Elamur Research⌕
☰
home / papers / 2011 / Poly-co: a multilayer perceptron approach for coreference detection

Poly-co: a multilayer perceptron approach for coreference detection

Summary

Poly-co is a coreference resolution system developed for the CoNLL-2011 Shared Task, competing in its closed-track setting. The task is to find which named entities, pronouns and noun phrases in English text point to the same real-world entity.

The authors keep the classic pairwise design of Soon et al. (2001). A rule-based pipeline over the Ontonotes annotation layers first extracts candidate mentions. Additional modules then cluster name aliases, measure word overlap between mentions and assign gender and number. Each mention pair becomes a 22-feature vector, and a classifier decides whether the two mentions corefer.

Training pairs come from a backward search: each mention is compared with up to ten earlier candidates, which yields one positive example and several negative ones. A multilayer perceptron was tested against SVM and a J48 decision tree on identical features. The MLP was retained, but its advantage over J48 is negligible.

The main takeaway is that, once the feature model is fixed, the choice of classifier changes results only slightly. Improving features and mention detection therefore seems more worthwhile.

Key points

  • Data: the Ontonotes corpus as provided for CoNLL-2011 (closed track). Gender lexicon from Bergsma's list, which the shared task permits.
  • Features: 22 in total.
    • Four describe the pair: IsAlias, IsSimilar, Distance and Sent.
    • Nine describe each mention: NE/PRP/NP flags, one of 18 NE types, 30 pronoun values, determiner, NP type, gender and number.
  • Pairing:
    • At most n = 10 preceding mentions are examined.
    • At inference, the first antecedent whose MLP probability exceeds 0.5 is linked.
    • If no antecedent qualifies, the mention is treated as non-coreferring.
  • Gold dev-set, MLP results (F-scores): mentions 65.37, B3 64.27, CEAF 50.18, MUC 52.60.
    • J48 reached B3 64.33 and MUC 52.36.
    • SVM was weaker, with B3 61.24.
  • Predicted dev-set, MLP results (F-scores): mentions 63.97, B3 63.79, CEAF 49.12, MUC about 50.90.

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

5 papers of our library cite it (of the 140673 whose bibliographies we read); 3 of them have a page here, 2 not yet.

Cite

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

@inproceedings{charton2011poly,
  title = {{Poly-co: a multilayer perceptron approach for coreference detection}},
  author = {Charton, Eric and Gagnon, Michel},
  booktitle = {Proceedings of the Fifteenth Conference on Computational Natural Language Learning: Shared Task},
  year = {2011},
  url = {https://aclanthology.org/W11-1915/}
}

References

7 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Eric Charton, Michel Gagnon, and Benoit Ozell. 2010. Poly-co : an unsupervised co-reference detection system. In INLG 2010-GREC, Dublin. ACL SIGGEN. find
  2. Hoifung Poon and Pedro Domingos. 2008. Joint unsupervised coreference resolution with Markov logic. In Proceedings of the Conference on Empirical Methods in Natural Language Processing - EMNLP ’08, page 650, Morristown, NJ, USA. Association for Computational Linguistics. find our card
  3. Sameer Pradhan, Lance Ramshaw, Ralph Weischedel, Jessica MacBride, and Linnea Micciulla. 2007. Unrestricted coreference: Identifying entities and events in OntoNotes. In International Conference on Semantic Computing, 2007. find
  4. Sameer Pradhan, Lance Ramshaw., Mitchell Marcus, Martha Palmer, Ralph Weischedel, and Xue Nianwen. 2011. CoNLL-2011 Shared Task: Modeling Unrestricted Coreference in OntoNotes. In Proceedings of the Fifteenth Conference on Computational Natural Language Learning (CoNLL 2011), Portland, Oregon. find our card
  5. Emili Sapena, L. Padró, and Jordi Turmo. 2010. Relax-Cor: A global relaxation labeling approach to coreference resolution. In Proceedings of the 5th International Workshop on Semantic Evaluation, number July, pages 88–91. Association for Computational Linguistics. find our card
  6. Wee Meng Soon, Hwee Tou Ng, and Daniel Chung Yong Lim. 2001. A Machine Learning Approach to Coreference Resolution of Noun Phrases. Computational Linguistics, 27(4):521–544, December. find our card
  7. Yannick Versley, S.P. Ponzetto, Massimo Poesio, Vladimir Eidelman, Alan Jern, Jason Smith, Xiaofeng Yang, and Alessandro Moschitti. 2008. BART: A modular toolkit for coreference resolution. In Proceedings of the Sixth International Language Resources and Evaluation (LREC’08), number 2006, pages 9–12… find our card

This page in Markdown → https://md.research.elamur.ai/free/papers/charton-2011-poly-co-multilayer-perceptron-approach/