Elamur Research⌕
☰
home / papers / 2008 / Unsupervised Models for Coreference Resolution

Unsupervised Models for Coreference Resolution

Summary

Supervised coreference resolvers depend heavily on annotated corpora, which makes them hard to transfer to languages with little labelled data. Earlier attempts to bootstrap pairwise classifiers also drifted toward tagging nearly every pair as non-coreferent. Ng instead proposes a generative, document-level model that treats coreference as clustering learned with EM. Each document is represented through its mention pairs. Every pair is described by seven automatically computed features, split into three conditionally independent groups.

Enumerating all partitions is intractable, so the E-step keeps only the N most probable ones, found with Luo et al.'s (2004) Bell tree beam search. Learning is seeded with a single coreference-annotated document. The author also analyses Haghighi and Klein's (2007) Bayesian model and adds three fixes to it.

On ACE broadcast news and newswire data, scored with MUC and CEAF, the EM model clearly beats the original H&K model and roughly matches the modified one. It still trails a supervised resolver. The results suggest that constraint-aware unsupervised clustering is a workable path for low-resource coreference.

Key points

  • Features: the seven pair features fall into three groups.
    • Strong indicators: STR MATCH, ALIAS, APPOSITIVE.
    • Linguistic constraints: GENDER, NUMBER, SEM CLASS.
    • Mention type: NPTYPE.
    • The model's parameters are multinomials over each group, conditioned on whether the pair is coreferent, estimated with add-one smoothing.
  • Fixes to H&K: the three modifications are relaxed head generation, agreement constraints and pronoun-only salience. Together they raise H&K's F-score by 5–8% (MUC) and 7–12% (CEAF), mainly through better precision.
  • Gain over original H&K: the EM model outperforms it by 6–16% MUC F-score and 6–14% CEAF F-score.
  • True-mention results (MUC F): the EM model scores 62.8 on BNEWS and 55.7 on NWIRE. The fully supervised model scores 61.1 and 59.6.
  • Supervised comparison system: a C4.5 decision tree with the 34 features of Ng (2007), trained on the full ACE 2003 coreference training corpus and clustered with the Bell tree search.

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

26 papers of our library cite it (of the 140673 whose bibliographies we read); 16 of them have a page here, 10 not yet.

Cite

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

@inproceedings{ng2008unsupervised,
  title = {{Unsupervised Models for Coreference Resolution}},
  author = {Ng, Vincent},
  booktitle = {Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing},
  year = {2008},
  doi = {10.3115/1613715.1613795},
  url = {https://doi.org/10.3115/1613715.1613795}
}

References

23 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Avrim Blum and Tom Mitchell. 1998. Combining labeled and unlabeled data with co-training. In Proceedings of COLT, pages 92–100. find our card
  2. Colin Cherry and Shane Bergsma. 2005. An expectation maximization approach to pronoun resolution. In Proceedings of CoNLL, pages 88–95. find our card
  3. Hal Daumé III and Daniel Marcu. 2005. A large-scale exploration of effective global features for a joint entity detection and tracking model. In Proceedings of HLT/EMNLP, pages 97–104. find
  4. Arthur P. Dempster, Nan M. Laird, and Donald B. Rubin. 1977. Maximum likelihood from incomplete data via the EM algorithm. Journal of the Royal Statistical Society, Series B, 39(1):1–38. find
  5. Pascal Denis and Jason Baldridge. 2007. Global, joint determination of anaphoricity and coreference resolution using integer programming. In Proceedings of NAACL/HLT, pages 236–243. find our card
  6. Stuart Geman and Donald Geman. 1984. Stochastic relaxation, Gibbs distributions and the Bayesian restoration of images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 6:721–741. find
  7. Aria Haghighi and Dan Klein. 2007. Unsupervised coreference resolution in a nonparametric Bayesian model. In Proceedings of the ACL, pages 848–855. find our card
  8. Heng Ji, David Westbrook, and Ralph Grishman. 2005. Using semantic relations to refine coreference decisions. In Proceedings of HLT/EMNLP, pages 17–24. find our card
  9. Andrew Kehler, Douglas Appelt, Lara Taylor, and Aleksandr Simma. 2004. Competitive self-trained pronoun interpretation In Proceedings of HLT-NAACL 2004: Short Papers, pages 33–36. find
  10. Xiaoqiang Luo, Abe Ittycheriah, Hongyan Jing, Nanda Kambhatla, and Salim Roukos. 2004. A mention-synchronous coreference resolution algorithm based on the Bell tree. In Proceedings of the ACL, pages 135–142. find our card
  11. Xiaoqiang Luo. 2005. On coreference resolution performance metrics. In Proceedings of HLT/EMNLP, pages 25–32. find our card
  12. Christoph Müller, Stefan Rapp, and Michael Strube. 2002. Applying co-training to reference resolution. In Proceedings of the ACL, pages 352–359. find
  13. Vincent Ng. 2007. Shallow semantics for coreference resolution. In Proceedings of IJCAI, pages 1689– 1694. find
  14. Vincent Ng and Claire Cardie. 2002. Improving machine learning approaches to coreference resolution. In Proceedings of the ACL, pages 104–111. find our card
  15. Vincent Ng and Claire Cardie. 2003. Weakly supervised natural language learning without redundant views. In HLT-NAACL: Main Proceedings, pages 173–180. find
  16. Simone Paolo Ponzetto and Michael Strube. 2006. Exploiting semantic role labeling, WordNet and Wikipedia for coreference resolution. In Proceedings of HLT/NAACL, pages 192–199. find our card
  17. J. Ross Quinlan. 1993. C4.5: Programs for Machine Learning. Morgan Kaufmann. find our card
  18. 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
  19. Michael Strube, Stefan Rapp, and Christoph Müller. 2002. The influence of minimum edit distance on reference resolution. In Proceedings of EMNLP, pages 312–319. find
  20. Yee Whye Teh, Michael Jordan, Matthew Beal, and David Blei. 2006. Hierarchical Dirichlet processes. Journal of the American Statistical Association, 101(476):1527–1554. find
  21. Renata Vieira and Massimo Poesio. 2000. An empirically-based system for processing definite descriptions. Computational Linguistics, 26(4):539– 593. find
  22. Marc Vilain, John Burger, John Aberdeen, Dennis Connolly, and Lynette Hirschman. 1995. A model-theoretic coreference scoring scheme. In Proceedings of MUC-6, pages 45–52. find our card
  23. Xiaofeng Yang, GuoDong Zhou, Jian Su, and Chew Lim Tan. 2003. Coreference resolution using competitive learning approach. In Proceedings of the ACL, pages 176–183. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ng-2008-unsupervised-models-coreference-resolution/