Elamur Research⌕
☰
home / papers / 2012 / Data-driven Multilingual Coreference Resolution using Resolver Stacking

Data-driven Multilingual Coreference Resolution using Resolver Stacking

Summary

This CoNLL 2012 Shared Task system performs coreference resolution for Arabic, Chinese and English. Its goal is to reduce a known weakness of pair-wise resolvers, which judge each mention pair in isolation. The pipeline begins with a deliberately over-generating mention extractor built on parse trees and, for English, named entities. In English, a MaxEnt filter then removes clearly non-referential uses of it, we and you. A LIBLINEAR logistic regression model scores mention pairs, using Soon et al.-style training instances.

The authors add a new cluster-aware decoder, AverageMaxProb (AMP). It links a mention to the existing cluster that has the highest geometric-mean link probability across all of its members. AMP yields smaller but more consistent clusters. Its output then becomes an extra feature for a second, conventional pair-wise decoder (BF or PCF), in a stacking setup. Decoders also block clusters formed by transitively merging nested mentions.

The stacked system scored 58.25 officially, second among sixteen teams. This shows that cluster-level and pair-wise decoding complement each other across typologically different languages.

Key points

  • Official result: 58.25 overall (mean of MUC, B3 and CEAFE over three languages), ranked 2nd of 16. Test CoNLL scores with predicted mentions were Arabic 53.55, Chinese 59.97 and English 61.24.
  • English pronoun filter: separate Mallet MaxEnt classifiers handle it, we and you. A strict 0.95 posterior threshold keeps false removals rare and added about 0.5 CoNLL points. Similar filters for Arabic and Chinese brought no benefit.
  • Nesting constraint: forbidding transitive clustering of nested mentions improved English by roughly 0.6 points, while Arabic and Chinese each rose by around 0.4.
  • Stacking design:
    • AMP runs first with a fixed 0.5 threshold.
    • The second resolver receives a five-valued feature describing the first resolver's decision (NEITHER, IONLY, JONLY, COREF, DISREF).
    • Arabic uses PCF at 0.60, Chinese uses BF at 0.65, and English uses PCF at 0.65.
    • Stacking two pair-wise decoders gave no improvement.
  • Development-set CoNLL scores (Stacked): Arabic 51.84, Chinese 60.30 and English 61.71, each beating the BF, PCF and AMP decoders used alone. The final models were trained on train+dev, which notably boosted Arabic on the test set.

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

21 papers of our library cite it (of the 140673 whose bibliographies we read); 13 of them have a page here, 8 not yet.

Cite

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

@inproceedings{bjorkelund2012data,
  title = {{Data-driven Multilingual Coreference Resolution using Resolver Stacking}},
  author = {Björkelund, Anders and Farkas, Richárd},
  booktitle = {Joint Conference on EMNLP and CoNLL - Shared Task},
  year = {2012},
  url = {https://aclanthology.org/W12-4503/}
}

References

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

  1. Anders Björkelund and Pierre Nugues. 2011. Exploring lexicalized features for coreference resolution. In Proceedings of the Fifteenth Conference on Computational Natural Language Learning: Shared Task, pages 45–50, June. find
  2. Adriane Boyd, Whitney Gegg-Harrison, and Donna Byron. 2005. Identifying non-referential it: A machine learning approach incorporating linguistically motivated patterns. In Proceedings of the ACL Workshop on Feature Engineering for Machine Learning in Natural Language Processing, pages 40–47, June. find
  3. Jinho D. Choi and Martha Palmer. 2010. Robust Constituent-to-Dependency Conversion for English. In Proceedings of 9th Treebanks and Linguistic Theories Workshop (TLT), pages 55–66. find
  4. Rong-En Fan, Kai-Wei Chang, Cho-Jui Hsieh, Xiang-Rui Wang, and Chih-Jen Lin. 2008. LIBLINEAR: A library for large linear classification. Journal of Machine Learning Research, 9:1871–1874. find
  5. Xiaoqiang Luo and Imed Zitouni. 2005. Multi-lingual coreference resolution with syntactic features. In Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing, pages 660–667, October. find our card
  6. Andrew Kachites McCallum. 2002. Mallet: A machine learning for language toolkit. http://mallet.cs.umass.edu. ↗
  7. Eugene W. Myers. 1986. An O(ND) difference algorithm and its variations. Algorithmica, 1:251–266. find
  8. Vincent Ng. 2010. Supervised noun phrase coreference research: The first fifteen years. In Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 1396–1411, July. find our card
  9. 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). find our card
  10. Emili Sapena, Lluı́s Padró, and Jordi Turmo. 2010. A global relaxation labeling approach to coreference resolution. In Coling 2010: Posters, pages 1086–1094, August. find our card
  11. 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. find our card
  12. Yue Zhang and Stephen Clark. 2011. Syntactic processing using the generalized perceptron and beam search. Computational Linguistics, 37(1):105–151. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bjorkelund-2012-data-driven-multilingual-coreference-resolution/