Inducing a Discriminative Parser to Optimize Machine Translation Reordering

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 843–853, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Inducing a Discriminative Parser to Optimize Machine Translation Reordering Graham Neubig1,2, Taro Watanabe2, Shinsuke Mori1 1Graduate School of Informatics, Kyoto University Yoshida Honmachi, Sakyo-ku, Kyoto, Japan 2National Institute of Information and Communication Technology 3-5 Hikari-dai, Seika-cho, Soraku-gun, Kyoto, Japan Abstract This paper proposes a method for learning a discriminative parser for machine trans- lation reordering using only aligned par- allel text. This is done by treating the parser’s derivation tree as a latent variable in a model that is trained to maximize re- ordering accuracy. We demonstrate that efficient large-margin training is possible by showing that two measures of reorder- ing accuracy can be factored over the parse tree. Using this model in the pre-ordering framework results in significant gains in translation accuracy over standard phrase- based SMT and previously proposed unsu- pervised syntax induction methods. 1 Introduction Finding the appropriate word ordering in the target language is one of the most difficult prob- lems for statistical machine translation (SMT), particularly for language pairs with widely di- vergent syntax. As a result, there is a large amount of previous research that handles the problem of reordering through the use of im- proved reordering models for phrase-based SMT (Koehn et al., 2005), hierarchical phrase-based translation (Chiang, 2007), syntax-based trans- lation (Yamada and Knight, 2001), or pre- ordering (Xia and McCord, 2004). In particular, systems that use source- language syntax allow for the handling of long- distance reordering without large increases in The first author is now affiliated with the Nara Institute of Science and Technology. decoding time. However, these require a good syntactic parser, which is not available for many languages. In recent work, DeNero and Uszko- reit (2011) suggest that unsupervised grammar induction can be used to create source-sentence parse structure for use in translation as a part of a pre-ordering based translation system. In this work, we present a method for inducing a parser for SMT by training a discriminative model to maximize reordering accuracy while treating the parse tree as a latent variable. As a learning framework, we use online large-margin methods to train the model to directly minimize two measures of reordering accuracy. We pro- pose a variety of features, and demonstrate that learning can succeed when no linguistic informa- tion (POS tags or parse structure) is available in the source language, but also show that this lin- guistic information can be simply incorporated when it is available. Experiments find that the proposed model improves both reordering and translation accuracy, leading to average gains of 1.2 BLEU points on English-Japanese and Japanese-English translation without linguistic analysis tools, or up to 1.5 BLEU points when these tools are incorporated. In addition, we show that our model is able to effectively max- imize various measures of reordering accuracy, and that the reordering measure that we choose has a …

Content truncated to 500 words