Efficient Solutions for Word Reordering in German-English Phrase-Based Statistical Machine Translation

Proceedings of the Eighth Workshop on Statistical Machine Translation, pages 440–451, Sofia, Bulgaria, August 8-9, 2013 c⃝2013 Association for Computational Linguistics Efficient solutions for word reordering in German-English phrase-based statistical machine translation Arianna Bisazza and Marcello Federico Fondazione Bruno Kessler Trento, Italy {bisazza,federico}@fbk.eu Abstract Despite being closely related languages, German and English are characterized by important word order differences. Long- range reordering of verbs, in particular, represents a real challenge for state-of-the- art SMT systems and is one of the main reasons why translation quality is often so poor in this language pair. In this work, we review several solutions to improve the accuracy of German-English word re- ordering while preserving the efficiency of phrase-based decoding. Among these, we consider a novel technique to dynamically shape the reordering search space and effectively capture long-range reordering phenomena. Through an extensive eval- uation including diverse translation qual- ity metrics, we show that these solutions can significantly narrow the gap between phrase-based and hierarchical SMT. 1 Introduction Modeling the German-English language pair is known to be a challenging task for state-of-the- art statistical machine translation (SMT) methods. A major factor of difficulty is given by word or- der differences that yield important long-range re- ordering phenomena. Thanks to specific reordering modeling compo- nents, phrase-based SMT (PSMT) systems (Zens et al., 2002; Koehn et al., 2003; Och and Ney, 2002) are generally good at handling local re- ordering phenomena that are not captured inside phrases. However, they typically fail to predict long reorderings. On the other hand, hierarchi- cal SMT (HSMT) systems (Chiang, 2005) can learn reordering patterns by means of discontinu- ous translation rules, and are therefore considered a better choice for language pairs characterized by massive and hierarchical reordering. Looking at the results of the Workshop of Machine Translation’s last edition (WMT12) (Callison-Burch et al., 2012), no particular SMT approach appears to be clearly dominating. In both language directions (official results excluding the online systems) the rule-based systems outper- formed all SMT approaches, and among the best SMT systems we find a variety of approaches: pure phrase-based, phrase-based and hierarchical systems combination, n-gram based, a rich syntax- based approach, and a phrase-based system cou- pled with POS-based pre-ordering. This gives an idea of how challenging this language pair is for SMT and raises the question of which SMT ap- proach is best suited to model it. In this work, we aim at answering this ques- tion by focussing on the word reordering problem, which is known to be an important factor of SMT performance (Birch et al., 2008). We hypothe- size that PSMT can be as successful for German- English as the more computationally costly HSMT approach, provided that the reordering-related pa- rameters are carefully chosen and the best avail- able reordering models are used. More specifi- cally, our study covers the following topics: dis- tortion functions and limits, and dynamic shaping of the reordering search space based on a discrim- inative reordering model. We first review these topics, and then evaluate …

Content truncated to 500 words