Elamur Research⌕
☰
home / papers / undated / Neural Japanese Zero Anaphora Resolution using Smoothed Large-scale Case Frames with Word Embedding

Neural Japanese Zero Anaphora Resolution using Smoothed Large-scale Case Frames with Word Embedding

Summary

Japanese texts often omit verb arguments, so a system has to recover these zero anaphors and link them to antecedents. Most earlier neural work handled only antecedents within the same sentence. The authors show that in the balanced corpus BCCWJ, roughly half of zero arguments lie in earlier sentences. Covering those cross-sentence cases makes the set of candidate antecedents enormous, which hurts both training and inference.

The authors filter candidates using Kyoto University Case Frames (KUCF) together with word embeddings. They then compare two resolvers on the reduced set: a ranking SVM based on Hangyo et al. (2013), and feed-forward networks with optional features. One FNN variant adds a local-attention RNN that encodes preceding sentences. The models treat nominative, accusative and dative cases together in one model. They are evaluated on the multi-genre BCCWJ core data rather than the newspaper-only NAIST Text Corpus. Case-frame pruning clearly beats a naive filter, and the FNN models do better on inter-sentential anaphora. The work shows that full-text zero anaphora resolution can be trained at scale and tested across genres.

Key points

  • Data: The experiments rely on the BCCWJ core data, a collection of 2,000 documents (about 2M words) spanning six genres: blogs, social QA, books, magazines, newspapers and white papers. It is annotated with predicate-argument structure and coreference and split roughly 80/5/15 into train/dev/test. Only 50.4% of the 46,091 zero arguments are intra-sentential.
  • Candidate reduction: The case-frame and embedding filter cuts candidates to about 1/1000 of the original number, leaving roughly 5 per slot on average. About 70% of correct answers survive the filtering.
  • Pruning result: Ranking SVM with the proposed pruning (S0) beat a baseline that keeps the 5 nearest preceding nouns (S0′) in every evaluation column. The abstract reports an accuracy gain of 0.056, and a McNemar test gave p < 0.001.
  • FNN feature effects: Adding argument and predicate embeddings (F1) lowered accuracy. Replacing the predicate embedding with case-frame-based vectors (F2, "MVC") improved results. The context RNN with Luong-style local attention (F3) gave a further small gain.
  • Overall comparison: F3 remains below S0 overall but is stronger on inter-sentential cases. The paper (PACLIC 32, 2018) covers verbs only; adjectives and event nouns are left for future work.

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: Coreference resolution

Cited by

8 papers of our library cite it (of the 140673 whose bibliographies we read); 4 of them have a page here, 4 not yet.

