A Unified and Discriminative Soft Syntactic Constraint Model for Hierarchical Phrase-based Translation

In the last decade, there have been a countless number of researches in soft syntactic features many of which have led to the improved performance for Hiero. However, it seems that all the syntactic constituent features cannot efficiently work together in the Hiero optimized by MERT. In this paper, we propose a more general soft syntactic constraint model based on discriminative classifiers for each constituent type and integrate all of them into the translation model with a unified form. The experimental results show that our method significantly improves the performance on the NIST05 Chinese-toEnglish translation task.