A POS-Based Model for Long-Range Reorderings in SMT

Proceedings of the Fourth Workshop on Statistical Machine Translation , pages 206–214, Athens, Greece, 30 March – 31 March 2009. c⃝2009 Association for Computational Linguistics A POS-Based Model for Long-Range Reorderings in SMT Jan Niehues and Muntsin Kolss Universit¨at Karlsruhe Karlsruhe, Germany {jniehues,kolss}@ira.uka.de Abstract In this paper we describe a new approach to model long-range word reorderings in statistical machine translation (SMT). Un- til now, most SMT approaches are only able to model local reorderings. But even the word order of related languages like German and English can be very different. In recent years approaches that reorder the source sentence in a preprocessing step to better match target sentences according to POS(Part-of-Speech)-based rules have been applied successfully. We enhance this approach to model long-range reorder- ings by introducing discontinuous rules. We tested this new approach on a German- English translation task and could signifi- cantly improve the translation quality, by up to 0.8 BLEU points, compared to a sys- tem which already uses continuous POS- based rules to model short-range reorder- ings. 1 Introduction Statistical machine translation (SMT) is currently the most promising approach to machine transla- tion of large vocabulary tasks. The approach was first presented by Brown et al. (1993) and has since been used in many translation systems (Wang and Waibel, 1998), (Och and Ney, 2000), (Yamada and Knight, 2000), (Vogel et al., 2003). State- of-the-art SMT systems often use translation mod- els based on phrases to describe translation corre- spondences and word reordering between two lan- guages. The reordering of words is one of the main difficulties in machine translation. Phrase-based translation models by themselves have only limited capability to model different word orders in the source and target language, by capturing local reorderings within phrase pairs. In addition, the decoder can reorder phrases, subject to constraints such as confining reorderings to a relatively small window. In combination with a distance-based distortion model, some short-range reorderings can be handled. But for many lan- guage pairs this is not sufficient, and several au- thors have proposed additional reordering mod- els as described in Section 2. In this work we present a new method that explicitly handles long- range word reorderings by applying discontinu- ous, POS-based reordering rules. The paper is structured as follows: In the next section we present related work that was carried out in this area. Afterwards, we describe the prob- lem of long-range reordering. In Section 4 the existing framework for reordering will be intro- duced. Section 5 describes the extraction of rules modeling long-range reorderings, and in the fol- lowing section the integration into the framework will be explained. Finally, the model will be eval- uated in Section 7, and a conclusion is given in Section 8. 2 Related Work Several approaches have been proposed to ad- dress the problem of word reordering in SMT. Wu (1996) and Berger et al. (1996), for example, re- strict the possible reorderings either during decod- ing time or during the alignment, but do not use any …

Content truncated to 500 words