodel can be computationally expensive, and canresult in overtraining. Therefore, in...