Factored Soft Source Syntactic Constraints for Hierarchical Machine Translation

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 556–566, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Factored Soft Source Syntactic Constraints for Hierarchical Machine Translation Zhongqiang Huang Raytheon BBN Technologies 50 Moulton St Cambridge, MA, USA zhuang@bbn.com Jacob Devlin Raytheon BBN Technologies 50 Moulton St Cambridge, MA, USA jdevlin@bbn.com Rabih Zbib Raytheon BBN Technologies 50 Moulton St Cambridge, MA, USA rzbib@bbn.com Abstract This paper describes a factored approach to incorporating soft source syntactic constraints into a hierarchical phrase-based translation system. In contrast to traditional approaches that directly introduce syntactic constraints to translation rules by explicitly decorating them with syntactic annotations, which often ex- acerbate the data sparsity problem and cause other problems, our approach keeps transla- tion rules intact and factorizes the use of syn- tactic constraints through two separate mod- els: 1) a syntax mismatch model that asso- ciates each nonterminal of a translation rule with a distribution of tags that is used to measure the degree of syntactic compatibil- ity of the translation rule on source spans; 2) a syntax-based reordering model that predicts whether a pair of sibling constituents in the constituent parse tree of the source sentence should be reordered or not when translated to the target language. The features produced by both models are used as soft constraints to guide the translation process. Experiments on Chinese-English translation show that the proposed approach significantly improves a strong string-to-dependency translation sys- tem on multiple evaluation sets. 1 Introduction Hierarchical phrase-based translation models (Chi- ang, 2007) are widely used in machine translation systems due to their ability to achieve local flu- ency through phrasal translation and handle non- local phrase reordering using synchronous context- free grammars. A large number of previous works have tried to introduce grammaticality to the trans- lation process by incorporating syntactic constraints into hierarchical translation models. Despite some differences in the granularity of syntax units (e.g., tree fragments (Galley et al., 2004; Liu et al., 2006), treebank tags (Shen et al., 2008; Chiang, 2010), and extended tags (Zollmann and Venugopal, 2006)), most previous work incorporates syntax into hier- archical translation models by explicitly decorating translation rules with syntactic annotations. These approaches inevitably exacerbate the data sparsity problem and cause other problems such as increased grammar size, worsened derivational ambiguity, and unavoidable parsing errors (Hanneman and Lavie, 2013). In this paper, we propose a factored approach that incorporates soft source syntactic constraints into a hierarchical string-to-dependency translation model (Shen et al., 2008). The general ideas are ap- plicable to other hierarchical models as well. Instead of enriching translation rules with explicit syntactic annotations, we keep the original translation rules intact, and factorize the use of source syntactic con- straints through two separate models. The first is a syntax mismatch model that intro- duces source syntax into the nonterminals of transla- tion rules, and measures the degree of syntactic com- patibility between a translation rule and the source spans it is applied to during decoding. When a …

Content truncated to 500 words