The RWTH Aachen Machine Translation System for WMT 2013

Proceedings of the Eighth Workshop on Statistical Machine Translation, pages 193–199, Sofia, Bulgaria, August 8-9, 2013 c⃝2013 Association for Computational Linguistics The RWTH Aachen Machine Translation System for WMT 2013 Stephan Peitz, Saab Mansour, Jan-Thorsten Peter, Christoph Schmidt, Joern Wuebker, Matthias Huck, Markus Freitag and Hermann Ney Human Language Technology and Pattern Recognition Group Computer Science Department RWTH Aachen University D-52056 Aachen, Germany <surname>@cs.rwth-aachen.de Abstract This paper describes the statistical ma- chine translation (SMT) systems devel- oped at RWTH Aachen University for the translation task of the ACL 2013 Eighth Workshop on Statistical Machine Translation (WMT 2013). We partici- pated in the evaluation campaign for the French-English and German-English lan- guage pairs in both translation directions. Both hierarchical and phrase-based SMT systems are applied. A number of dif- ferent techniques are evaluated, including hierarchical phrase reordering, translation model interpolation, domain adaptation techniques, weighted phrase extraction, word class language model, continuous space language model and system combi- nation. By application of these methods we achieve considerable improvements over the respective baseline systems. 1 Introduction For the WMT 2013 shared translation task1 RWTH utilized state-of-the-art phrase-based and hierarchical translation systems as well as an in- house system combination framework. We give a survey of these systems and the basic meth- ods they implement in Section 2. For both the French-English (Section 3) and the German- English (Section 4) language pair, we investigate several different advanced techniques. We con- centrate on specific research directions for each of the translation tasks and present the respec- tive techniques along with the empirical results they yield: For the French→English task (Sec- tion 3.2), we apply a standard phrase-based sys- tem with up to five language models including a 1http://www.statmt.org/wmt13/ translation-task.html word class language model. In addition, we em- ploy translation model interpolation and hierarchi- cal phrase reordering. For the English→French task (Section 3.1), we train translation mod- els on different training data sets and augment the phrase-based system with a hierarchical re- ordering model, a word class language model, a discriminative word lexicon and a insertion and deletion model. For the German→English (Section 4.3) and English→German (Section 4.4) tasks, we utilize morpho-syntactic analysis to pre- process the data (Section 4.1), domain-adaptation (Section 4.2) and a hierarchical reordering model. For the German→English task, an augmented hi- erarchical phrase-based system is set up and we rescore the phrase-based baseline with a continu- ous space language model. Finally, we perform a system combination. 2 Translation Systems In this evaluation, we employ phrase-based trans- lation and hierarchical phrase-based translation. Both approaches are implemented in Jane (Vilar et al., 2012; Wuebker et al., 2012), a statistical ma- chine translation toolkit which has been developed at RWTH Aachen University and is freely avail- able for non-commercial use.2 2.1 Phrase-based System In the phrase-based decoder (source cardinality synchronous search, SCSS), we use the standard set of models with phrase translation probabilities and lexical smoothing in both directions, word and phrase penalty, distance-based distortion model, an n-gram target language model and three bi- nary …

Content truncated to 500 words