Syntax-Augmented Machine Translation using Syntax-Label Clustering

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 165–171, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Syntax-Augmented Machine Translation using Syntax-Label Clustering Hideya Mino, Taro Watanabe and Eiichiro Sumita National Institute of Information and Communications Technology 3-5 Hikaridai, Seika-cho, Soraku-gun, Kyoto, JAPAN {hideya.mino, taro.watanabe, eiichiro.sumita}@nict.go.jp Abstract Recently, syntactic information has helped significantly to improve statistical ma- chine translation. However, the use of syn- tactic information may have a negative im- pact on the speed of translation because of the large number of rules, especially when syntax labels are projected from a parser in syntax-augmented machine translation. In this paper, we propose a syntax-label clus- tering method that uses an exchange algo- rithm in which syntax labels are clustered together to reduce the number of rules. The proposed method achieves clustering by directly maximizing the likelihood of synchronous rules, whereas previous work considered only the similarity of proba- bilistic distributions of labels. We tested the proposed method on Japanese-English and Chinese-English translation tasks and found order-of-magnitude higher cluster- ing speeds for reducing labels and gains in translation quality compared with pre- vious clustering method. 1 Introduction In recent years, statistical machine translation (SMT) models that use syntactic information have received significant research attention. These models use syntactic information on the source side (Liu et al., 2006; Mylonakis and Sima’an, 2011), the target side (Galley et al., 2006; Huang and Knight, 2006) or both sides (Chiang, 2010; Hanneman and Lavie, 2013) produce syntactically correct translations. Zollmann and Venugopal (2006) proposed syntax-augmented MT (SAMT), which is a MT system that uses syntax labels of a parser. The SAMT grammar directly encodes syn- tactic information into the synchronous context- free grammar (SCFG) of Hiero (Chiang, 2007), which relies on two nonterminal labels. One prob- lem in adding syntax labels to Hiero-style rules is that only partial phrases are assigned labels. It is common practice to extend labels by us- ing the idea of combinatory categorial grammar (CCG) (Steedman, 2000) on the problem. Al- though this extended syntactical information may improve the coverage of rules and syntactic cor- rectness in translation, the increased grammar size causes serious speed and data-sparseness prob- lems. To address these problems, Hanneman and Lavie (2013) coarsen syntactic labels using the similarity of the probabilistic distributions of la- bels in synchronous rules and showed that perfor- mance improved. In the present work, we follow the idea of label- set coarsening and propose a new method to group syntax labels. First, as an optimization criterion, we use the logarithm of the likelihood of syn- chronous rules instead of the similarity of prob- abilistic distributions of syntax labels. Second, we use exchange clustering (Uszkoreit and Brants, 2008), which is faster than the agglomerative- clustering algorithm used in the previous work. We tested our proposed method on Japanese- English and Chinese-English translation tasks and observed gains comparable to those of previous work with similar reductions in grammar size. 2 Syntax-Augmented Machine Translation SAMT is an …

Content truncated to 500 words