Augmenting String-to-Tree Translation Models with Fuzzy Use of Source-side Syntax

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 204–215, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Augmenting String-to-Tree Translation Models with Fuzzy Use of Source-side Syntax Jiajun Zhang, Feifei Zhai and Chengqing Zong Institute of Automation, Chinese Academy of Sciences Beijing, China {jjzhang, ffzhai, cqzong}@nlpr.ia.ac.cn Abstract Due to its explicit modeling of the grammaticality of the output via target-side syntax, the string-to-tree model has been shown to be one of the most successful syntax-based translation models. However, a major limitation of this model is that it does not utilize any useful syntactic information on the source side. In this paper, we analyze the difficulties of incorporating source syntax in a string-to- tree model. We then propose a new way to use the source syntax in a fuzzy manner, both in source syntactic annotation and in rule matching. We further explore three algorithms in rule matching: 0-1 matching, likelihood matching, and deep similarity matching. Our method not only guarantees grammatical output with an explicit target tree, but also enables the system to choose the proper translation rules via fuzzy use of the source syntax. Our extensive experiments have shown significant improvements over the state-of-the-art string-to-tree system. 1 Introduction In recent years, statistical translation models based upon linguistic syntax have shown promising progress in improving translation quality. It appears that encoding syntactic annotations on either side or both sides in translation rules can increase the expressiveness of rules and can produce more accurate translations with improved reordering. One of the most successful syntax-based models is the string-to-tree model (Galley et al., 2006; Marcu et al., 2006; Shen et al., 2008; Chiang et al., 2009). Since it explicitly models the grammaticality of the output via target-side syntax, the string-to-tree model (Xiao et al., 2010) significantly outperforms both the state-of-the-art phrase-based system Moses (Koehn et al., 2007) and the formal syntax-based system Hiero (Chiang, 2007). However, there is a major limitation in the string-to-tree model: it does not utilize any useful source-side syntactic information, and thus to some extent lacks the ability to distinguish good translation rules from bad ones. The source syntax is well-known to be helpful in improving translation accuracy, as shown especially by tree-to-string systems (Quirk et al., 2005; Liu et al., 2006; Huang et al., 2006; Mi et al., 2008; Zhang et al., 2009). The tree-to-string systems are simple and efficient, but they also have a major limitation: they cannot guarantee the grammaticality of the translation output because they lack target-side syntactic constraints. Thus a promising solution is to combine the advantages of the tree-to-string and string-to-tree approaches. A natural idea is the tree-to-tree model (Ding and Palmer, 2005; Cowan et al., 2006; Liu et al., 2009). However, as discussed by Chiang (2010), while tree-to-tree translation is indeed promising in theory, in practice it usually ends up over-constrained. Alternatively, Mi and Liu (2010) proposed to enhance the tree-to-string model with target dependency structures (as a language model). In this paper, we …

Content truncated to 500 words