The RWTH Aachen Machine Translation System for WMT 2012

Proceedings of the 7th Workshop on Statistical Machine Translation, pages 304–311, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics The RWTH Aachen Machine Translation System for WMT 2012 Matthias Huck, Stephan Peitz, Markus Freitag, Malte Nuhn 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 developed at RWTH Aachen University for the translation task of the NAACL 2012 Seventh Workshop on Statistical Machine Translation (WMT 2012). We participated in the evaluation campaign for the French-English and German-English language pairs in both translation directions. Both hierarchical and phrase-based SMT sys- tems are applied. A number of different tech- niques are evaluated, including an insertion model, different lexical smoothing methods, a discriminative reordering extension for the hierarchical system, reverse translation, and system combination. By application of these methods we achieve considerable improve- ments over the respective baseline systems. 1 Introduction For the WMT 2012 shared translation task1 RWTH utilized state-of-the-art phrase-based and hierarchi- cal translation systems as well as an in-house sys- tem combination framework. We give a survey of these systems and the basic methods they implement in Section 2. For both the French-English (Sec- tion 3) and the German-English (Section 4) language pair, we investigate several different advanced tech- niques. We concentrate on specific research direc- tions for each of the translation tasks and present the respective techniques along with the empirical re- sults they yield: For the French→English task (Sec- tion 3.1), we apply a standard phrase-based system. 1http://www.statmt.org/wmt12/ translation-task.html For the English→French task (Section 3.2), we aug- ment a hierarchical phrase-based setup with a num- ber of enhancements like an insertion model, dif- ferent lexical smoothing methods, and a discrimina- tive reordering extension. 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) and employ sys- tem combination to produce a consensus hypothesis from normal and reverse translations (Section 4.2) of phrase-based and hierarchical phrase-based setups. 2 Translation Systems 2.1 Phrase-Based System The phrase-based translation (PBT) system used in this work is an in-house implementation of the state-of-the-art decoder described in (Zens and Ney, 2008). We use the standard set of models with phrase translation probabilities and lexical smooth- ing in both directions, word and phrase penalty, distance-based distortion model, an n-gram target language model and three binary count features. The parameter weights are optimized with minimum er- ror rate training (MERT) (Och, 2003). 2.2 Hierarchical Phrase-Based System For our hierarchical phrase-based translation (HPBT) setups, we employ the open source trans- lation toolkit Jane (Vilar et al., 2010; Stein et al., 2011; Vilar et al., 2012), which has been developed at RWTH and is freely available for non-commercial use. In hierarchical phrase-based translation (Chiang, 2007), a weighted synchronous context-free grammar is induced from parallel text. 304 In addition to contiguous lexical phrases, hierar- chical phrases with up to two gaps are extracted. The …

Content truncated to 500 words