Insertion Position Selection Model for Flexible Non-Terminals in Dependency Tree-to-Tree Machine Translation

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 2271–2277, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics Insertion Position Selection Model for Flexible Non-Terminals in Dependency Tree-to-Tree Machine Translation Toshiaki Nakazawa Japan Science and Technology Agency 5-3, Yonbancho, Chiyoda-ku, Tokyo, 102-8666, Japan nakazawa@pa.jst.jp John Richardson and Sadao Kurohashi Kyoto University Yoshida-honmachi, Sakyo-ku, Kyoto, 606-8501, Japan john@nlp.ist.i.kyoto-u.ac.jp kuro@i.kyoto-u.ac.jp Abstract Dependency tree-to-tree translation models are powerful because they can naturally han- dle long range reorderings which is important for distant language pairs. The translation pro- cess is easy if it can be accomplished only by replacing non-terminals in translation rules with other rules. However it is sometimes necessary to adjoin translation rules. Flex- ible non-terminals have been proposed as a promising solution for this problem. A flex- ible non-terminal provides several insertion position candidates for the rules to be ad- joined, but it increases the computational cost of decoding. In this paper we propose a neu- ral network based insertion position selection model to reduce the computational cost by selecting the appropriate insertion positions. The experimental results show the proposed model can select the appropriate insertion po- sition with a high accuracy. It reduces the de- coding time and improves the translation qual- ity owing to reduced search space. 1 Introduction Tree-to-tree machine translation models currently receive limited attention. However, we believe that using target-side syntax is important to achieve high- quality translations between distant language pairs which require long range reorderings. Especially, using dependency trees on both source and tar- get sides is promising for this purpose (Menezes and Quirk, 2007; Nakazawa and Kurohashi, 2010; Richardson et al., 2014). Tree-based translation models naturally realize word reorderings using the non-terminals or anchors for the attachment in the translation rules: therefore they do not need a re- ordering model which string-based models require. For example, suppose we have a translation rule with the word alignment shown in Figure 1, it is easy to translate a new input sentence which has “図書館 (library)” instead of “公園 (park)” because we can accomplish it by simply substituting “library” for the target word “park” without considering the reorder- ing. In this case, the source word “公園” and target word “park” work as the non-terminals. On the other hand, it is problematic when we need to adjoin a subtree which is not presented in train- ing sentences, which we call floating subtree in this paper. The floating subtrees are not necessarily ad- juncts, but any words or phrases. For example, sup- pose the Japanese input sentence in Figure 1 has “ 突然 (suddenly)”, but the training corpus provides only a translation rule without the word. In this case we cannot directly use the rule for the translation be- cause it does not know where to insert the translation of the floating word in the output. As another exam- ple, there is no context information available for the children of the OOV word in the input sentence, so we need some special …

Content truncated to 500 words