2019Q3 Reports: Publications Chairs
Summary
The publication chairs produced the proceedings of the main conference, tutorials, demos and SRW; and coordinated the work of all book chairs for the individual workshops. We mostly relied on documentation from aclpub (https://github.com/acl-org/acl-pub) as well as on a very useful FAQ written by the advisory publication chairs (Kevin Gimpel and Shay Cohen).
Changes from last year
We made changes to the style files: 1. We fixed the bug that had been plaguing the ACL community for a while now, caused by inappropriate handling of page wraps. 2. We fixed other minor bugs in handling e.g. colors across page wraps in footnotes and removed obsolete latex comments (as per l2tabu - https://ctan.org/pkg/l2tabu). 3. Appendices were included after the references in the main PDF of the camera-ready version, rather than as supplementary material. At submission appendices still had to be uploaded as the supplementary material. The different handling of appendices at submission and in the final versions caused some confusion with the authors, and this difference should be stressed more in the future *ACL call for papers.
We did not incorporate DOI, as was done in 2017 (but not in 2018), because we did not have clear documentation on what to do there, and the general advice received was to bypass DOI.
We did not have separate proceedings for long and short papers, but only main proceedings. This decision came after consulting with the general chair and the program chairs who were all in favor of having a single volume for the main conference. This facilitated the entire compilation process in light of the fact that both long and short papers are submitted to the same START page and with separate proceedings they have to be manually separated on START (e.g., separate order files are needed). However, a lot of older documentation still assumes that long and short papers are separate.
Problems/Observations/Suggestions
The main issue this year was that with the growth of ACL, the role of publication chair is just too much work for two people. If this trend continues, future editions of the conference require at least 1-2 additional people for this role. For instance, this year we had to take care of 660 papers for the main conference only (and the number of workshops and the number of workshop papers is also on the rise), which is a large increase compared even to some recent conferences. For instance, this is an increase of more than 100% in the number of papers compared to ACL 2016, while ACL 2016 employed 3 publication chairs. Manually checking all the proceedings (main conference and workshops) was a massive amount of work. What is more, we found style violations in one in every three (!) papers, and manually had to contact the authors to get these fixed. We strongly recommend bringing the paper check form back in START, so that people can and should check margin violations by themselves (and be penalized for not obliging). We also strongly encourage the development of some useful scripts that could facilitate the checks and automatise the process for some standard errors (e.g., margin violations, table captions above tables, font size, incorrect font size for figure and table captions).
Instructions around the copyright form are unclear. The publication chair instructions make note of having to check that author names and addresses are correct. However, very many papers did not have any filled out copyright information at all. We strongly recommend making both the name and address fields required in START in future editions. This will reduce the communication overhead between the publication chairs, the workshop organizers and the final authors.
There are subtle differences between style files from different conferences. It was useful to have a kind of master style file, for which we used NAACL19, but we then had to incorporate lots of ACL-specific things from the ACL18 style files. We recommend that the ACL community centers on one single style file, where we basically have \conferencename and \conferenceyear commands that are updated, and everything else stays the same as much as possible. We also suggest for the style files to fix the gender imbalance in citations included in the current files.
The existence of a Word template makes it harder to check the proceedings for correctness, because they look visually wrong when inspecting the proceedings. We recommend NOT using a Word template in the future. The rationale for having this made sense in the past (e.g., people from other fields might not be able to use LaTeX), but nowadays it is extremely easy to write papers even for LaTeX-novices on e.g. Overleaf, which in our opinion makes Word superfluous.
We need better instructions for workshop organizers, especially around the metadata. Despite plenty of preventive information obtained from the advisory publication chairs, the workshop organizers still did not use the correct Abbrev, did not check their own .bib files, or location and month information were inconsistent, etc. One suggestion based on our experience is to perhaps seed this automatically with appropriate information in START, so that people basically only have to enter their own names? For instance, if we are principled with workshop names, and generate bibtex URLs, abbreviations and ISBNs for workshops once they’re accepted, we can automatically fill in almost all metadata, in the correct way.
Names of authors and titles of papers are often incorrectly cased. START offers a hacky solution to fix some of this, but the solution is incomplete (it does not fix the index), so we had to do a lot of this manually. In the case of author names, we think that START should just use the correct casing automatically, or notify authors if they diverge. To standardize casing for titles, the START solution is somewhat functional, but still buggy (it seems to rely on a faulty POS tagger). We also had quite a few abstracts that contained LaTeX commands, which had to be fixed manually.
We were asked by the PCs to include links to TACL papers in the proceedings. While this is certainly a useful practice, this was made very problematic by the fact that not every TACL paper had a corresponding link yet. We also had to hack the program.tex underlying LaTeX code to allow it to use \href without erroring out. All of this created a lot of extra work and required last minute changes as preprint URLs became available.
We got a lot of requests for last-minute camera ready changes, and the policy on how to handle such individual cases and when to stop accommodating for such requests is unclear. We spent a lot of time replacing the PDFs on START with the corrected versions after the official camera-ready deadline.
Another part where some improvement is necessary is handling the mismatches between the START metadata and the data in the actual uploaded PDFs. Some authors have made changes to the list of authors and the titles only in PDFs without ever reporting the changes to the publication chairs and the program chairs. This caused the mismatch between the data on START and the data in the PDF files. What is more, the change of the metadata on START can be done only by the program chairs, while many such requests were sent to the publication chairs. We feel that a general policy on how to handle these requests is needed in the future.
Generating the final proceedings is very easy, using the instructions on acl-pub (note that there are duplicate instructions at https://github.com/acl-org/ACLPUB/blob/master/anthologize/README.md - we should probably only keep one of those before they start diverging). One problem, however, is that the scripts do not easily facilitate small changes. It would be good to have download-proceedings.sh accept single acronyms to upload new versions of proceedings while everything else stays the same (currently, the script complains about data/ already existing). We also had minor issues with sharing the proceedings with the Antology director via Dropbox - it might be useful to standardize the way that this sharing happens.
There were problems interacting with the PCs and handbook, web and app chairs, centered around the latest version of the order files. We would recommend tracking this on a git repository somewhere. In our case, things were made more problematic by the program.tex file not corresponding to the order file, because we had to make in-place edits to support TACL links and last-minute changes to the program. One solution might be to use program.tex as the master for handbook/web/app creation, rather than order.txt?
Related to the above, while the instructions for pub chairs are great, it is unclear what to do for the handbook, website and app, and how these relate to pub chair duties. There has definitely been some re-inventing of the wheel between chairs, e.g. for getting sponsor logos, etc. It would be good to have more visibility on what the other chairs are doing/supposed to do.
Finally, we plan to update and publicly release the extremely useful FAQ document prepared for us by the advisory publication chairs Kevin Gimpel and Shay Cohen. In fact, this document was one of our primary sources of knowledge and practical know-how, and it helped us navigate through the entire process and steered our communication with the handbook chair, the program chairs, the workshop chairs, the tutorial chairs, etc.