MaTrEx: the DCU machine translation system for IWSLT 2007

MaTrEx: the DCU Machine Translation System for IWSLT 2007 Hany Hassan Yanjun Ma School of ComputingAndy Way Dublin City University Dublin, Ireland {hhasan,yma,away}@computing.dcu.ie Abstract In this paper, we give a description of the machine translation system devel- oped at DCU that was used for our sec- ond participation in the evaluation cam- paign of the International Workshop on Spoken Language Translation (IWSLT 2007). In this participation, we focus on some new methods to improve sys- tem quality. Specifically, we try our word packing technique for different language pairs, we smooth our translation tables with out-of-domain word translations for the Arabic–English and Chinese–English tasks in order to solve the high number of out of vocabulary items, and finally we deploy a translation-based model for case and punctuation restoration. We participated in both the classical and challenge tasks for the following translation directions: Chinese–English, Japanese–English and Arabic–English. For the last two tasks, we translated both the single-best ASR hypotheses and the correct recognition results; for Chinese– English, we just translated the correct recognition results. We report the results of the system for the provided evaluation sets, together with some additional ex- periments carried out following identifi- cation of some simple tokenisation errors in the official runs. 1 Introduction In this paper, we describe some new exten- sions to the data-driven MT system developed at DCU, MaTrEx (Machine Translation using examples), subsequent to our participation at IWSLT 2006 (Stroppa and Way, 2006). Firstly, we extend our word packing tech- nique (Ma et al., 2007a) to Japanese and Ara- bic. Secondly, we demonstrate that smoothing the Arabic–English translation tables with out- of-domain data improves BLEU score by 9.67% relative, and 2.28% relative for Chinese–English, compared to the baseline systems. Thirdly, we treat case and punctuation as a translation task by using the translations with case differences and punctuation marks as the ‘target’ text, and the equivalent texts with these removed as the ‘source’. The system then learns how to restore case and punctuation from this bitext for this pseudo-translation pair. We participated in both the classical and chal- lenge tasks for the following translation direc- tions: Chinese–English, Japanese–English and Arabic–English. For the last two tasks, we translated both the single-best ASR hypotheses and the correct recognition results; for Chinese– English, we just translated the correct recogni- tion results. We report the results of the system for the provided evaluation sets. The remainder of the paper is organized as fol- lows. In section 2, we describe the various com- ponents of the system; in particular, we give de- tails about the various novel extensions to Ma- TrEx as summarised above. In section 3, we re- port experimental results obtained for the three language pairs, while in section 4, we conclude, and provide avenues for further research. 2 The MaTrEx System The MaTrEx system is a hybrid system which exploits both EBMT and SMT techniques to ex- tract a dataset of aligned chunks (Armstrong et al., 2006). It is a modular data-driven MT en- gine, built …

Content truncated to 500 words