Elamur Research⌕
☰
home / papers / 2019 / Incorporating Context and External Knowledge for Pronoun Coreference Resolution

Incorporating Context and External Knowledge for Pronoun Coreference Resolution

Summary

Pronouns have little semantic content, so finding their antecedents often requires both an understanding of the surrounding text and world knowledge that is hard to learn from annotated corpora. Examples include number agreement, gender and animacy, and which arguments a verb typically takes. The authors propose a two-layer neural architecture that combines these signals. The first layer builds span representations for the pronoun and candidate noun phrases using a biLSTM with inner-span attention, following the End2end model of Lee et al. (2017), and produces a context-based score. A softmax pruning step keeps only high-confidence candidates. The second layer compares the remaining candidates pair by pair using features from several external knowledge sources. A knowledge attention module, conditioned on context, decides how much each source should count. This makes the model robust to noisy, automatically built resources and lets it adapt knowledge use to each case. On a widely used benchmark, the model beats state-of-the-art baselines by a large margin. Its attention weights also show which knowledge type drove each decision, which makes the model interpretable.

Key points

  • The final score is the sum of a context score (F_c) and a knowledge score (F_k). Candidates are all non-pronominal noun phrases from the three most recent sentences before the pronoun.
  • There are three knowledge types: Plurality, animacy/gender (AG), and selectional preference (SP) drawn from a parsed SP knowledge base.
  • Pairwise comparison costs O(|N|²), so softmax pruning between the layers keeps computation manageable.
  • In case studies against End2end, SP resolves cases where Plurality and AG cannot separate the candidates. For example, "son" appears 100 times as an argument of "child" in the SP base, and "accident" appears 26 times with "fault." End2end either overpredicts in these cases or returns no antecedent.
  • Code and data are released at github.com/HKUST-KnowComp/Pronoun-Coref.

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

20 papers of our library cite it (of the 140673 whose bibliographies we read); 3 of them have a page here, 17 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/n19-1093 · arXiv 1905.10238

@inproceedings{zhang2019incorporating,
  title = {{Incorporating Context and External Knowledge for Pronoun Coreference Resolution}},
  author = {Zhang, Hongming and Song, Yan and Song, Yangqiu},
  booktitle = {Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)},
  year = {2019},
  doi = {10.18653/v1/n19-1093},
  eprint = {1905.10238},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.18653/v1/n19-1093}
}

References

5 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2015. Neural machine translation by jointly learning to align and translate. In Proceedings of ICLR 2015. find our card
  2. Eugene Charniak and Micha Elsner. 2009. Em works for pronoun anaphora resolution. In Proceedings of EACL 2009, pages 148–156. find our card
  3. Kevin Clark and Christopher D Manning. 2015. Entity-centric coreference resolution with model stacking. In Proceedings of ACL-IJCNLP 2015, volume 1, pages 1405–1415. find our card
  4. Kevin Clark and Christopher D. Manning. 2016. Deep reinforcement learning for mention-ranking coreference models. In Proceedings of EMNLP 2016, pages 2256–2262. find our card
  5. Richard J Edens, Helen L Gaylard, Gareth JF Jones, and Adenike M Lam-Adesina. 2003. An investigation of broad coverage automatic pronoun resolution for information retrieval. In Proceedings of SIGIR 2003, pages 381–382. ACM. \#\# Conclusion find

This page in Markdown → https://md.research.elamur.ai/free/papers/zhang-2019-incorporating-context-external-knowledge-pronoun/