Elamur Research⌕
☰
home / papers / 2012 / Using Syntactic Dependencies to Solve Coreferences

Using Syntactic Dependencies to Solve Coreferences

Summary

The paper presents the LTH coreference solver, developed for the CoNLL 2012 shared task's closed-track setting, which covered English, Chinese, and Arabic. It extends Björkelund and Nugues (2011)'s English system to all three languages. Instead of writing head-finding rules for each feature, the authors first convert every constituent tree into a generic dependency graph. They use the LTH converter for English, Penn2Malt for Chinese, and CATiB for Arabic, which required mapping simplified Arabic POS tags to the Penn Arabic tagset.

Candidate mentions are then paired, and a Soon et al. (2001)-style binary classifier decides whether each pair corefers. The classifier is trained with logistic regression in LIBLINEAR. The authors also redesigned the mention detection step and added new features: edit distance, speaker agreement, document genre, and a crude cluster model built from the first mention of each entity. Features were chosen per language through cross-validated forward and backward selection.

The system achieved an official combined score of 55.21. This suggests that dependency-derived features carry over well across typologically different languages.

Key points

  • Mention detection improvements (English, development set):

    • A pleonastic-it classifier is merged with the coreference probability using a threshold of 0.4. It raised the score from 59.15 to 59.57.
    • Adding named entities (excluding CARDINAL, ORDINAL, PERCENT, and QUANTITY) raised it from 58.85 to 59.57.
    • Together, the mention-stage extensions contributed +2.35, from 57.22 to 59.57.

  • Pruning same-head mentions: removing the smaller of two mentions that share a head helped English (56.42 → 59.57) and Chinese (50.94 → 56.62). It hurt Arabic (48.25 → 47.10), which the authors tentatively attribute to dependency conversion errors.

  • Decoding: pronouns use closest-first clustering. Other mentions use best-first clustering for Chinese and Arabic, and an averaged best-first variant for English, which scores whole chains by the geometric mean of pair probabilities.

  • CoNLL scores (mean of MUC, BCUB, CEAFE):

    LanguageDevelopmentTestTest with gold mentions
    English59.5759.3668.66
    Chinese56.6256.8566.97
    Arabic48.2549.4353.35

  • Gain over the 2011 system: on English, the score rose from 54.53 to 59.36 (+4.83), though the authors note the two results are not strictly comparable.

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 · Part-of-speech tagging

Cited by

5 papers of our library cite it (of the 140673 whose bibliographies we read); 2 of them have a page here, 3 not yet.

Cite

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

@inproceedings{stamborg2012using,
  title = {{Using Syntactic Dependencies to Solve Coreferences}},
  author = {Stamborg, Marcus and Medved, Dennis and Exner, Peter and Nugues, Pierre},
  booktitle = {Joint Conference on EMNLP and CoNLL - Shared Task},
  year = {2012},
  url = {https://aclanthology.org/W12-4505/}
}

References

9 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Anders Björkelund and Pierre Nugues. 2011. Exploring lexicalized features for coreference resolution. In Proceedings of the Fifteenth Conference on Computational Natural Language Learning: Shared Task, pages 45–50, Portland, Oregon, USA, June. Association for Computational Linguistics. find
  2. Rong-En Fan, Kai-Wei Chang, Cho-Jui Hsieh, Xiang-Rui Wang, and Chih-Jen Lin. 2008. LIBLINEAR: A library for large linear classification. Journal of Machine Learning Research, 9:1871–1874. find
  3. Nizar Habash and Ryan Roth. 2009. CATiB: The Columbia Arabic treebank. In Proceedings of the ACL-IJCNLP 2009 Conference Short Papers, pages 221–224, Suntec, Singapore, August. Association for Computational Linguistics. find our card
  4. Richard Johansson and Pierre Nugues. 2007. Extended constituent-to-dependency conversion for English. In Joakim Nivre, Heiki-Jaan Kaalep, Kadri Muischnek, and Mare Koit, editors, NODALIDA 2007 Conference Proceedings, pages 105–112, Tartu, May 25-26. find our card
  5. Vincent Ng and Claire Cardie. 2002. Improving machine laerning approaches to coreference resolution. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 104–111. find
  6. Joakim Nivre. 2006. Inductive Dependency Parsing. Springer, Dordrecht, The Netherlands. find
  7. Sameer Pradhan, Lance Ramshaw, Mitchell Marcus, Martha Palmer, Ralph Weischedel, and Nianwen Xue. 2011. CoNLL-2011 shared task: Modeling unrestricted coreference in OntoNotes. In Proceedings of the Fifteenth Conference on Computational Natural Language Learning: Shared Task, pages 1–27, Portland, Or… find our card
  8. Sameer Pradhan, Alessandro Moschitti, Nianwen Xue, Olga Uryupina, and Yuchen Zhang. 2012. CoNLL-2012 shared task: Modeling multilingual unrestricted coreference in OntoNotes. In Proceedings of the Sixteenth Conference on Computational Natural Language Learning (CoNLL 2012), Jeju, Korea. find our card
  9. Wee Meng Soon, Hwee Tou Ng, and Daniel Chung Yong Lim. 2001. A machine learning approach to coreference resolution of noun phrases. Computational Linguistics, 27(4):521–544. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/stamborg-2012-syntactic-dependencies-solve-coreferences/