Japanese Word Reordering Integrated with Dependency Parsing

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1186–1196, Dublin, Ireland, August 23-29 2014. Japanese Word Reordering Integrated with Dependency Parsing Kazushi Yoshida1,a) Tomohiro Ohno2,b) Yoshihide Kato3,c) Shigeki Matsubara1,d) 1Graduate School of Information Science, Nagoya University, Japan 2Information Technology Center, Nagoya University, Japan 3Information & Communications, Nagoya University, Japan a)yoshida@db.ss.is.nagoya-u.ac.jp b)ohno@nagoya-u.jp c)yoshihide@icts.nagoya-u.ac.jp d)matubara@nagoya-u.jp Abstract Although Japanese has relatively free word order, Japanese word order is not completely arbitrary and has some sort of preference. Since such preference is incompletely understood, even native Japanese writers often write Japanese sentences which are grammatically well-formed but not easy to read. This paper proposes a method for reordering words in a Japanese sentence so that the sentence becomes more readable. Our method can identify more suitable word order than conventional word reordering methods by concurrently performing dependency parsing and word reordering instead of sequentially performing the two processing steps. As the result of an experiment on word reordering using newspaper articles, we confirmed the effectiveness of our method. 1 Introduction Japanese has relatively free word order, and thus Japanese sentences which make sense can be written without having a strong awareness of word order. However, Japanese word order is not completely arbitrary and has some sort of preference. Since such preference is incompletely understood, even native Japanese writers often write Japanese sentences which are grammatically well-formed but not easy to read. The word reordering of such sentences enables the readability to be improved. There have been proposed some methods for reordering words in a Japanese sentence so that the sentence becomes easier to read (Uchimoto et al., 2000; Yokobayashi et al., 2004). In addition, there exist a lot of researches for estimating appropriate word order in various languages (Filippova and Strube, 2007; Harbusch et al., 2006; Kruijff et al., 2001; Ringger et al., 2004; Shaw and Hatzivassiloglou, 1999). Although most of these previous researches used syntactic information, the sentences they used there were what had been previously parsed. It is a problem that word reordering suffers the influence of parsing errors. Furthermore, as the related works, there are various researches on word reordering for improving the performance of statistical machine translation (Goto et al., 2012; Elming, 2008; Ge, 2010; Christoph and Hermann, 2003; Nizar, 2007). These researches consider information as to both a source language and a target language to handle word order differences between them. Therefore, their problem setting is different from that for improving the readability of a single language. This paper proposes a method for reordering words in a Japanese sentence so that the sentence becomes easier to read for revision support. Our proposed method concurrently performs dependency parsing and word reordering for an input sentence of which the dependency structure is still unknown. Our method can identify more suitable word order than conventional word reordering methods because it can concurrently consider the preference of both word order and dependency. An experiment using newspaper articles showed the effectiveness of our method. 2 Word Order and …

Content truncated to 500 words