A Tree Sequence Alignment-based Tree-to-Tree Translation Model

Proceedings of ACL-08: HLT, pages 559–567, Columbus, Ohio, USA, June 2008. c⃝2008 Association for Computational Linguistics A Tree Sequence Alignment-based Tree-to-Tree Translation Model Min Zhang1 Hongfei Jiang2 Aiti Aw1 Haizhou Li1 Chew Lim Tan3 and Sheng Li2 1Institute for Infocomm Research 2Harbin Institute of Technology 3National University of Singapore mzhang@i2r.a-star.edu.sg hfjiang@mtlab.hit.edu.cn tancl@comp.nus.edu.sg aaiti@i2r.a-star.edu.sg lisheng@hit.edu.cn hli@i2r.a-star.edu.sg Abstract This paper presents a translation model that is based on tree sequence alignment, where a tree sequence refers to a single sequence of sub- trees that covers a phrase. The model leverages on the strengths of both phrase-based and lin- guistically syntax-based method. It automati- cally learns aligned tree sequence pairs with mapping probabilities from word-aligned bi- parsed parallel texts. Compared with previous models, it not only captures non-syntactic phrases and discontinuous phrases with lin- guistically structured features, but also sup- ports multi-level structure reordering of tree typology with larger span. This gives our model stronger expressive power than other re- ported models. Experimental results on the NIST MT-2005 Chinese-English translation task show that our method statistically signifi- cantly outperforms the baseline systems. 1 Introduction Phrase-based modeling method (Koehn et al., 2003; Och and Ney, 2004a) is a simple, but power- ful mechanism to machine translation since it can model local reorderings and translations of multi- word expressions well. However, it cannot handle long-distance reorderings properly and does not exploit discontinuous phrases and linguistically syntactic structure features (Quirk and Menezes, 2006). Recently, many syntax-based models have been proposed to address the above deficiencies (Wu, 1997; Chiang, 2005; Eisner, 2003; Ding and Palmer, 2005; Quirk et al, 2005; Cowan et al., 2006; Zhang et al., 2007; Bod, 2007; Yamada and Knight, 2001; Liu et al., 2006; Liu et al., 2007; Gildea, 2003; Poutsma, 2000; Hearne and Way, 2003). Although good progress has been reported, the fundamental issues in applying linguistic syn- tax to SMT, such as non-isomorphic tree align- ment, structure reordering and non-syntactic phrase modeling, are still worth well studying. In this paper, we propose a tree-to-tree transla- tion model that is based on tree sequence align- ment. It is designed to combine the strengths of phrase-based and syntax-based methods. The pro- posed model adopts tree sequence 1 as the basic translation unit and utilizes tree sequence align- ments to model the translation process. Therefore, it not only describes non-syntactic phrases with syntactic structure information, but also supports multi-level tree structure reordering in larger span. These give our model much more expressive power and flexibility than those previous models. Experiment results on the NIST MT-2005 Chinese- English translation task show that our method sig- nificantly outperforms Moses (Koehn et al., 2007), a state-of-the-art phrase-based SMT system, and other linguistically syntax-based methods, such as SCFG-based and STSG-based methods (Zhang et al., 2007). In addition, our study further demon- strates that 1) structure reordering rules in our model are very useful for performance improve- ment while discontinuous phrase rules have less contribution and 2) tree sequence rules are able to model non-syntactic phrases with syntactic struc- …

Content truncated to 500 words