Wrapper Syntax for Example-based Machine Translation

Wrapper Syntax for Example-Based Machine Translation Karolina Owczarzak Bart Mellebeek Declan Groves Josef Van Genabith Andy Way National Centre for Language Technology School of Computing, Dublin City University, Dublin 9 {owczarzak,mellebeek,dgroves,josef,away}@computing.dcu.ie Abstract TransBooster is a wrapper technology designed to improve the performance of wide-coverage machine translation systems. Using linguistically motivated syntactic information, it automatically decomposes source language sentences into shorter and syntactically simpler chunks, and recomposes their translation to form target language sentences. This generally improves both the word order and lexical selection of the translation. To date, TransBooster has been successfully applied to rule-based MT, statistical MT, and multi-engine MT. This paper presents the application of TransBooster to Example-Based Machine Translation. In an experiment conducted on test sets extracted from Europarl and the Penn II Treebank we show that our method can raise the BLEU score up to 3.8% relative to the EBMT baseline. We also conduct a manual evaluation, showing that TransBooster-enhanced EBMT produces a better output in terms of fluency than the baseline EBMT in 55% of the cases and in terms of accuracy in 53% of the cases. 1 Introduction Almost all research in Machine Translation (MT) carried out today is corpus-based, and one of the most promising research directions in that area is Example-Based Machine Translation (EBMT). EBMT models have recently achieved considerable improvements in translation quality; however, like other statistical MT systems, they still face difficulty when it comes to modelling long-distance dependencies or differences in word order between source and target languages. Our approach uses TransBooster, a wrapper technology designed to improve the output of wide-coverage MT systems (Mellebeek et al., 2005a) by exploiting the fact that both rule-based and statistical MT systems tend to perform better when translating shorter sentences than longer ones. TransBooster decomposes source language sentences into shorter, syntactically simpler chunks, sends the chunks to a baseline MT system and recomposes the translated output into target language sentences. It has already proved successful in experiments with rule-based and statistical MT systems (Mellebeek et al., 2005b, Mellebeek et al., 2006), as well as in experiments with multi-engine MT (Mellebeek et al., this volume). In this paper we apply the TransBooster wrapper technology to an Example-Based MT system. Even though we see a relative improvement of 3.8% in BLEU and 0.5% in NIST scores over the baseline EBMT system, we argue that these metrics are not able to reflect fully the improvement our method introduces. In a preliminary manual evaluation, we show that TransBooster helps obtain better translation fluency in 55% of the cases, and accuracy in 53% of the cases. This paper is organized as follows: Section 2 describes background research in EBMT; Section 3 presents the architecture of TransBooster; Section 4 describes the experimental setup; Section 5 gives the results of the experiment; Section 6 concludes. 2 Related research Example-based MT is based on bitexts, i.e. a set of sentences in one language aligned with their translations in another. Taking a corpus of source– target aligned sentence pairs, EBMT models …

Content truncated to 500 words