A Reordering Approach for Statistical Machine Translation

This paper presents a Markov based hierarchical reordering scheme for lexical reordering to incorporate into phrase-based statistical machine translation system. The goal is to reorder the words and phrases in source language syntactic structure into their corresponding target language syntactic order for making translation easy. Without reordering during language translation, sentences can only be translated properly into a language with similar word order. An effective reordering scheme is essential to model translation between languages with different word orders, such as SVO-languages (English or Chinese) and SOVlanguages (Japanese or Turkish or Myanmar). In earlier work we have shown a phrase based statistical translation model for English Myanmar language translation [21]. In this paper, we focus on reorder scheme for lexical reordering to incorporate into our translation model. For a given source text sentence, reordering performs at three different lexical hierarchical levels; words level where words in each chunk are reordered in target language word order, chunks level where chunks in specific sentence clauses are reordered and clauses level where clauses in the sentence are reordered. All reordering processes will be performed by using reorder Markov model to predicate reordering of neighbor blocks (phrase pairs).