Dynamically Shaping the Reordering Search Space of Phrase-Based Statistical Machine Translation

Dynamically Shaping the Reordering Search Space of Phrase-Based Statistical Machine Translation Arianna Bisazza and Marcello Federico Fondazione Bruno Kessler Trento, Italy {bisazza,federico}@fbk.eu Abstract Defining the reordering search space is a cru- cial issue in phrase-based SMT between dis- tant languages. In fact, the optimal trade- off between accuracy and complexity of de- coding is nowadays reached by harshly lim- iting the input permutation space. We pro- pose a method to dynamically shape such space and, thus, capture long-range word movements without hurting translation qual- ity nor decoding time. The space defined by loose reordering constraints is dynamically pruned through a binary classifier that predicts whether a given input word should be trans- lated right after another. The integration of this model into a phrase-based decoder im- proves a strong Arabic-English baseline al- ready including state-of-the-art early distor- tion cost (Moore and Quirk, 2007) and hierar- chical phrase orientation models (Galley and Manning, 2008). Significant improvements in the reordering of verbs are achieved by a sys- tem that is notably faster than the baseline, while BLEU and METEOR remain stable, or even increase, at a very high distortion limit. 1 Introduction Word order differences are among the most impor- tant factors determining the performance of statisti- cal machine translation (SMT) on a given language pair (Birch et al., 2009). This is particularly true in the framework of phrase-based SMT (PSMT) (Zens et al., 2002; Koehn et al., 2003; Och and Ney, 2002), an approach that remains highly competitive despite the recent advances of the tree-based approaches. During the PSMT decoding process, the output sentence is built from left to right, while the input sentence positions can be covered in different or- ders. Thus, reordering in PSMT can be viewed as the problem of choosing the input permutation that leads to the highest-scoring output sentence. Due to efficiency reasons, however, the input permutation space cannot be fully explored, and is therefore lim- ited with hard reordering constraints. Although many solutions have been proposed to explicitly model word reordering during decoding, PSMT still largely fails to handle long-range word movements in language pairs with different syntac- tic structures1. We believe this is mostly not due to deficiencies of the existing reordering models, but rather to a very coarse definition of the reorder- ing search space. Indeed, the existing reordering constraints are rather simple and typically based on word-to-word distances. Moreover, they are uni- form throughout the input sentence and insensitive to the actual words being translated. Relaxing this kind of constraints means dramatically increasing the size of the search space and making the reorder- ing model’s task extremely complex. As a result, even in language pairs where long reordering is reg- ularly observed, PSMT quality degrades when long word movements are allowed to the decoder. We address this problem by training a binary classifier to predict whether a given input position should be translated right after another, given the words at those positions and their contexts. When this model is integrated into the …

Content truncated to 500 words