The Maximum Entropy based Rule Selection Model for Statistical Machine Translation (Invited Paper)

This paper presents a novel rule selection model for statistical machine translation (SMT) that uses the maximum entropy approach to predict target-side for an ambiguous source-side. The maximum entropy based rule selection (MERS) model combines rich contextual information as features, thus can help SMT systems perform context-dependent rule selection. We incorporate the MERS model into two kinds of the state-of-the-art syntax-based SMT models: the hierarchical phrase-based model and the tree-to-string alignment template model. Experiments show that our approach achieves significant improvements over both the baseline systems.