Elamur ResearchβŒ•
☰
home / papers / 2009 / Graph-Cut-Based Anaphoricity Determination for Coreference Resolution

Graph-Cut-Based Anaphoricity Determination for Coreference Resolution

Summary

This paper tackles anaphoricity determination: deciding which mentions have an antecedent, so that a coreference resolver can skip the non-anaphoric ones. Earlier work fell into two camps. Ng (2004) tuned a classifier threshold to maximize the coreference metric directly but ignored pairwise coreference probabilities. ILP-based joint inference used those probabilities but optimized an objective unrelated to the evaluation metric. Ng proposes a minimum-cut formulation that combines both strengths. It builds a graph from the outputs of a MaxEnt anaphoricity model and a MaxEnt pairwise coreference model. Two parameters are tuned on held-out data against the target metric. The method is compared with five existing approaches on three ACE data sets, scored with both MUC and CEAF. Unlike several prior studies, mentions come from an NP chunker rather than gold answer keys. The graph-cut system achieves the best F-score in all six data-set/scorer combinations, and none of the earlier methods wins consistently. This makes the paper a useful systematic benchmark of anaphoricity filtering under realistic, automatically extracted mentions.

Key points

  • Baselines compared: the five approaches are Ng and Cardie (2002a), Ng (2004), Luo (2007), Denis and Baldridge (2007, ILP), and Finkel and Manning (2008, ILP with transitivity constraints).
  • Models: the coreference model is MaxEnt with 33 features (Ng 2007), trained on Soon et al. (2001) instances with closest-first clustering. The anaphoricity model is MaxEnt with 37 features (Ng and Cardie 2002a).
  • Data and evaluation: the ACE data sets are Broadcast News, Newspaper and Newswire. One third of the training data was held out to tune the parameters c and c2, searched in steps of 0.1. Significance was tested with Approximate Randomization at p=0.05.
  • Main results: the graph-cut system reaches MUC F of 59.4–63.9 and CEAF F of 59.4–63.8. Over the no-anaphoricity baseline this is a significant gain in all six cases: +2.0–4.4 MUC and +2.6–4.1 CEAF. The improvement comes mainly from higher precision, with a smaller loss in recall.
  • Comparison with baselines: the graph-cut system significantly beats the strongest baseline (Denis and Baldridge) except on Newspaper/MUC and Newswire/MUC. Adding transitivity constraints (Finkel and Manning) never improved F-score over Denis and Baldridge.

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

Fields: Linguistics & languages

Topics: Coreference resolution

Cited by

17 papers of our library cite it (of the 140673 whose bibliographies we read); 10 of them have a page here, 7 not yet.

Cite

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

@inproceedings{ng2009graph,
  title = {{Graph-Cut-Based Anaphoricity Determination for Coreference Resolution}},
  author = {Ng, Vincent},
  booktitle = {Proceedings of Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2009},
  doi = {10.3115/1620754.1620838},
  url = {https://doi.org/10.3115/1620754.1620838}
}

References

30 references, 15 of them with a card of ours. β†— leads to the work, Β«findΒ» searches for one we have no id for yet.

