Elamur Research⌕
☰
home / papers / 2007 / Coreference or Not: A Twin Model for Coreference Resolution

Coreference or Not: A Twin Model for Coreference Resolution

Summary

The paper tackles a frequently neglected part of coreference resolution: deciding when a mention has no antecedent at all, either because it is discourse-new or because it opens a new coreference chain. Earlier systems handled this by rejecting candidates whose best link score fell under a manually tuned threshold, or by training a separate anaphoricity classifier. Luo instead proposes a twin model with two jointly trained maximum-entropy (MaxEnt) components. The link model scores whether a mention belongs to a given entity. The creation model scores the probability that the mention starts a new entity. Because the creation decision depends on every preceding entity, its features would be costly to compute directly. The author therefore restricts feature forms so that creation features can be built from link-model feature values. Decoding is a beam search over a Bell Tree. On ACE 2005 data, the twin model matches or beats a thresholding baseline on both ACE-Value and entity F-measure without any free parameter to tune. This makes the approach easier to carry across tasks and datasets.

Key points

  • Each step of the coreference process either links the current mention to an existing entity, scored by the link model P_l, or creates a new entity, scored by the creation model P_c. A full outcome's score is the product of the scores of its steps.
  • Mentions are processed in a fixed order: names first, then nominals, then pronouns. The search uses a Bell Tree with a beam of about 20 hypotheses and a relative pruning threshold of 10⁻⁵.
  • Unlike the starting model of Luo et al. (2004), which reused link scores in an ad-hoc way, the creation model here is fully trained. Unlike the anaphoricity classifiers of Ng and Cardie (2002a) and Ng (2004), it is trained together with the link model.
  • The best baseline threshold depends on the evaluation metric. θ=0.3 is best for entity F-measure (84.7 F, 87.5 ACE-Value), while θ=0.5 is best for ACE-Value (81.1 F, 88.0 ACE-Value).
  • The twin model scores 84.8 entity F-measure and 88.4 ACE-Value, reaching the top result on both metrics at once with no threshold.

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

12 papers of our library cite it (of the 140673 whose bibliographies we read); 7 of them have a page here, 5 not yet.

Cite

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

@inproceedings{luo2007coreference,
  title = {{Coreference or Not: A Twin Model for Coreference Resolution}},
  author = {Luo, Xiaoqiang},
  booktitle = {Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics; Proceedings of the Main Conference},
  year = {2007},
  url = {https://aclanthology.org/N07-1010/}
}

References

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

  1. David L. Bean and Ellen Riloff. 1999. Corpus-based identification of non-anaphoric noun phrases. In Proc. ACL. find our card
  2. E.T. Bell. 1934. Exponential numbers. Amer. Math. Monthly, pages 411–419. find
  3. Adam L. Berger, Stephen A. Della Pietra, and Vincent J. Della Pietra. 1996. A maximum entropy approach to natural language processing. Computational Linguistics, 22(1):39–71, March. find our card
  4. Hal Daumé III and Daniel Marcu. 2005. A large-scale exploration of effective global features for a joint entity detection and tracking model. In Proc. of HLT and EMNLP, pages 97–104, Vancouver, British Columbia, Canada, October. Association for Computational Linguistics. find
  5. Joshua Goodman. 2002. Sequential conditional generalized iterative scaling. In Pro. of the 40th ACL. find
  6. Christopher Kennedy and Branimir Boguraev. 1996. Anaphora for everyone: Pronominal anaphora resolution without a parser. In Proceedings of COLING-96 (16th International Conference on Computational Linguistics), Copenhagen,DK. find our card
  7. Shalom Lappin and Herbert J. Leass. 1994. An algorithm for pronominal anaphora resolution. Computational Linguistics, 20(4), December. find our card
  8. Xiaoqiang Luo and Imed Zitouni. 2005. Multilingual coreference resolution with syntactic features. In Proc. of Human Language Technology (HLT)/Empirical Methods in Natural Language Processing (EMNLP). find
  9. Xiaoqiang Luo, Abe Ittycheriah, Hongyan Jing, Nanda Kambhatla, and Salim Roukos. 2004. A mention-synchronous coreference resolution algorithm based on the bell tree. In Proc. of ACL. find our card
  10. Vincent Ng and Claire Cardie. 2002a. Identifying anaphoric and non-anaphoric noun phrases to improve coreference resolution. In Proceedings of COLING. find
  11. Vincent Ng and Claire Cardie. 2002b. Improving machine learning approaches to coreference resolution. In Proc. of ACL, pages 104–111. find our card
  12. Vincent Ng. 2004. Learning noun phrase anaphoricity to improve conference resolution: Issues in representation and optimization. In Proceedings of the 42nd Meeting of the Association for Computational Linguistics (ACL’04), Main Volume, pages 151–158, Barcelona, Spain, July. find our card
  13. Cristina Nicolae and Gabriel Nicolae. 2006. BEST-CUT: A graph algorithm for coreference resolution. In Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing, pages 275–283, Sydney, Australia, July. Association for Computational Linguistics. find
  14. NIST. 2005. ACE 2005 evaluation. www.nist.gov/speech/tests/ace/ace05/index.htm. find
  15. 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 ACL 2004: Workshop on Reference Resolution and its Applications, pages 47–54, Barcelona, Spain, July. find
  16. Wee Meng Soon, Hwee Tou Ng, and Chung Yong Lim. 2001. A machine learning approach to coreference resolution of noun phrases. Computational Linguistics, 27(4):521–544. find our card
  17. Xiaofeng Yang, Guodong Zhou, Jian Su, and Chew Lim Tan. 2003. Coreference resolution using competition learning approach. In Proc. of the 41 st ACL. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/luo-2007-coreference-or-not-twin-model/