Elamur Research⌕
☰
home / papers / 2011 / Inference Protocols for Coreference Resolution

Inference Protocols for Coreference Resolution

Summary

This paper presents Illinois-Coref, the system the University of Illinois submitted to the closed-track competition of CoNLL-2011, a shared task on resolving coreference in OntoNotes-4.0. The system is built with Learning Based Java and extends the Bengtson and Roth (2008) pipeline. A rule-based mention detector feeds a pairwise compatibility scorer, and a decoder then groups the mentions into entities. The authors compare two decoders, both cast as integer linear programs. Best-Link attaches each mention to its highest-scoring preceding candidate and is trained on pairwise examples. All-Link (correlational clustering) enforces transitivity across whole clusters and is trained with a structured perceptron. The scorer can also take hand-written, high-precision constraints aimed at recall errors involving definite NPs and named-entity heads. Despite its simpler design, Best-Link beat All-Link on both predicted and gold mentions. The constraints helped only marginally. The error analysis points to unlinked coreferent mentions and pronoun ambiguity as the main remaining problems. The work matters because it tests whether global clustering actually helps coreference and offers a modular way to add linguistic knowledge.

Key points

  • Because OntoNotes-4.0 does not annotate singletons, mention detection keeps every NE and parser NP, giving about 90% recall and about 35% precision. Mentions left as singletons are removed after inference.
  • On DEV with predicted mentions, the averaged MUC/BCUB/CEAF score was 56.35 for Best-Link with constraints, 56.22 for Best-Link alone and 55.07 for All-Link. With gold mentions, Best-Link reached 73.65 and All-Link 70.18.
  • On TEST, the submitted Best-Link system averaged 55.96 with predicted mentions and boundaries, 56.62 with gold boundaries and 73.83 with gold mentions.
  • Using gold rather than predicted mentions improves results by roughly 15–18 absolute points, so mention detection is a major bottleneck.
  • Pronouns account for about 35% of mentions, and resolving them when several antecedents are plausible remains a key weakness.

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

18 papers of our library cite it (of the 140673 whose bibliographies we read); 7 of them have a page here, 11 not yet.

Cite

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

@inproceedings{chang2011inference,
  title = {{Inference Protocols for Coreference Resolution}},
  author = {Chang, Kai-Wei and Samdani, Rajhans and Rozovskaya, Alla and Rizzolo, Nick and Sammons, Mark and Roth, Dan},
  booktitle = {Proceedings of the Fifteenth Conference on Computational Natural Language Learning: Shared Task},
  year = {2011},
  url = {https://aclanthology.org/W11-1904/}
}

References

8 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. N. Bansal, A. Blum, and S. Chawla. 2002. Correlation clustering. In Proceedings of the 43rd Symposium on Foundations of Computer Science. find
  2. E. Bengtson and D. Roth. 2008. Understanding the value of features for coreference resolution. In EMNLP, 10. find our card
  3. T. Finley and T. Joachims. 2005. Supervised clustering with support vector machines. In Proceedings of the International Conference on Machine Learning (ICML). find
  4. A. Mccallum and B. Wellner. 2003. Toward conditional models of identity uncertainty with application to proper noun coreference. In The Conference on Advances in Neural Information Processing Systems (NIPS). find
  5. D. Pascal and J. Baldridge. 2009. Global joint models for coreference resolution and named entity classification. In Procesamiento del Lenguaje Natural. find
  6. S. Pradhan, L. Ramshaw, R. Weischedel, J. MacBride, and L. Micciulla. 2007. Unrestricted Coreference: Identifying Entities and Events in OntoNotes. In in Proceedings of the IEEE International Conference on Semantic Computing (ICSC), September 17-19. find
  7. S. Pradhan, L. Ramshaw, M. Marcus, M. Palmer, R. Weischedel, and N. Xue. 2011. Conll-2011 shared task: Modeling unrestricted coreference in ontonotes. In Proceedings of the Annual Conference on Computational Natural Language Learning (CoNLL). find our card
  8. N. Rizzolo and D. Roth. 2010. Learning Based Java for Rapid Development of NLP Systems. In Proceedings of the International Conference on Language Resources and Evaluation (LREC), Valletta, Malta, 5. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/chang-2011-inference-protocols-coreference-resolution/