Combining the Best of Two Worlds: A Hybrid Approach to Multilingual Coreference Resolution

Summary

This paper describes the University of Texas at Dallas entry to the CoNLL-2012 shared task, which covers coreference resolution in OntoNotes for three typologically distant languages: English, Chinese, and Arabic. The authors argue that hand-written rules and statistical learning each have distinct advantages, so they build a system that uses both.

Mention detection works in two stages. Language-specific heuristics first extract candidates from parse trees and named entities. A learned filter then removes candidates whose head nouns are rarely annotated as coreferent in the training data.

Resolution builds on the Stanford multi-pass sieve architecture. The authors adapt the sieves for Chinese with a new head-match sieve, extra abbreviation rules, and bootstrapped gender, number, and animacy lists. They also add lexical knowledge learned from annotated data. All thresholds and rule settings are tuned jointly for each genre against the evaluation metric.

The system achieved an official combined score of 56.35 and the best Chinese result among participants. This shows that rule-based and data-driven components can complement each other across languages.

Key points

The team entered four tracks: the closed track for English, Chinese, and Arabic, plus the open track for Chinese, which used automatically predicted NEs.
Ten of Stanford's sieves were reimplemented for English, omitting the WordNet, Wikipedia, and Freebase-based ones. The Chinese resolver uses nine sieves.
Before learned pruning, mention extraction on the development set reached recall of 88.59 (English), 85.74 (Chinese), and 81.49 (Arabic), but precision was low (about 21–43).
On English development data, the full system averaged 62.3 F, compared with 60.6 for the re-implemented Lee et al. (2011) baseline. Removing the lexical and threshold extensions lowered scores, whether each was removed alone or both together.
With gold mentions on the test set, average F rose to 70.5 for English, 77.8/79.0 for Chinese (closed/open), and 52.3 for Arabic. Arabic remained the weakest language.

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

15 papers of our library cite it (of the 140673 whose bibliographies we read); 12 of them have a page here, 3 not yet.

Cite

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

@inproceedings{chen2012combining,
  title = {{Combining the Best of Two Worlds: A Hybrid Approach to Multilingual Coreference Resolution}},
  author = {Chen, Chen and Ng, Vincent},
  booktitle = {Joint Conference on EMNLP and CoNLL - Shared Task},
  year = {2012},
  url = {https://aclanthology.org/W12-4504/}
}

References

10 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

Michael John Collins. 1999. Head-Driven Statistical Models for Natural Language Parsing. Ph.D. thesis, University of Pennsylvania, Philadelphia, PA. find
Aria Haghighi and Dan Klein. 2009. Simple coreference resolution with rich syntactic and semantic features. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 1152--1161. find our card
Heeyoung Lee, Yves Peirsman, Angel Chang, Nathanael Chambers, Mihai Surdeanu, and Dan Jurafsky. 2011. find
Incorrect that; that idea (ours,they); (both of you,us) (China,its); (Taiwan,its) (himself,he); (My,I) (she,the woman); (Taiwan,the island) Incorrect --- (我们这儿人, 他们); (爸爸, 我) (台湾, 你们); (我国, 我) (咱们妈, 她妈); (咱们, 咱) (中国, 中) ; (亚洲地区, 亚洲) Stanford's multi-pass sieve coreference resolution system at the Co… find
Vincent Ng. 2004. Learning noun phrase anaphoricity to improve conference resolution: Issues in representation and optimization. In Proceedings of the 42nd Meeting of the Association for Computational Linguistics, pages 151--158. find our card
Vincent Ng. 2009. Graph-cut-based anaphoricity determination for coreference resolution. In Proceedings of the 2009 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 575--583. find our card
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. find our card
Altaf Rahman and Vincent Ng. 2011a. Coreference resolution with world knowledge. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, pages 814--824. find our card
Altaf Rahman and Vincent Ng. 2011b. Narrowing the modeling gap: A cluster-ranking approach to coreference resolution. Journal of Artificial Intelligence Research, 40:469--521. find
Marta Recasens, Lluís Màrquez, Emili Sapena, M. Antònia Martí, Mariona Taulé, Véronique Hoste, Massimo Poesio, and Yannick Versley. 2010. Semeval-2010 task 1: Coreference resolution in multiple languages. In Proceedings of the 5th International Workshop on Semantic Evaluation, pages 1--8. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/chen-2012-combining-best-two-worlds-hybrid/

📄 Combining the Best of Two Worlds: A Hybrid Approach to Multilingual Coreference Resolution