Head Finalization: A Simple Reordering Rule for SOV Languages

Proceedings of the Joint 5th Workshop on Statistical Machine Translation and MetricsMATR, pages 244–251, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics Head Finalization: A Simple Reordering Rule for SOV Languages Hideki Isozaki, Katsuhito Sudoh, Hajime Tsukada, Kevin Duh NTT Communication Science Laboratories, NTT Corporation 2-4 Hikaridai, Seikacho, Sorakugun, Kyoto, 619-0237, Japan {isozaki,sudoh,tsukada,kevinduh}@cslab.kecl.ntt.co.jp Abstract English is a typical SVO (Subject-Verb- Object) language, while Japanese is a typ- ical SOV language. Conventional Statis- tical Machine Translation (SMT) systems work well within each of these language families. However, SMT-based translation from an SVO language to an SOV lan- guage does not work well because their word orders are completely different. Re- cently, a few groups have proposed rule- based preprocessing methods to mitigate this problem (Xu et al., 2009; Hong et al., 2009). These methods rewrite SVO sen- tences to derive more SOV-like sentences by using a set of handcrafted rules. In this paper, we propose an alternative single re- ordering rule: Head Finalization. This is a syntax-based preprocessing approach that offers the advantage of simplicity. We do not have to be concerned about part- of-speech tags or rule weights because the powerful Enju parser allows us to imple- ment the rule at a general level. Our ex- periments show that its result, Head Final English (HFE), follows almost the same order as Japanese. We also show that this rule improves automatic evaluation scores. 1 Introduction Statistical Machine Translation (SMT) is useful for building a machine translator between a pair of languages that follow similar word orders. How- ever, SMT does not work well for distant language pairs such as English and Japanese, since English is an SVO language and Japanese is an SOV lan- guage. Some existing methods try to solve this word- order problem in language-independent ways. They usually parse input sentences and learn a re- ordering decision at each node of the parse trees. For example, Yamada and Knight (2001), Quirk et al. (2005), Xia and McCord (2004), and Li et al. (2007) proposed such methods. Other methods tackle this problem in language- dependent ways (Katz-Brown and Collins, 2008; Collins et al., 2005; Nguyen and Shimazu, 2006). Recently, Xu et al. (2009) and Hong et al. (2009) proposed rule-based preprocessing methods for SOV languages. These methods parse input sen- tences and reorder the words using a set of hand- crafted rules to get SOV-like sentences. If we could completely reorder the words in in- put sentences by preprocessing to match the word order of the target language, we would be able to greatly reduce the computational cost of SMT sys- tems. In this paper, we introduce a single reordering rule: Head Finalization. We simply move syntac- tic heads to the end of the corresponding syntactic constituents (e.g., phrases and clauses). We use only this reordering rule, and we do not have to consider part-of-speech tags or rule weights be- cause the powerful Enju parser allows us to im- plement the rule at a general level. Why do …

Content truncated to 500 words