References

23 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Chen Chen and Vincent Ng. 2016. Chinese Zero Pronoun Resolution with Deep Neural Networks. In ACL, pages 778–788. find our card
  2. Masatsugu Hangyo, Daisuke Kawahara, and Sadao Kurohashi. 2012. Building a Diverse Document Leads Corpus Annotated with Semantic Relations. In PACLIC, pages 535–544. find
  3. Masatsugu Hangyo, Daisuke Kawahara, and Sadao Kurohashi. 2013. Japanese Zero Reference Resolution Considering Exophora and Author/Reader Mentions. In EMNLP, pages 924–934. find
  4. Ryu Iida, Mamoru Komachi, Kentaro Inui, and Yuji Matsumoto. 2007. Annotating a Japanese Text Corpus with Predicate-Argument and Coreference Relations. In Proceedings of the Linguistic Annotation Workshop, pages 132–139. find our card
  5. Ryu Iida and Massimo Poesio. 2011. A Cross-Lingual ILP Solution to Zero Anaphora Resolution. In ACL, pages 804–813. find our card
  6. Ryu Iida, Kentaro Torisawa, Jong-Hoon Oh, Canasai Kruengkrai, and Julien Kloetzer. 2016. Intra-Sentential Subject Zero Anaphora Resolution using Multi-Column Convolutional Neural Network. In EMNLP, pages 1244–1254. find our card
  7. Kenji Imamura, Kuniko Saito, and Tomoko Izumi. 2009. Discriminative Approach to Predicate-Argument Structure Analysiswith Zero-Anaphora Resolution. In ACL-IJCNLP, pages 85–88. find
  8. Thorsten Joachims. 2006. Training Linear SVMs in Linear Time. In Proceedings of the 12th ACM SIGKDD, pages 217–226. find
  9. Daisuke Kawahara and Sadao Kurohashi. 2006. A Fully-Lexicalized Probabilistic Model for Japanese Syntactic and Case Structure Analysis. In ACL, pages 176–183. find
  10. Taku Kudo, Hiroshi Ichikawa, and Hideto Kazawa. 2014. A joint inference of deep case analysis and zero subject generation for Japanese-to-English statistical machine translation. In ACL, pages 557–562. find
  11. Omer Levy and Yoav Goldberg. 2014. Dependency-Based Word Embeddings. In ACL, pages 302– 308. 777 32nd Pacific Asia Conference on Language, Information and Computation Hong Kong, 1-3 December 2018 Copyright 2018 by the authors Minh-Thang Luong, Hieu Pham, and Christopher D. Manning. 2015. Effective A… find
  12. Kikuo Maekawa, Makoto Yamazaki, Toshinobu Ogiso, Takehiko Maruyama, Hideki Ogura, Wakako Kashino, Hanae Koiso, Masaya Yamaguchi, Makiro Tanaka, and Yasuharu Den. 2014. Balanced corpus of contemporary written Japanese. Language Resources and Evaluation, 48(2):345–371. find
  13. Yuichiroh Matsubayashi and Kentaro Inui. 2017. Revisiting the Design Issues of Local Models for Japanese Predicate-Argument Structure Analysis. In IJCNLP, pages 128–133. find
  14. Tomas Mikolov, Kai Chen, Grag Corrado, and Jeffrey Dean. 2013. Efficient Estimation of Word Representations in Vector Space. arXiv preprint arXiv:1301.3781. find
  15. Hiroki Ouchi, Hiroyuki Shindo, Kevin Duh, and Yuji Matsumoto. 2015. Joint Case Argument Identification for Japanese Predicate Argument Structure Analysis. In ACL-IJCNLP, pages 961– 970. find
  16. Hiroki Ouchi, Hiroyuki Shindo, and Yuji Matsumoto. 2017. Neural Modeling of Multi-Predicate Interactions for Japanese Predicate Argument Structure Analysis. In ACL, pages 1591– 1600. find
  17. Luz Rello, Ricardo Baeza-Yates, and Ruslan Mitkov. 2012. Elliphant: Improved Automatic Detection of Zero Subjects and Impersonal Constructions in Spanish. In EACL, pages 706–715. find
  18. Ryohei Sasano, Daisuke Kawahara, and Sadao Kurohashi. 2008. A Fully-Lexicalized Probabilistic Model for Japanese Zero Anaphora Resolution. In COLING, pages 769–776. find our card
  19. Ryohei Sasano and Sadao Kurohashi. 2011. A Discriminative Approach to Japanese Zero Anaphora Resolution with Large-scale Lexicalized Case Frames. In IJCNLP, pages 758–766. find our card
  20. Tomohide Shibata, Daisuke Kawahara, and Sadao Kurohashi. 2016. Neural Network-Based Model for Japanese Predicate Argument Structure Analysis. In ACL, pages 1235–1244. find
  21. Kiyoshi Sudo, Satoshi Sekine, and Ralph Grishman. 2001. Automatic pattern acquisition for Japanese information extraction. In HLT, pages 1–7. find
  22. Hiroaki Yamada, Simone Teufel, and Takenobu Tokunaga. 2017. Designing an annotation scheme for summarizing Japanese judgment documents. In KSE, pages 275–280. find
  23. Qingyu Yin, Yu Zhang, Weinan Zhang, and Ting Liu. 2017. Chinese Zero Pronoun Resolution with Deep Memory Network. In EMNLP, pages 1309– 1318. 778 32nd Pacific Asia Conference on Language, Information and Computation Hong Kong, 1-3 December 2018 Copyright 2018 by the authors find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/coref--yamashiro-neural-japanese-zero-anaphora-resolution/