Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Updated Coreference Resolution SoTA #400

Merged
merged 4 commits into from Dec 30, 2019

Conversation

@AndreFCruz
Copy link
Contributor

AndreFCruz commented Dec 21, 2019

Updated coreference resolution SoTA with 3 recently published works:

  • Fei et al. (2019): End-to-end Deep Reinforcement Learning Based Coreference Resolution
  • Kantor and Globerson (2019): Coreference Resolution with Entity Equalization
  • Joshi et al. (2019): BERT for Coreference Resolution: Baselines and Analysis

Set a footnote for the most recent work to follow previous repo. organization, as it was getting difficult to read the train of improvements (Lee et al (2017) + BERT + ...)

@sebastianruder

This comment has been minimized.

Copy link
Owner

sebastianruder commented Dec 30, 2019

Hi André, that's fantastic! Thanks so much for adding these recent papers! Good idea re the footnote. 👍

@sebastianruder sebastianruder merged commit fe823b9 into sebastianruder:master Dec 30, 2019
@Yorko

This comment has been minimized.

Copy link

Yorko commented Jan 27, 2020

How about SpanBERT? It's actually SoTA on OntoNotes

As for GAP (Gendered Ambiguous Pronouns) mentioned in the 3rd paper "BERT for Coreference Resolution: Baselines and Analysis", it's hard to speak about SoTA, there was a competition held with GAP data but it's not a scientific leaderboard (or what else is needed to claim SoTA), and winners presented higher performing solution. In our paper results reported for GAP are higher than in "BERT for Coreference Resolution: Baselines and Analysis" (~0.92 as opposed to 0.87 А1). However, ours are crazy Kaggle-style ensembles :)

@mosicr

This comment has been minimized.

Copy link

mosicr commented Jan 27, 2020

Second SpanBERT addition for OntoNotes. It might do well if fine tuned to QA too.

@rakeshchada

This comment has been minimized.

Copy link
Contributor

rakeshchada commented Jan 27, 2020

@sebastianruder : I just wanted to expand on @Yorko's note and add a reference to the work that I published in June 2019 at ACL workshop. I frame co-reference resolution as an extractive question answering task and show great results (similar or better than the one's shown on the Joshi et al. paper mentioned above) on the GAP co-reference dataset without any ensembling. Here's the link to my paper: https://arxiv.org/abs/1906.03695 . To the best of my knowledge, this was one of the earliest works to frame co-reference resolution effectively as an extractive question answering problem.

@sebastianruder

This comment has been minimized.

Copy link
Owner

sebastianruder commented Feb 2, 2020

@Yorko, thanks for the suggestions! Could you add SpanBERT to the OntoNotes page? Also feel free to add the winner of the GAP competition to the corresponding table.

@rakeshchada: Feel free to create a PR with your work for the GAP table.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Linked issues

Successfully merging this pull request may close these issues.

None yet

5 participants
You can’t perform that action at this time.