A POS-based model for long-range reorderings in SMT

Abstract

In this paper we describe a new approach to model long-range word reorderings in statistical machine translation (SMT). Until 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 reorderings by introducing discontinuous rules. We tested this new approach on a German-English translation task and could significantly improve the translation quality, by up to 0.8 BLEU points, compared to a system which already uses continuous POSbased rules to model short-range reorderings.

Estimating Word Alignment Quality for SMT Reordering Tasks

“…There has been much work on automatic learning of reordering rules, which can be based on different levels of annotation, such as part-of-speech tags (Rottmann and Vogel, 2007;Niehues and Kolss, 2009;Genzel, 2010), chunks or parse trees (Xia and McCord, 2004). In general, all these approaches lead to improvements of translation quality.…”

“…We learn rules from our entire SMT training corpus varying alignment models and symmetrization. To investigate only the effect of word alignment for creating reordering rules, we do not 4 Note that we do not use words (Rottmann and Vogel, 2007) or wild cards (Niehues and Kolss, 2009) Table 7: Bleu scores for part-of-speech-based reordering for different alignments change the SMT system, which is trained based on model 4+gdfa alignments. The only thing that varies for the translation task is thus the input lattice given to this SMT system.…”

“…Our first reordering task is a part-of-speech-based preordering method described by Rottmann and Vogel (2007) and Niehues and Kolss (2009), which was successfully used for German-English translation. Rules are learnt from a word aligned POS-tagged corpus.…”

The Karlsruhe Institute of Technology Translation Systems for the WMT 2015

“…As described in Rottmann and Vogel (2007), continuous reordering rules are extracted. This modeling of short-range reorderings was extended so that it can cover also long-range reorderings with noncontinuous rules (Niehues and Kolss, 2009), for German↔English systems.…”

The KIT-LIMSI Translation System for WMT 2015

“…The short-range reordering (Rottmann and Vogel, 2007) and long-range reordering (Niehues and Kolss, 2009) rules are extracted from POS-tagged versions of parallel EPPS and NC. The POS tags of those corpora are produced using the TreeTagger (Schmid, 1994).…”