Part-of-Speech Induction in Dependency Trees for Statistical Machine Translation

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 841–851, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Part-of-Speech Induction in Dependency Trees for Statistical Machine Translation Akihiro Tamura†,‡, Taro Watanabe†, Eiichiro Sumita†, Hiroya Takamura‡, Manabu Okumura‡ † National Institute of Information and Communications Technology {akihiro.tamura, taro.watanabe, eiichiro.sumita}@nict.go.jp † Precision and Intelligence Laboratory, Tokyo Institute of Technology {takamura, oku}@pi.titech.ac.jp Abstract This paper proposes a nonparametric Bayesian method for inducing Part-of- Speech (POS) tags in dependency trees to improve the performance of statistical machine translation (SMT). In particular, we extend the monolingual infinite tree model (Finkel et al., 2007) to a bilin- gual scenario: each hidden state (POS tag) of a source-side dependency tree emits a source word together with its aligned tar- get word, either jointly (joint model), or independently (independent model). Eval- uations of Japanese-to-English translation on the NTCIR-9 data show that our in- duced Japanese POS tags for dependency trees improve the performance of a forest- to-string SMT system. Our independent model gains over 1 point in BLEU by re- solving the sparseness problem introduced in the joint model. 1 Introduction In recent years, syntax-based SMT has made promising progress by employing either depen- dency parsing (Lin, 2004; Ding and Palmer, 2005; Quirk et al., 2005; Shen et al., 2008; Mi and Liu, 2010) or constituency parsing (Huang et al., 2006; Liu et al., 2006; Galley et al., 2006; Mi and Huang, 2008; Zhang et al., 2008; Cohn and Blunsom, 2009; Liu et al., 2009; Mi and Liu, 2010; Zhang et al., 2011) on the source side, the target side, or both. However, dependency parsing, which is a popular choice for Japanese, can incorporate only shallow syntactic information, i.e., POS tags, compared with the richer syntactic phrasal cate- gories in constituency parsing. Moreover, exist- ing POS tagsets might not be optimal for SMT because they are constructed without considering the language in the other side. Consider the ex- amples in Figure 1. The Japanese noun “利用” in 私 が 利用 利用 利用 利用 料金 を 払う あなた は インターネット が 利用 利用 利用 利用 でき ない You can not use the Internet . I pay usage fees . noun particle particle noun noun verb auxiliary verb noun particle noun noun verb particle [Example 1] [Example 2] Japanese POS: Japanese POS: Figure 1: Examples of Existing Japanese POS Tags and Dependency Structures Example 1 corresponds to the English verb “use”, while that in Example 2 corresponds to the English noun “usage”. Thus, Japanese nouns act like verbs in English in one situation, and nouns in English in another. If we could discriminate POS tags for two cases, we might improve the performance of a Japanese-to-English SMT system. In the face of the above situations, this pa- per proposes an unsupervised method for inducing POS tags for SMT, and aims to improve the perfor- mance of syntax-based SMT by utilizing the in- duced POS tagset. The proposed method is based on the infinite tree …

Content truncated to 500 words