Sentence Type Based Reordering Model for Statistical Machine Translation

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 1089–1096 Manchester, August 2008 Sentence Type Based Reordering Model for Statistical Machine Translation Jiajun Zhang, Chengqing Zong, Shoushan Li National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China {jjzhang, cqzong, sshanli}@nlpr.ia.ac.cn Abstract Many reordering approaches have been proposed for the statistical machine translation (SMT) system. However, the information about the type of source sentence is ignored in the previous works. In this paper, we propose a group of novel reordering models based on the source sentence type for Chinese-to- English translation. In our approach, an SVM-based classifier is employed to classify the given Chinese sentences into three types: special interrogative sen- tences, other interrogative sentences, and non-question sentences. The different reordering models are developed ori- ented to the different sentence types. Our experiments show that the novel re- ordering models have obtained an im- provement of more than 2.65% in BLEU for a phrase-based spoken language translation system. 1 Introduction The phrase-based translation approach has been the popular and widely used strategy to the sta- tistical machine translation (SMT) since Och, et al. (2002) proposed the log-linear model. How- ever, reordering is always a key issue in the de- coding process. A number of models have been developed to deal with the problem of reorder- ing. The existing reordering approaches could be divided into two categories: one is integrated into the decoder and the other is employed as a preprocessing module. © 2008. Licensed under the Creative Commons Attribu- tion-Noncommercial-Share Alike 3.0 Unported license (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. Many reordering methods belong to the for- mer category. Distortion model was first em- ployed by Koehn et al. (2003); a lexicalized re- ordering model was proposed by Och et al. (2004) and Koehn et al. (2005); and the formal syntax-based reordering models were proposed by Chiang (2005) and Xiong et al. (2006). It is worthy to note that little syntactic knowledge is used in the models mentioned above. Compared to the reordering models that are integrated into the decoder, the reordering at the source side can utilize more syntactic knowl- edge, with the goal of adjusting the source lan- guage sentence to make its word order closer to that of the target language. The most notable models are given by Xia and McCord (2004), Collins et al. (2005), Li et al. (2007) and Wang et al. (2007). Xia and McCord (2004) parsed the source and target sides of the training data and then automatically extracted the rewriting pat- terns. The rewriting patterns are employed on the input source sentence to make the word or- der more accordant to target language. Collins et al. (2005) described an approach to reorder Ger- man in German-to-English translation. The method concentrates on the German clauses and six types of transforming rules are applied to the parsed source sentence. However, all the rules are manually built. Li et al. (2007) used a parser to get the syntactic tree …

Content truncated to 500 words