MATREX: The DCU MT System for WMT 2010

Proceedings of the Joint 5th Workshop on Statistical Machine Translation and MetricsMATR, pages 143–148, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics MATREX: The DCU MT System for WMT 2010 Sergio Penkale, Rejwanul Haque, Sandipan Dandapat, Pratyush Banerjee, Ankit K. Srivastava, Jinhua Du, Pavel Pecina, Sudip Kumar Naskar, Mikel L. Forcada, Andy Way CNGL, School of Computing Dublin City University, Dublin 9, Ireland { spenkale, rhaque, sdandapat, pbanerjee, asrivastava, jdu, ppecina, snaskar, mforcada, away }@computing.dcu.ie Abstract This paper describes the DCU machine translation system in the evaluation cam- paign of the Joint Fifth Workshop on Sta- tistical Machine Translation and Metrics in ACL-2010. We describe the modular design of our multi-engine machine trans- lation (MT) system with particular focus on the components used in this partici- pation. We participated in the English– Spanish and English–Czech translation tasks, in which we employed our multi- engine architecture to translate. We also participated in the system combination task which was carried out by the MBR decoder and confusion network decoder. 1 Introduction In this paper, we present the DCU multi-engine MT system MATREX (Machine Translation using Examples). This system exploits example-based MT, statistical MT (SMT), and system combina- tion techniques. We participated in the English–Spanish (en– es) and English–Czech (en–cs) translation tasks. For these two tasks, we employ several individual MT systems: 1) Baseline: phrase- based SMT (Koehn et al., 2007); 2) EBMT: Monolingually chunking both source and target sides of the dataset using a marker-based chunker (Gough and Way, 2004); 3) Factored translation model (Koehn and Hoang, 2007); 4) Source-side context-informed (SSCI) systems (Stroppa et al., 2007); 5) the moses-chart (a Moses imple- mentation of the hierarchical phrase-based (HPB) approach of Chiang (2007)) and 6) Apertium (For- cada et al., 2009) rule-based machine translation (RBMT). Finally, we use a word-level combina- tion framework (Rosti et al., 2007) to combine the multiple translation hypotheses and employ a new rescoring model to generate the final translation. For the system combination task, we first use the minimum Bayes-risk (MBR) (Kumar and Byrne, 2004) decoder to select the best hypoth- esis as the alignment reference for the confusion network (CN) (Mangu et al., 2000). We then build the CN using the TER metric (Snover et al., 2006), and finally search for the best translation. The remainder of this paper is organised as fol- lows: Section 2 details the various components of our system, in particular the multi-engine strate- gies used for the shared task. In Section 3, we outline the complete system setup for the shared task and provide evaluation results on the test set. Section 4 concludes the paper. 2 The MATREX System 2.1 System Architecture The MATREX system is a combination-based multi-engine architecture, which exploits as- pects of both the EBMT and SMT paradigms. The architecture includes various individual sys- tems: phrase-based, example-based, hierarchical phrase-based and tree-based MT. The combination structure uses the MBR and CN decoders, and is based on a word-level com- bination strategy (Du et al., 2009). In the …

Content truncated to 500 words