A Discriminative Latent Variable-Based “DE” Classifier for Chinese-English SMT

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 286–294, Beijing, August 2010 A Discriminative Latent Variable-Based “DE” Classifier for Chinese–English SMT Jinhua Du and Andy Way CNGL, School of Computing Dublin City University {jdu, away}@computing.dcu.ie Abstract Syntactic reordering on the source-side is an effective way of handling word or- der differences. The � (DE) construc- tion is a flexible and ubiquitous syntac- tic structure in Chinese which is a ma- jor source of error in translation quality. In this paper, we propose a new classi- fier model — discriminative latent vari- able model (DPLVM) — to classify the DE construction to improve the accuracy of the classification and hence the transla- tion quality. We also propose a new fea- ture which can automatically learn the re- ordering rules to a certain extent. The ex- perimental results show that the MT sys- tems using the data reordered by our pro- posed model outperform the baseline sys- tems by 6.42% and 3.08% relative points in terms of the BLEU score on PB-SMT and hierarchical phrase-based MT respec- tively. In addition, we analyse the impact of DE annotation on word alignment and on the SMT phrase table. 1 Introduction Syntactic structure-based reordering has been shown to be significantly helpful for handling word order issues in phrase-based machine trans- lation (PB-SMT) (Xia and McCord, 2004; Collins et al., 2005; Wang et al., 2007; Li et al., 2007; Elming, 2008; Chang et al., 2009). It is well- known that in MT, it is difficult to translate be- tween Chinese–English because of the different word orders (cf. the different orderings of head nouns and relative clauses). Wang et al. (2007) pointed out that Chinese differs from English in several important respects, such as relative clauses appearing before the noun being modified, prepo- sitional phrases often appearing before the head they modify, etc. Chang et al. (2009) argued that many of the structural differences are re- lated to the ubiquitous Chinese structural parti- cle phrase � (DE) construction, used for a wide range of noun modification constructions (both single word and clausal) and other uses. They pointed out that DE is a major source of word order error when a Chinese sentence is translated into English due to the different ways that the DE construction can be translated. In this paper, we focus on improving the clas- sification accuracy of DE constructions in Chi- nese as well as investigating its impact on trans- lation quality. From the grammatical perspective, the �(DE) in Chinese represents the meaning of “noun modification” which generally is shown in the form of a Noun phrase (NP) [A DE B]. A in- cludes all the words in the NP before DE and B contains all the words in the NP after DE. Wang et al. (2007) first introduced a reordering of the DE construction based on a set of rules which were generated manually and achieved significant improvements in translation quality. Chang et al. (2009) extended this work by …

Content truncated to 500 words