Sunday, November 03, 2019

Call for opinions: Length of papes in conference proceedings in TCS

As current chair of the editorial board of LIPIcs, Leibniz International Proceedings in Informatics, I have been looking at some data about the length of the papers published in the series. The average length of the articles published in LIPIcs in 2019 so far is of 15.8 pages, including front matter and bibliography. However, three of the published papers are over 40 pages and three conferences have an average article length above 20 pages (22, 23.9 and 28.4, respectively).

I have seen that some conferences in TCS have no limit on the length of the submitted papers. A colleague whose opinions I hold in high esteem wrote to me saying:
Some people in the "conference name removed" community strongly feel there should be no page limit. My opinion may not be as strong as some, but I believe these people have a point.
Whether we like it or not, most papers in the TOC community only appear in conferences. Many of those conferences have no page limit for submissions, and encourage authors to include all proofs while making sure that the key ideas are presented in the first 10 pages or so. Scientific progress is hindered when authors are then forced to take out parts of their writeups for the conference proceedings in order for their papers to fit within the page limits. One may wish that they'd publish a full version of their paper in a journal subsequently, but most of them won't bother. The net effect is that there is no actual paper with all the details, which is no good.
I share the point made in the last sentence, but I am somewhat bothered by the fact that full versions of papers in TCS are increasingly not being submitted to journals. I am probably very old fashioned, but I feel that it is desirable to see our published results vetted by a journal-strength review process. I still view conferences as means for the rapid dissemination of results and as a meeting point for their community of reference, and I consider journals as the media for final archival publication of mature pieces of research. However, with my LIPIcs EB chair hat on and out of personal interest, I am keen to hear your opinion on whether it is good for the TCS community to publish only in conferences and to publish conference papers without page limits, bearing in mind that, quoting from the FOCS 2019 call for papers:
Although there is no bound on the length of a submission, material other than the title page, references, and the first ten pages will be read at the committee’s discretion.
I'd be grateful if you could post your thoughts on this matter in the comment section. Let's focus on what is best for the dissemination of science, even though long articles are more expensive than those whose average length is around 15 pages.

I look forward to hearing your opinions. Thanks in advance!


JS said...

I think what LIPIcs is doing is a good strategy:

1. There is a page limit (so it is easy to prepare an extended journal version that has additional content in comparison with the conference version).

2. Any reference to the "full version" has to be accompanied by an arXiv reference or something similar (so the full version is always available somewhere, even if the authors do not ever prepare a journal version).

I think full-length conference papers are bad, as they make journal submissions more difficult. Many publishers have a policy that the journal version has to contain some substantial amount of new material in comparison with the conference version. Hence if there is already a full-length conference version, it is going to be a nontrivial effort to prepare a journal submission.

Hans said...

Page limits have traditionally served two purposes. One is obsolete, namely that printed proceedings were costly. Another is still sensible: There is a limit to the effort that one can reasonably expect from a reviewer.

There is already a well-established approach to dealing with this. The proceedings of ACM conferences such as POPL and SPLASH appear in the journal ACM Proceedings on Programming Languages and papers get a status as both conference papers and journal papers. In my opinion, this is a preferable solution. The page limit is 25 pages plus bibliography, which is usually enough if one wants to include background material and the gory details of proofs.

Luca Aceto said...

Hans, I do welcome the approach taken by POPL an other PL conferences that publish their proceedings in the journal ACM Proceedings on Programming Languages. However, one has to bear in mind what you pointed out in your comment, namely "There is a limit to the effort that one can reasonably expect from a reviewer." This is true at the best of time and is exacerbated by the tight time limits of conference paper reviewing.

In his CV available here, Derek Dreyer writes:

"Note: Since August 2017, ACM SIGPLAN has published the proceedings of POPL, ICFP, and OOPSLA as special issues of the newly-formed journal PACMPL (Proceedings of the ACM on Programming Languages). The rationale behind the creation of PACMPL is that conference publications in programming languages are of the same quality—and reviewed with at least as much diligence—as journal publications in other scientific disciplines, and as such should be counted as journal publications. Although I concur with this rationale (and in fact supported the transition to PACMPL when I served on the ACM SIGPLAN Executive Committee), there is nonetheless a qualitative difference between PACMPL publications and traditional long-form journal publications in programming languages: PACMPL publications must obey strict page limits, and the PACMPL reviewing process is tightly constrained by the demands of conference reviewing. For this reason, I list my PACMPL publications below together with my conference publications, even though formally they are journal publications."

IMHO, this is very sensible.

Let me also note that I am told that long papers cost more to publish than "standard-length conference papers" even in a digital form, as they require more effort from the LIPIcs publishing team.