Elamur Research⌕
☰
home / papers / 2021 / Bridging Resolution: Making Sense of the State of the Art

Bridging Resolution: Making Sense of the State of the Art

Summary

Bridging resolution means linking an anaphor to a related but non-identical antecedent, as in "the games" linked to "baseball." Yu and Poesio's (2020) neural multi-task learning (MTL) model, trained jointly on coreference and bridging, is the current state of the art. It was unclear why it beats rule-based systems and where it still fails. Kobayashi and Ng build a sieve-style hybrid. First, the hand-crafted rules (Hou's rules for ISNotes, Rösiger's rulesets for BASHI and ARRAU) propose links. The MTL model then handles the remaining mentions. They evaluate full bridging resolution with gold mentions on ISNotes, BASHI and ARRAU RST. The hybrid has much higher recall than either component and achieves the best F-scores on all three corpora. This shows that the rules and MTL make complementary errors, so they should be combined rather than treated as rivals. A manual error analysis identifies the main weaknesses of MTL. The authors also suggest that the annotation quality of BASHI may need to be re-examined.

Key points

  • Data:
    • ISNotes: 50 WSJ docs, 663 anaphors.
    • BASHI: 50 WSJ docs, 459 anaphors.
    • ARRAU RST: 413 docs, 3777 anaphors, standard train-test split.
    • ISNotes and BASHI use 10-fold cross-validation.
  • F-scores (recognition / resolution), Hybrid vs MTL:
    • ISNotes: 49.5 / 30.0 vs 43.8 / 25.2.
    • BASHI: 40.5 / 20.1 vs 35.1 / 18.1.
    • ARRAU RST: 37.4 / 25.4 vs 36.7 / 24.0. The gain is small here because the rules' precision is no better than MTL's.
  • Rules vs MTL on ISNotes: the rules win on most rule categories, but MTL wins on the large leftover "Other" category.
  • Main MTL error types:
    • Recognition precision: coreferent mentions mislabelled as bridging, 14–30% of these errors.
    • Recognition recall: indefinite anaphors confused with NEW mentions, 48–71% of these errors.
    • Resolution precision: short mentions without modifiers, 23–63% of these errors.
  • BASHI anomalies: documents are longer than in ISNotes (1154 vs 805 tokens/doc), and the authors show examples of bridging links that were not annotated.

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

10 papers of our library cite it (of the 140673 whose bibliographies we read); 3 of them have a page here, 7 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2021.naacl-main.131

@inproceedings{kobayashi2021bridging,
  title = {{Bridging Resolution: Making Sense of the State of the Art}},
  author = {Kobayashi, Hideo and Ng, Vincent},
  booktitle = {Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies},
  year = {2021},
  doi = {10.18653/v1/2021.naacl-main.131},
  url = {https://doi.org/10.18653/v1/2021.naacl-main.131}
}

References

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

  1. Pascal Denis and Jason Baldridge. 2008. Specialized models and ranking for coreference resolution. In Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 660–669. find our card
  2. Yufang Hou. 2020. Bridging anaphora resolution as question answering. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 1428–1438. find our card
  3. Yufang Hou, Katja Markert, and Michael Strube. 2014. A rule-based system for unrestricted bridging resolution: Recognizing bridging anaphora and finding links to antecedents. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, pages 2082–2093. find our card
  4. Yufang Hou, Katja Markert, and Michael Strube. 2018. Unrestricted bridging resolution. Computational Linguistics, 44(2):237–284. ↗ our card
  5. Eduard Hovy, Mitchell Marcus, Martha Palmer, Lance Ramshaw, and Ralph Weischedel. 2006. OntoNotes: The 90% solution. In Proceedings of the Human Language Technology Conference of the NAACL, Companion Volume: Short Papers, pages 57–60. find our card
  6. 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. find our card
  7. Heeyoung Lee, Angel Chang, Yves Peirsman, Nathanael Chambers, Mihai Surdeanu, and Dan Jurafsky. 2013. Deterministic coreference resolution based on entity-centric, precision-ranked rules. Computational Linguistics, 39(4):885–916. In Proceedings of the First Workshop on Computational Models of Reference, Anaphora and Coreference, pages 23–33. ↗ our card
  8. Ina Rösiger, Arndt Riester, and Jonas Kuhn. 2018. Bridging resolution: Task definition, corpus resources and rule-based experiments. In Proceedings of the 27th International Conference on Computational Linguistics, pages 3516–3528. find our card
  9. Olga Uryupina, Ron Artstein, Antonella Bristot, Federica Cavicchio, Francesca Delogu, Kepa Joseba Rodriguez, and Massimo Poesio. 2020. Annotating a broad range of anaphoric phenomena, in a variety of genres: the ARRAU corpus. Natural Language Engineering, 26(1):95–128. ↗
  10. Juntao Yu and Massimo Poesio. 2020. Multi-task learning based neural bridging reference resolution. In Proceedings of the 28th International Conference on Computational Linguistics, pages 3534–3546. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/kobayashi-2021-bridging-resolution-making-sense-state/