Translation with Source Constituency and Dependency Trees

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1066–1076, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Translation with Source Constituency and Dependency Trees Fandong Meng†§ Jun Xie† Linfeng Song†§ Yajuan L¨u† Qun Liu‡† †Key Laboratory of Intelligent Information Processing Institute of Computing Technology, Chinese Academy of Sciences §University of Chinese Academy of Sciences {mengfandong,xiejun,songlinfeng,lvyajuan}@ict.ac.cn ‡Centre for Next Generation Localisation Faculty of Engineering and Computing, Dublin City University qliu@computing.dcu.ie Abstract We present a novel translation model, which simultaneously exploits the constituency and dependency trees on the source side, to com- bine the advantages of two types of trees. We take head-dependents relations of dependency trees as backbone and incorporate phrasal n- odes of constituency trees as the source side of our translation rules, and the target side as strings. Our rules hold the property of long distance reorderings and the compatibility with phrases. Large-scale experimental result- s show that our model achieves significantly improvements over the constituency-to-string (+2.45 BLEU on average) and dependency- to-string (+0.91 BLEU on average) model- s, which only employ single type of trees, and significantly outperforms the state-of-the- art hierarchical phrase-based model (+1.12 BLEU on average), on three Chinese-English NIST test sets. 1 Introduction In recent years, syntax-based models have become a hot topic in statistical machine translation. Accord- ing to the linguistic structures, these models can be broadly divided into two categories: constituency- based models (Yamada and Knight, 2001; Graehl and Knight, 2004; Liu et al., 2006; Huang et al., 2006), and dependency-based models (Lin, 2004; Ding and Palmer, 2005; Quirk et al., 2005; Xiong et al., 2007; Shen et al., 2008; Xie et al., 2011). These two kinds of models have their own advan- tages, as they capture different linguistic phenome- na. Constituency trees describe how words and se- quences of words combine to form constituents, and constituency-based models show better compatibil- ity with phrases. However, dependency trees de- scribe the grammatical relation between words of the sentence, and represent long distance dependen- cies in a concise manner. Dependency-based mod- els, such as dependency-to-string model (Xie et al., 2011), exhibit better capability of long distance re- orderings. In this paper, we propose to combine the advan- tages of source side constituency and dependency trees. Since the dependency tree is structurally sim- pler and directly represents long distance depen- dencies, we take dependency trees as the backbone and incorporate constituents to them. Our mod- el employs rules that represent the source side as head-dependents relations which are incorporated with constituency phrasal nodes, and the target side as strings. A head-dependents relation (Xie et al., 2011) is composed of a head and all its dependents in dependency trees, and it encodes phrase pattern and sentence pattern (typically long distance reordering relations). With the advantages of head-dependents relations, the translation rules of our model hold the property of long distance reorderings and the com- patibility with phrases. Our new model (Section 2) extracts rules from word-aligned pairs of source …

Content truncated to 500 words