Elamur Research⌕
☰
home / papers / 2021 / Multilingual Coreference Resolution with Harmonized Annotations

Multilingual Coreference Resolution with Harmonized Annotations

Summary

This pilot study asks whether harmonized coreference annotations let one model learn from several languages at once. The authors use CorefUD (Nedoluzhko et al., 2021), which converts existing coreference corpora into one cluster-based scheme built on Universal Dependencies. They select six of its largest datasets: Czech, Russian, Polish, German, Spanish and Catalan. Their system is the end-to-end span-ranking model of Lee et al. (2017), taken from the Xu and Choi (2020) implementation without higher-order inference. It runs on multilingual BERT or Slavic BERT. They train monolingual models, a joint model for the three Slavic languages and a joint model for all six, and score them with the CoNLL F1 average over five runs. Pooling training data improved every language except Czech, which already has the largest training set. The biggest gains came for the smallest corpora, Russian and German. This suggests harmonized resources such as CorefUD can help lower-resource languages through joint multilingual training.

Key points

  • Data: six CorefUD datasets (Catalan-AnCora, Czech-PDT, German-PotsdamCC, Polish-PCC, Russian-RuCor, Spanish-AnCora).
    • The official test sets are not public, so the original dev sets served as test data.
    • A new dev set was carved from 10% of each training portion.
  • Model limitations: the architecture scores only continuous spans and cannot predict singletons.
    • It therefore misses discontinuous mentions entirely, which make up 6.3% of mentions in German-PotsdamCC.
    • The authors also report results with singletons filtered out of the test data. Singletons are frequent in Polish-PCC (82.6%), German-PotsdamCC (76.5%) and Catalan-AnCora (74.6%).
  • Main gains: the all-language Joined-mBERT model raised German F1 by about 6.1 points over its monolingual counterpart. Slavic-mBERT raised Russian by about 3.4 points. Czech showed no real benefit.
  • Results without singletons: German improves from 52.5 (monolingual mBERT) to 61.4 (joint model). The best Czech score is 65.8, achieved with monolingual Slavic BERT.
  • Future work: training needed 32GB GPUs, which limited how many languages could be covered. The authors plan to add the remaining CorefUD languages and test zero-shot cross-lingual transfer.

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); 7 of them have a page here, 14 not yet.

Cite

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

@inproceedings{prazak2021multilingual,
  title = {{Multilingual Coreference Resolution with Harmonized Annotations}},
  author = {Pražák, Ondřej and Konopík, Miloslav and Sido, Jakub},
  booktitle = {Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021)},
  year = {2021},
  doi = {10.26615/978-954-452-072-4\_125},
  url = {https://doi.org/10.26615/978-954-452-072-4_125}
}

References

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

  1. Mikhail Arkhipov, Maria Trofimova, Yuri Kuratov, and Alexey Sorokin. 2019. Tuning multilingual transformers for language-specific named entity recognition. In Proceedings of the 7th Workshop on Balto-Slavic Natural Language Processing, pages 89–93, Florence, Italy. Association for Computational Ling… find our card
  2. Andre Cruz, Gil Rocha, and Henrique Lopes Cardoso. 2018. Exploring spanish corpora for portuguese coreference resolution. pages 290–295. find
  3. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. BERT: pre-training of deep bidirectional transformers for language understanding. CoRR, abs/1810.04805. find
  4. Mandar Joshi, Danqi Chen, Yinhan Liu, Daniel S. Weld, Luke Zettlemoyer, and Omer Levy. 2020. find
  5. SpanBERT: Improving pre-training by representing and predicting spans. Transactions of the Association for Computational Linguistics, 8:64–77. find our card
  6. Mandar Joshi, Omer Levy, Luke Zettlemoyer, and Daniel Weld. 2019. BERT for coreference resolution: Baselines and analysis. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP… find our card
  7. Ben Kantor and Amir Globerson. 2019. Coreference resolution with entity equalization. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 673–677, Florence, Italy. Association for Computational Linguistics. find our card
  8. Gourab Kundu, Avi Sil, Radu Florian, and Wael Hamza. 2018. Neural cross-lingual coreference resolution and its application to entity linking. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 395–400, Melbourne, Australia. Asso… find our card
  9. Kenton Lee, Luheng He, Mike Lewis, and Luke Zettlemoyer. 2017. End-to-end neural coreference resolution. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 188–197, Copenhagen, Denmark. Association for Computational Linguistics. find our card
  10. Anna Nedoluzhko, Michal Novák, Martin Popel, Zdeněk Žabokrtský, and Daniel Zeman. 2021. Coreference meets Universal Dependencies – a pilot experiment on harmonizing coreference datasets for 11 languages. ÚFAL MFF UK, Praha, Czechia. find our card
  11. Bartłomiej Nitoń, Paweł Morawiecki, and Maciej Ogrodniczuk. 2018. Deep neural networks for coreference resolution for Polish. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA). find our card
  12. Marta Recasens, Lluı́s Màrquez, Emili Sapena, M. Antònia Martı́, 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, Uppsa… find our card
  13. A. Sboev, R. Rybka, and A. Gryaznov. 2020. Deep neural networks ensemble with word vector representation models to resolve coreference resolution in russian. In Advanced Technologies in Robotics and Intelligent Systems, pages 35–44, Cham. Springer International Publishing. find
  14. Ankit Srivastava, Sabine Weber, Peter Bourgonje, and Georg Rehm. 2018. Different german and english coreference resolution models for multi-domain content curation scenarios. In Language Technologies for the Challenges of the Digital Age, pages 48–61, Cham. Springer International Publishing. find
  15. Rhea Sukthanker, Soujanya Poria, Erik Cambria, and Ramkumar Thirunavukarasu. 2020. Anaphora and coreference resolution: A review. Information Fusion, 59:139–162. find
  16. Gorka Urbizu, Ander Soraluze, and Olatz Arregi. 2019. Deep cross-lingual coreference resolution for less-resourced languages: The case of Basque. In Proceedings of the Second Workshop on Computational Models of Reference, Anaphora and Coreference, pages 35–41, Minneapolis, USA. Association for Compu… find our card
  17. Liyan Xu and Jinho D. Choi. 2020. Revealing the myth of higher-order inference in coreference resolution. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 8527–8533, Online. Association for Computational Linguistics. find our card
  18. Michal Novák. 2017. Coreference resolution system not only for czech. In Proceedings of the 17th Conference on Information Technologies - Applications and Theory (ITAT 2017), Martinské hole, Slovakia, September 22-26, 2017, volume 1885 of CEUR Workshop Proceedings, pages 193–200. CEUR-WS.org. find
  19. Sameer Pradhan, Xiaoqiang Luo, Marta Recasens, Eduard Hovy, Vincent Ng, and Michael Strube. 2014. Scoring coreference partitions of predicted mentions: A reference implementation. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pag… find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/prazak-2021-multilingual-coreference-resolution-harmonized-annotations/