  1. D. Bean and E. Riloff. 1999. Corpus-based identification of non-anaphoric noun phrases. In Proc. of the ACL, pages 373–380. find our card
  2. A. Berger, S. Della Pietra, and V. Della Pietra. 1996. A maximum entropy approach to natural language processing. Computational Linguistics, 22(1):39–71. find our card
  3. S. Bergsma, D. Lin, and R. Goebel. 2008. Distributional identification of non-referential pronouns. In Proc. of ACL-08:HLT, pages 10–18. find
  4. S. Della Pietra, V. Della Pietra, and J. Lafferty. 1997. Inducing features of random fields. IEEE Transactions on Pattern Analysis and Machine Intelligence, 19(4):380–393. find
  5. P. Denis and J. Baldridge. 2007. Global, joint determination of anaphoricity and coreference resolution using integer programming. In Proc. of NAACL/HLT, pages 236–243. find our card
  6. M. Elsner and E. Charniak. 2007. A generative discourse-new model for text coherence. Technical Report CS-07-04, Brown University. find
  7. R. Evans. 2001. Applying machine learning toward an automatic classification of it. Literary and Linguistic Computing, 16(1):45–57. find
  8. J. R. Finkel and C. Manning. 2008. Enforcing transitivity in coreference resolution. In Proc. of ACL-08:HLT Short Papers (Companion Volume), pages 45–48. find our card
  9. C. Kennedy and B. Boguraev. 1996. Anaphor for everyone: Pronominal anaphora resolution without a parser. In Proc. of COLING, pages 113–118. find
  10. S. Lappin and H. Leass. 1994. An algorithm for pronominal anaphora resolution. Computational Linguistics, 20(4):535–562. find our card
  11. X. Luo, A. Ittycheriah, H. Jing, N. Kambhatla, and S. Roukos. 2004. A mention-synchronous coreference resolution algorithm based on the Bell tree. In Proc. of the ACL, pages 135–142. find our card
  12. X. Luo. 2005. On coreference resolution performance metrics. In Proc. of HLT/EMNLP, pages 25–32. find our card
  13. X. Luo. 2007. Coreference or not: A twin model for coreference resolution. In Proc. of NAACL-HLT, pages 73–80. find our card
  14. C. Müller. 2006. Automatic detection of nonreferential it in spoken multi-party dialog. In Proc. of EACL, pages 49–56. find
  15. V. Ng. 2007. Shallow semantics for coreference resolution. In Proceedings of IJCAI, pages 1689–1694. find
  16. V. Ng and C. Cardie. 2002a. Identifying anaphoric and non-anaphoric noun phrases to improve coreference resolution. In Proc. of COLING, pages 730–736. find
  17. V. Ng and C. Cardie. 2002b. Improving machine learning approaches to coreference resolution. In Proc. of the ACL, pages 104–111. find our card
  18. V. Ng. 2004. Learning noun phrase anaphoricity to improve conference resolution: Issues in representation and optimization. In Proc. of the ACL, pages 151–158. find our card
  19. E. W. Noreen. 1989. Computer Intensive Methods for Testing Hypothesis: An Introduction. John Wiley & Sons. find
  20. C. Paice and G. Husk. 1987. Towards the automatic recognition of anaphoric features in English text: the impersonal pronoun ’it’. Computer Speech and Language, 2. find
  21. B. Pang and L. Lee. 2004. A sentimental education: Sentiment analysis using subjectivity summarization based on minimum cuts. In Proc. of the ACL, pages 271–278. find
  22. M. Poesio, O. Uryupina, R. Vieira, M. Alexandrov-Kabadjov, and R. Goulart. 2004. Discourse-new detectors for definite description resolution: A survey and a preliminary proposal. In Proc. of the ACL Workshop on Reference Resolution. find
  23. S. P. Ponzetto and M. Strube. 2006. Exploiting semantic role labeling, WordNet and Wikipedia for coreference resolution. In Proc. of HLT/NAACL, pages 192–199. find our card
  24. W. M. Soon, H. T. Ng, and D. Lim. 2001. A machine learning approach to coreference resolution of noun phrases. Computational Linguistics, 27(4):521–544. find our card
  25. O. Uryupina. 2003. High-precision identification of discourse new and unique noun phrases. In Proc. of the ACL Student Research Workshop. find
  26. Y. Versley, A. Moschitti, M. Poesio, and X. Yang. 2008. Coreference systems based on kernels methods. In Proc. of COLING, pages 961–968. find our card
  27. R. Vieira and M. Poesio. 2000. Processing definite descriptions in corpora. In S. Botley and A. McEnery, editors, Corpus-based and Computational Approaches to Discourse Anaphora, pages 189–212. UCL Press. find
  28. M. Vilain, J. Burger, J. Aberdeen, D. Connolly, and L. Hirschman. 1995. A model-theoretic coreference scoring scheme. In Proc. of MUC-6, pages 45–52. find our card
  29. B. L. Webber. 1979. A Formal Approach to Discourse Anaphora. Garland Publishing, Inc. find
  30. X. Yang and J. Su. 2007. Coreference resolution using semantic relatedness information from automatically discovered patterns. In Proc. of ACL, pages 528–535. find our card

↑ back to top

This page in Markdown β†’