N-gram-based SMT System Enhanced with Reordering Patterns

Proceedings of the Workshop on Statistical Machine Translation, pages 162–165, New York City, June 2006. c⃝2006 Association for Computational Linguistics N-gram-based SMT System Enhanced with Reordering Patterns Josep M. Crego Marta R. Costa-juss`a Jos´e B. Mari˜no Adri`a de Gispert Maxim Khalilov Jos´e A. R. Fonollosa Department of Signal Theory and Communications TALP Research Center (UPC) Barcelona 08034, Spain {jmcrego,agispert,lambert,mruiz,khalilov,rbanchs,canton,adrian}@gps.tsc.upc.edu Patrik Lambert Rafael E. Banchs Abstract This work presents translation results for the three data sets made available in the shared task “Exploiting Parallel Texts for Statistical Machine Translation” of the HLT-NAACL 2006 Workshop on Statisti- cal Machine Translation. All results pre- sented were generated by using the N- gram-based statistical machine translation system which has been enhanced from the last year’s evaluation with a tagged target language model (using Part-Of-Speech tags). For both Spanish-English transla- tion directions and the English-to-French translation task, the baseline system al- lows for linguistically motivated source- side reorderings. 1 Introduction The statistical machine translation approach used in this work implements a log-linear combination of feature functions along with a translation model which is based on bilingual n-grams (de Gispert and Mari˜no, 2002). This translation model differs from the well known phrase-based translation approach (Koehn et al., 2003) in two basic issues: first, training data is monotonously segmented into bilingual units; and second, the model considers n-gram probabilities in- stead of relative frequencies. This translation ap- proach is described in detail in (Mari˜no et al., 2005). For those translation tasks with Spanish or En- glish as target language, an additional tagged (us- ing POS information) target language model is used. Additionally a reordering strategy that includes POS information is described and evaluated. Translation results for all six translation directions proposed in the shared task are presented and dis- cussed. Both translation directions are considered for the pairs: English-Spanish, English-French, and English-German. The paper is structured as follows: Section 2 briefly outlines the baseline system. Section 3 de- scribes in detail the implemented POS-based re- ordering strategy. Section 4 presents and discusses the shared task results and, finally, section 5 presents some conclusions and further work. 2 Baseline N-gram-based SMT System As already mentioned, the translation model used here is based on bilingual n-grams. It actually con- stitutes a language model of bilingual units, referred to as tuples, which approximates the joint probabil- ity between source and target languages by using bilingual n-grams (de Gispert and Mari˜no, 2002). Tuples are extracted from a word-to-word aligned corpus according to the following two constraints: first, tuple extraction should produce a monotonic segmentation of bilingual sentence pairs; and sec- ond, no smaller tuples can be extracted without vi- olating the previous constraint. See (Crego et al., 2004) for further details. For all experiments presented here, the translation model consisted of a 4-gram language model of tu- ples. In addition to this bilingual n-gram translation model, the baseline system implements a log linear combination of five feature functions. 162 These five additional models are: • A target language model. 5-gram of …

Content truncated to 500 words