Elamur Research
☰
home / papers / 2012 / Latent Structure Perceptron with Feature Induction for Unrestricted Coreference Resolution

Latent Structure Perceptron with Feature Induction for Unrestricted 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/W12-4502.pdf

Summary

The paper addresses unrestricted coreference resolution, which means grouping all mentions in a document that refer to the same entity or event. It does so for three languages in the CoNLL-2012 Shared Task closed track, using OntoNotes data with predicted annotation layers. Candidate mentions are noun phrases, pronouns and named entities, extracted with an existing heuristic. Clustering them directly under standard cluster metrics would lead to NP-hard prediction. The authors instead represent each cluster as a directed coreference tree hanging from an artificial document root. Because gold trees are never annotated, the trees are treated as hidden variables and learned with a large margin latent structure perceptron. Decoding becomes a maximum branching problem that Chu-Liu-Edmonds solves efficiently. Entropy guided feature induction automatically builds conjunctive feature templates, which lets a linear learner behave like a nonlinear model. Apart from small language-specific resources such as pronoun lists, the same system runs on Arabic, Chinese and English. It obtained the top official score in the shared task, which suggests that latent tree modeling is a practical and strong approach to multilingual coreference.

Key points

  • Results on the test sets, by language: Arabic 54.22, Chinese 58.49, English 63.37. The official mean of 58.69 ranked first in the CoNLL-2012 closed track.
  • Candidate arcs between mentions are pruned with recall-oriented filters adapted from the sieves of Lee et al. (2011). Decoding uses the Chu-Liu-Edmonds algorithm with an averaged perceptron.
  • A tunable "root loss value" penalizes arcs from the artificial root, which pushes the model toward larger clusters and a better precision/recall balance. Tuned values were 6 for Arabic, 2 for Chinese and 1.5 for English, with the largest gains for Arabic and Chinese.
  • Arabic test performance clearly exceeded its development score, likely because the small Arabic training set was enlarged with the development data for the final model.
  • In supplementary experiments, golden mentions lifted the mean F-score by nearly 8 points, and golden parses also helped every language, especially Chinese. Golden mention boundaries helped only Chinese, by about 3 points.

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

45 papers of our library cite it (of the 140673 whose bibliographies we read); 10 of them have a page here, 35 not yet.

Cite

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

@inproceedings{fernandes2012latent,
  title = {{Latent Structure Perceptron with Feature Induction for Unrestricted Coreference Resolution}},
  author = {Fernandes, Eraldo and Santos, Cícero dos and Milidiú, Ruy},
  booktitle = {Joint Conference on EMNLP and CoNLL - Shared Task},
  year = {2012},
  url = {https://aclanthology.org/W12-4502/}
}

References

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

  1. Y. J. Chu and T. H. Liu. 1965. On the shortest arborescence of a directed graph. Science Sinica, 14:1396– 1400. find
  2. Michael Collins. 2002. Discriminative training methods for hidden Markov models: theory and experiments with perceptron algorithms. In Proceedings of the ACL-02 Conference on Empirical Methods in Natural Language Processing, pages 1–8. find our card
  3. Cicero Nogueira dos Santos and Davi Lopes Carvalho. 2011. Rule and tree ensembles for unrestricted coreference resolution. In Proceedings of the Fifteenth Conference on Computational Natural Language Learning: Shared Task, pages 51–55, Portland, Oregon, USA, June. Association for Computational Linguistics. find
  4. J. Edmonds. 1967. Optimum branchings. Journal of Research of the National Bureau of Standards, 71B:233– 240. find
  5. Eraldo R. Fernandes and Ulf Brefeld. 2011. Learning from partially annotated sequences. In Proceedings of the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML-PKDD), Athens, Greece. find
  6. Eraldo R. Fernandes and Ruy L. Milidiú. 2012. Entropy-guided feature generation for structured learning of Portuguese dependency parsing. In Proceedings of the Conference on Computational Processing of the Portuguese Language (PROPOR), volume 7243 of Lecture Notes in Computer Science, pages 146–156. Springer Berlin / Heidelberg. find
  7. Heeyoung Lee, Yves Peirsman, Angel Chang, Nathanael Chambers, Mihai Surdeanu, and Dan Jurafsky. 2011. Stanford’s multi-pass sieve coreference resolution system at the CoNLL-2011 shared task. In Proceedings of the Fifteenth Conference on Computational Natural Language Learning: Shared Task, CoNLL Shared Task 2011, pages 28–34, Stroudsburg, PA, USA. Association for Computational Linguistics. find our card
  8. Ryan McDonald, Koby Crammer, and Fernando Pereira. 2005. Online large-margin training of dependency parsers. In Proceedings of the 43rd Annual Meeting on Association for Computational Linguistics, ACL’05, pages 91–98. find our card
  9. Ryan McDonald, Kevin Lerman, and Fernando Pereira. 2006. Multilingual dependency analysis with a two-stage discriminative parser. In Proceedings of the Conference on Computational Natural Language Learning (CoNLL), pages 216–220. find
  10. Ruy L. Milidiú, Cı́cero N. dos Santos, and Julio C. Duarte. 2008. Phrase chunking using entropy guided transformation learning. In Proceedings of ACL2008, Columbus, Ohio. find
  11. Sameer Pradhan, Alessandro Moschitti, Nianwen Xue, Olga Uryupina, and Yuchen Zhang. 2012. CoNLL-2012 shared task: Modeling multilingual unrestricted coreference in OntoNotes. In Proceedings of the Sixteenth Conference on Computational Natural Language Learning (CoNLL 2012), Jeju, Korea. find our card
  12. Frank Rosenblatt. 1957. The Perceptron – a perceiving and recognizing automaton. Technical report, Cornell Aeronautical Laboratory. Report 85-460-1. find
  13. I. Tsochantaridis, T. Joachims, T. Hofmann, and Y. Altun. 2005. Large margin methods for structured and interdependent output variables. Journal of Machine Learning Research, 6:1453–1484. find
  14. Chun-Nam Yu and Thorsten Joachims. 2009. Learning structural SVMs with latent variables. In Proceedings of the International Conference on Machine Learning (ICML). find