Elamur Research
☰
home / papers / 2008 / Understanding the Value of Features for Coreference Resolution

Understanding the Value of Features for Coreference Resolution

Written for the Research library (claude-opus-5-5) from the paper text, in our own words: the license "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)" does not allow redistributing the text itself. Original — aclanthology.org/D08-1031.pdf

Summary

Most coreference research before this paper focused on new model structures and usually compared them against weak pairwise baselines. Bengtson and Roth argue that carefully engineered features may matter as much as the model. They build a deliberately simple pairwise classifier: an averaged perceptron, trained in Learning Based Java, scores pairs of mentions. The Best-Link strategy then attaches each mention to its highest-scoring preceding candidate if that score clears a threshold. Pronouns are not allowed to serve as antecedents for non-pronominal mentions. The features cover string matching, semantic agreement based on WordNet and word lists, apposition, distance, aligned modifiers, memorization, and predicted entity types, and all pairwise conjunctions of these are added. Trained and tested on ACE data, the system beats the more complex cluster-level model of Culotta et al. (2007) by 1.5 B-Cubed F points. The authors offer it as a strong baseline for checking whether structural modeling innovations still pay off. An ablation study and an end-to-end version that runs on raw text are also included.

Key points

  • On the development set, adding feature groups one at a time raises B-Cubed F from 75.76 to 81.79. Distance, apposition and entity-type features give the largest statistically significant gains.
  • Predicted entity types are highly discriminative. When types agree, 13% of mention pairs are coreferent, against about 6% overall. Pairs that fire the relative-pronoun feature are positive 93% of the time.
  • For training, each mention is paired with only its closest preceding coreferent mention as a positive example. Every earlier non-coreferent mention gives a negative example.
  • The end-to-end pipeline has a chunking-based head detector (about 90% F1), an extent classifier and a mention-type classifier. On unseen ACE test data it reaches P 84.91, R 72.53 and B-Cubed F 78.24.
  • The full feature package and the LBJ code were released publicly by the Cognitive Computation Group at UIUC.

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

98 papers of our library cite it (of the 140673 whose bibliographies we read); 19 of them have a page here, 79 not yet.

Cite

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

@inproceedings{bengtson2008understanding,
  title = {{Understanding the Value of Features for Coreference Resolution}},
  author = {Bengtson, Eric and Roth, Dan},
  booktitle = {Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing},
  year = {2008},
  doi = {10.3115/1613715.1613756},
  url = {https://doi.org/10.3115/1613715.1613756}
}

References

15 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. A. Bagga and B. Baldwin. 1998. Algorithms for scoring coreference chains. In MUC7. find our card
  2. A. Culotta, M. Wick, R. Hall, and A. McCallum. 2007. First-order probabilistic models for coreference resolution. In HLT/NAACL, pages 81–88. find
  3. P. Denis and J. Baldridge. 2007. Joint determination of anaphoricity and coreference resolution using integer programming. In HLT/NAACL, pages 236–243, Rochester, New York. find
  4. C. Fellbaum. 1998. WordNet: An Electronic Lexical Database. MIT Press. find
  5. Y. Freund and R. E. Schapire. 1998. Large margin classification using the Perceptron algorithm. In COLT, pages 209–217. find
  6. H. Ji, D. Westbrook, and R. Grishman. 2005. Using semantic relations to refine coreference decisions. In EMNLP/HLT, pages 17–24, Vancouver, British Columbia, Canada. find
  7. X. Luo and I. Zitouni. 2005. Multi-lingual coreference resolution with syntactic features. In HLT/EMNLP, pages 660–667, Vancouver, British Columbia, Canada. find
  8. X. Luo, A. Ittycheriah, H. Jing, N. Kambhatla, and S. Roukos. 2004. A mention-synchronous coreference resolution algorithm based on the bell tree. In ACL, page 135, Morristown, NJ, USA. find
  9. V. Ng and C. Cardie. 2002a. Identifying anaphoric and non-anaphoric noun phrases to improve coreference resolution. In COLING-2002. find
  10. V. Ng and C. Cardie. 2002b. Improving machine learning approaches to coreference resolution. In ACL. find our card
  11. NIST. 2004. The ace evaluation plan. www.nist.gov/speech/tests/ace/index.htm. find
  12. V. Punyakanok and D. Roth. 2001. The use of classifiers in sequential inference. In The Conference on Advances in Neural Information Processing Systems (NIPS), pages 995–1001. MIT Press. find
  13. N. Rizzolo and D. Roth. 2007. Modeling Discriminative Global Inference. In Proceedings of the First International Conference on Semantic Computing (ICSC), pages 597–604, Irvine, California. find
  14. W. M. Soon, H. T. Ng, and C. Y. Lim. 2001. A machine learning approach to coreference resolution of noun phrases. Computational Linguistics, 27(4):521– 544. find our card
  15. M. Vilain, J. Burger, J. Aberdeen, D. Connolly, and L. Hirschman. 1995. A model-theoretic coreference scoring scheme. In MUC6, pages 45–52. find our card