Elamur Research⌕
☰
home / papers / 2010 / The Same-Head Heuristic for Coreference

The Same-Head Heuristic for Coreference

Summary

Many coreference systems, unsupervised ones in particular, assume that two full NPs sharing a head noun refer to the same entity. Elsner and Charniak test this assumption on the MUC-6 corpus. They compare a "link-all" heuristic with an oracle across three mention-detection settings: gold mentions, parser-detected base NPs, and all nouns. With gold mentions, few same-head pairs fail to corefer, because annotators rarely mark the problematic NPs. With realistic detection, about half of the same-head pairs are not coreferent. The authors also propose an unsupervised generative model, a mixture of two IBM model 2 variants. It is trained with EM, starts from a pronoun-alignment model, and uses syntactic features to decide which same-head NPs should stay unlinked. In every setting it raises precision but lowers recall. This lowers b³ but improves mention CEAF when mentions are detected by the parser. The authors argue that the field should report several metrics and use realistic mention detection, since b³ can hide the cost of this heuristic.

Key points

  • How common the errors are: In the nps setting, only 864 of the 1592 same-head pairs found within a 10-sentence window are truly coreferent. As a result, link-all b³ precision falls to 67.2 (nps) and 56.6 (nouns), compared with 80.6 for gold mentions.
  • What the errors look like: A manual check of 100 non-coreferent pairs found:
    • 39 distinct entities told apart by modifiers;
    • 24 time or measure phrases;
    • 12 partitives or quantities;
    • 12 generics;
    • 9 annotator errors;
    • 4 proper-name or abbreviation cases.
  • Model design: The model builds on the Charniak and Elsner (2009) pronoun model, recast as a log-linear alignment. The features cover distance, syntactic role, the parent nonterminal, modifiers, and determiner type. A uniform-alignment "non-coreferent" component is mixed in with a weight of λ = 0.9, the system's only hand-set parameter.
  • Precision gains: The model reaches b³ precision of 87.2 (nps) and 83.0 (nouns), at the cost of lower recall. It is too conservative with gold mentions.
  • Mentions and parsing: Mentions are extracted from parse trees generated by Charniak's parser in its self-trained version (McClosky et al., 2006).

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

6 papers of our library cite it (of the 140673 whose bibliographies we read); 2 of them have a page here, 4 not yet.

Cite

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

@inproceedings{elsner2010same,
  title = {{The Same-Head Heuristic for Coreference}},
  author = {Elsner, Micha and Charniak, Eugene},
  booktitle = {Proceedings of the ACL 2010 Conference Short Papers},
  year = {2010},
  url = {https://aclanthology.org/P10-2007/}
}

References

13 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Amit Bagga and Breck Baldwin. 1998. Algorithms for scoring coreference chains. In LREC Workshop on Linguistics Coreference, pages 563–566. find our card
  2. Eugene Charniak and Micha Elsner. 2009. EM works for pronoun anaphora resolution. In Proceedings of EACL, Athens, Greece. find our card
  3. Colin Cherry and Shane Bergsma. 2005. An Expectation Maximization approach to pronoun resolution. In Proceedings of CoNLL, pages 88–95, Ann Arbor, Michigan. find our card
  4. Kenneth W. Church. 2000. Empirical estimates of adaptation: the chance of two Noriegas is closer to p/2 than p 2 . In Proceedings of ACL, pages 180–186. find
  5. Aria Haghighi and Dan Klein. 2007. Unsupervised coreference resolution in a nonparametric Bayesian model. In Proceedings of ACL, pages 848–855. find our card
  6. Aria Haghighi and Dan Klein. 2009. Simple coreference resolution with rich syntactic and semantic features. In Proceedings of EMNLP, pages 1152–1161. find our card
  7. Percy Liang and Dan Klein. 2009. Online EM for unsupervised models. In HLT-NAACL. find
  8. Xiaoqiang Luo. 2005. On coreference resolution performance metrics. In Proceedings of HLT-EMNLP, pages 25–32, Morristown, NJ, USA. Association for Computational Linguistics. find our card
  9. David McClosky, Eugene Charniak, and Mark Johnson. 2006. Effective self-training for parsing. In Proceedings of HLT-NAACL, pages 152–159. find our card
  10. Vincent Ng. 2008. Unsupervised models for coreference resolution. In Proceedings of EMNLP, pages 640–649, Honolulu, Hawaii. Association for Computational Linguistics. find our card
  11. Hoifung Poon and Pedro Domingos. 2008. Joint unsupervised coreference resolution with Markov Logic. In Proceedings of EMNLP, pages 650–659, Honolulu, Hawaii, October. Association for Computational Linguistics. find our card
  12. Veselin Stoyanov, Nathan Gilbert, Claire Cardie, and Ellen Riloff. 2009. Conundrums in noun phrase coreference resolution: Making sense of the state-of-the-art. In Proceedings of ACL-IJCNLP, pages 656–664, Suntec, Singapore, August. Association for Computational Linguistics. find our card
  13. David Vadas and James Curran. 2007. Adding noun phrase structure to the penn treebank. In Proceedings of ACL, pages 240–247, Prague, Czech Republic, June. Association for Computational Linguistics. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/elsner-2010-same-head-heuristic-coreference/