Handling retractions

[mjpost]

Following up on the discussion here, this PR implements retractions. It:

• Adds a <retracted/> tag to mark the paper status
• Processes the PDF explanation as a revision, so that it gets returned by default when accessing the PDF
• Strikes the paper in the volume and event listings, and suppresses it entirely from the author listing

Comments welcome!

[mjpost]

Here are a few screenshots.

Volume listing:

Paper page:

Author page (absent):

[aryamccarthy]

Me gusta. Will we still serve BibTeX (and MODS, etc.) for retracted papers?

[mjpost]

Hmm, good thought. The ACM policy doesn't have anything to say about this directly. At the very least, perhaps we should add "[RETRACTED]" to the title. But since the purpose is to effectively remove the paper from the body of scientific literature, it seems fitting to me to remove citation formats, too.

[aryamccarthy]

My thinking was similar. The one argument against is if someone wants to cite X article as retracted, e.g. in some meta-analysis of retracted papers.

[mjpost]

In that situation, they'd just have to refer to it by title, or URL. They'd need a custom bib format anyway, since it is no longer part of its original proceedings.

[akoehn]

Looks good, the only thing that is bothering me a bit is that the retraction notice is linked as "PDF (v2)" and not as "Retraction notice". The second text is a bit to long OTOH, so I am okay with "PDF (v2)".

Maybe also change "can be found [here]" to "can be found [in this retraction notice]" because otherwise people could assume that "here" just links to a page about retractions in general.

[davidweichiang]

I feel that this whole issue is something that should be run by the ACL Exec.

[mjpost]

This is a good idea. I will write up a policy and incorporate our plans here for Exec approval.

@akoehn, on the technical side, it is tricky to change the name of just the last revision in the template. An alternative:

• Modify the <retracted> tag to work more like <revision>, taking a date and the file name, e.g., "2020.acl-main.563.retracted.pdf"
• This file will also be copied over top of the main file, aux revisions
• Then I can add a button labeled "Retraction Notice" or something to this, independent of the revisions

[mbollmann]

Is it desired that the retracted paper itself is still accessible? In your approach it would be (as "v1"), no? That is not currently the case with existing (yet unlinked) retractions on the server.

[mjpost]

Under the ACM policy, which I think makes sense, the original retracted paper remains accessible.

I think handling the back-catalog of (suspected or confirmed) retractions is a separate issue. I don't know that we'd go through the effort of bringing them into line with the new policy we adopt.

[mjpost]

This has come up again with EMNLP. I am making this one of my agenda items for the next ACL meeting, but in the meantime, this needs to be dealt with, and I don't think we're doing any permanent harm while we wait for a formal policy. Everything here can be altered.

After consulting the arXiv withdrawal policy, I propose to do the following:

• Display the paper as above with strikeouts
• Strike out in all paper listings
• Remove from author listing
• Add a <withdrawn> tag marking the date
• Process the paper as a revision. The revision will be the previous version with a retraction notice stamped on the top
• Continue to serve citation materials No longer serve citation materials

[mjpost]

Updated to use a dummy example

I've written some code to produce a watermarked version of the file. It creates a revision of just the first page, watermarked, and a link at the top to the paper page.

[github-actions]

Build successful. You can preview it here: https://preview.aclanthology.org/retractions
This preview will be removed when the branch is merged.

[mbollmann]

This LGTM now, but I noticed that on the preview branch, only the 2020.acl-main.563v2.pdf carried the watermark, while 2020.acl-main.563.pdf (without versioning) was still identical to the v1. Not sure if this is something you may need to handle explicitly in your retraction script @mjpost.

[mbollmann]

Direct link to retracted paper on the branch preview for convenience.

[akoehn]

A bit late, but this looks very good indeed.

@mjpost as you probably want to update the retraction script anyway: currently it assumes that there is only one version and will create a "v2" even when a v2 is already there. This will probably be a very rare occurrence, but this also means the person using that script will probably not have this peculiarity in mind.

[mjpost]

Thanks! I'm on vacation but will update the script when I return and then merge this.

[mjpost]

This has been approved by the ACL Exec. I plan to update the documentation today and would then like to merge in short order, if anyone has time to review.

[mjpost]

@akoehn my latest push addresses your concerns (the revision piece is now automatically handled).