Elamur Research⌕
☰
home / papers / 2018 / Universal Dependencies Version 2 for Japanese

Universal Dependencies Version 2 for Japanese

Summary

The paper describes how the UD Japanese team built Japanese resources under version 2 of the Universal Dependencies guidelines. Almost none of the data was re-annotated by hand. Instead, it was automatically converted from existing Japanese treebanks. The sources were bunsetsu dependencies, constituency trees and predicate–argument structure annotation. As a result, moving from the v1 scheme to v2 only required changing the conversion rules. The authors present five corpora that differ in origin and size. They also discuss the problems that came up when adapting Japanese to UD. First, Japanese is written without spaces, so it is unclear which unit should count as a syntactic word. Second, the guidelines do not specify whether POS tags should depend on the lexicon or on the context. Third, the scheme is designed for subject-prominent languages such as English, whereas Japanese is a topic-prominent language with strictly head-final word order. Fourth, the source texts are subject to copyright. The paper shows how a cross-linguistic standard is adapted to a typologically distant language. It also makes concrete proposals for improving UD.

Key points

  • Five resources, size in SUW tokens:
    • UD Japanese-BCCWJ — ~1 098K; newspapers, books, magazines, government white papers, Yahoo! Answers and Yahoo! Blogs; surface forms are hidden for copyright reasons.
    • UD Japanese-KTC — ~189K; newspaper texts; still in the v1 scheme as of February 2018.
    • UD Japanese-GSD — ~186K; Wikipedia; CC BY-NC-SA.
    • UD Japanese-PUD — ~26K; parallel corpus; CC BY-SA.
    • UD Japanese-Modern — ~14K; magazines from the Meiji–Taishō period; CC BY-NC-SA.
  • How the data was obtained:
    • BCCWJ and Modern were converted from the BCCWJ-DepPara bunsetsu annotation and the BCCWJ-PAS predicate–argument annotation.
    • KTC was converted from the NTT Japanese Phrase Structure Treebank.
    • GSD was segmented with an IBM tool and parsed with a bunsetsu parser. Version 2.0 was used in the CoNLL 2017 Shared Task, and v2.1 (November 2017) added manual corrections.
  • Segmentation: UniDic defines three levels of units — SUW, LUW and bunsetsu. SUW are obtained with MeCab, while LUW and bunsetsu are obtained with Comainu. All resources are currently built on SUW, although the authors consider this unit too fine-grained for syntactic annotation.
  • POS: UniDic tags for SUW are lexicon-based, while those for LUW are based on usage in context. For UD, the usage-dependent tags were chosen.
  • Typology: a topic-prominent language needs a separate label for the topic marker, while the expl label is not needed. The UD guidelines require the head of a coordinate structure to be placed first, which conflicts with the strictly head-final order of Japanese. The authors therefore propose adding a head-direction parameter to UD.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Topics: Universal Dependencies · Treebanks

Cited by

24 papers of our library cite it (of the 140673 whose bibliographies we read); 1 of them have a page here, 23 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{asahara2018universal,
  title = {{Universal Dependencies Version 2 for Japanese}},
  author = {Asahara, Masayuki and Kanayama, Hiroshi and Tanaka, Takaaki and Miyao, Yusuke and Uematsu, Sumire and Mori, Shinsuke and Matsumoto, Yuji and Omura, Mai and Murawaki, Yugo},
  booktitle = {Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)},
  year = {2018},
  doi = {10.63317/4uskraiweqvd},
  url = {https://doi.org/10.63317/4uskraiweqvd}
}

References

16 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Asahara, M. and Matsumoto, Y. (2016). BCCWJ-DepPara: A Syntactic Annotation Treebank on the ‘Balanced Corpus of Contemporary Written Japanese’. In Proceedings of the 12th Workshop on Asian Langauge Resources (ALR12), pages 49–58. find our card
  2. Haspelmath, M. (2011). The indeterminacy of word segmentation and the nature of morphology and syntax. Folia Linguistica, 45(1):31–80. find
  3. Hattori, S., (1960). Gengogaku no hoho, chapter Fuzokugo to Fuzokukeishiki, pages 461–491. Iwanami Shoten. (in Japanese). find
  4. Iida, R., Komachi, M., Inui, K., and Matsumoto, Y. (2007). Annotating a Japanese Text Corpus with Predicate-argument and Coreference Relations. In LAW ’07: Proceedings of the Linguistic Annotation Workshop. find our card
  5. Kanayama, H., Torisawa, K., Mitsuishi, Y., and Tsujii, J. (2000). A hybrid Japanese parser with hand-crafted grammar and statistics. In Proceedings of the 18th International Conference on Computational Linguistics, pages 411–417. find
  6. Kurohashi, S. and Nagao, M., (2003). Building a Japanese Parsed Corpus – while Improving the Parsing System, chapter 14, pages 249–260. find
  7. Kluwer Academic Publishers. Maekawa, K., Yamazaki, M., Ogiso, T., Maruyama, T., Ogura, H., Kashino, W., Koiso, H., Yamaguchi, M., Tanaka, M., and Den, Y. (2014). Balanced Corpus of Contemporary Written Japanese. Language Resources and Evaluation, 48:345–371. find
  8. Masuoka, T. and Takubo, Y. (1992). Kiso Nihongo Bunpou – Kaitei Ban. Kuroshio Publisher. McDonald, R., Nivre, J., Quirmbach-Brundage, Y., Goldberg, Y., Das, D., K.Ganchev, Hall, K., Petrov, S., Zhang, H., Täckström, O., Bedini, C., Castelló, N. B., and Lee, J. (2013). Universal dependency annotat… find
  9. Miyaoka, O. (2015). Go to wa Nani ka Saiko (Reconsidering What is the “Word”?). Sanseido. (in Japanese). find
  10. Ogiso, T., Kondo, A., Mabuchi, Y., and Hattori, N. (2017). Construction of the “Corpus of Historical Japanese: Meiji-Taisho Series I - Magazines”. In Proceedings of Digital Humanities 2017. find
  11. Omura, M., Takahashi, Y., and Asahara, M. (2017). Universal Dependency for Modern Japanese. In Proceedings of the 7th Conference of Japanese Association for Digital Humanities (JADH2017). find
  12. Tanaka, T. and Nagata, M. (2013). Constructing a Practical Constituent Parser from a Japanese Treebank with Function Labels. In Proceedings of the Fourth Workshop on Statistical Parsing of Morphologically Rich Language, pages 108–118. find
  13. Tanaka, T., Miyao, Y., Asahara, M., Uematsu, S., Kanayama, H., Mori, S., and Matsumoto, Y. (2016). Universal Dependencies for Japanese. In Proceedings of LREC 2016. find
  14. Uchimoto, K. and Den, Y. (2008). Word-level dependency-structure annotation to corpus of spontaneous japanese and its application. In Proceedings of LREC 2008. find
  15. Ueda, Y., Iida, R., Asahara, M., Matsumoto, Y., and Tokunaga, T. (2015). A Predicate-Argument and Coreference Relation Annotation on the ‘Balanced Corpus of Contemporary Written Japanese’, gendai nihongo kakikotoba kinkou koopasu ni taisuru jutugo koukouzou kyosanshou kankei annoteeshon (in Japanese… find
  16. CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies. In Proceedings of the CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies. Association for Computational Linguistics. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/asahara-2018-universal-dependencies-version-2-japanese/