Elamur Research⌕
☰
home / papers / 2021 / The Pipeline Model for Resolution of Anaphoric Reference and Resolution of Entity Reference

The Pipeline Model for Resolution of Anaphoric Reference and Resolution of Entity Reference

Summary

This paper presents a Konkuk University system for the CODI-CRAC 2021 shared task, which targets anaphora phenomena in dialogue that go beyond identity coreference in written text. The authors address two of the three subtasks: anaphoric identity resolution (including singletons and split antecedents) and bridging reference resolution.

For identity anaphora, they use a pipeline. Mentions, which can be nested, are found by predicting opening and closing brackets independently for each token. This step uses an ELECTRA-large encoder, BiLSTMs and a label attention network. Mention vectors built from averaged token representations then feed a pointer network, which links each mention to its nearest antecedent. Chains are then merged sequentially.

Bridging is recast as machine reading comprehension. A RoBERTa-large model pre-trained on QUOREF reads a query built from each cluster's first mention, together with the full dialogue, and extracts the related entity span.

The system ranked third on identity anaphora and second in both bridging tracks. This suggests that MRC reformulation is a workable approach to bridging in dialogue, though its precision is limited.

Key points

  • Subtask 1 averaged 65.9% CoNLL F1 on the test sets (3rd place). Per-corpus scores were roughly 69.1 (light), 57.5 (AMI), 71.0 (Persuasion) and 65.6 (Switchboard).
  • Test-set mention detection F1 was 94.18 (light), 90.88 (AMI), 92.93 (Persuasion) and 91.07 (Switchboard).
  • Bridging overall scores were 11.76% in Eval-Br(Pred) and about 17.3% in Eval-Br(Gold), 2nd place in both. The paper reports both 17.27 and 17.26 for the Gold track.
  • Bridging queries follow the template "What is related of x?". Here x is the earliest mention of a subtask-1 cluster, chosen so that the query does not contain an ambiguous pronoun.
  • Recall exceeded precision in the bridging tracks because the MRC model cannot reject unanswerable queries. Precision improved when gold mentions were provided.

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

4 papers of our library cite it (of the 140673 whose bibliographies we read); 1 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. DOI 10.18653/v1/2021.codi-sharedtask.4

@inproceedings{kim2021pipeline,
  title = {{The Pipeline Model for Resolution of Anaphoric Reference and Resolution of Entity Reference}},
  author = {Kim, Hongjin and Kim, Damrin and Kim, Harksoo},
  booktitle = {Proceedings of the CODI-CRAC 2021 Shared Task on Anaphora, Bridging, and Discourse Deixis in Dialogue},
  year = {2021},
  doi = {10.18653/v1/2021.codi-sharedtask.4},
  url = {https://doi.org/10.18653/v1/2021.codi-sharedtask.4}
}

References

2 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Oriol Vinyals, Meire Fortunato, and Navdeep Jaitly. 2015. Pointer networks. arXiv preprint arXiv:1506.03134. find
  2. Amit Bagga and Breck Baldwin. 1998. Algorithms for scoring coreference chains. In The first international conference on language resources and evalua- find our card

This page in Markdown → https://md.research.elamur.ai/free/en/papers/kim-2021-pipeline-model-resolution-anaphoric-reference/