Wednesday, April 13, 2005

A Modest Proposal

A guest post by Michael Mitzenmacher

Lance nicely invited me to expand on my views on the format for conference submissions. Currently, I am on a program committee using the standard theory call:

A submission for a regular presentation must be no longer than 10 pages on letter-size paper using at least 11-point font…additional details may be included in a clearly marked appendix, which will be read at the discretion of the program committee.
We have actually had discussions on whether to reject out of hand papers that use 10 point font or otherwise violate this standard.

The problem is that many, including myself, think that this formatting rule is silly, and so it has been widely ignored or at least painfully abused for many years. I would like to propose a simple and logical alternative: conference submissions should be in the same format (or as near an equivalent as possible) as the final conference version. Many other conferences (such as AAAI and Sigmetrics) use this approach with great success.

The advantages of this approach include:

  1. It reduces the work of the authors. Right now, authors have to create entirely distinct submission versions and final versions of conference papers using various formats. Most authors find this a hassle, and this is my main reason for the proposal. I hate writing the same conference paper multiple times just to cope with formatting issues.
  2. It gives the reviewer a more accurate picture of the conference paper. Reviewers will have a very good idea of what the paper will look like in the conference proceedings, making it easier to judge. When you're staring at 20+ pages of appendices, it is hard to tell what the final paper will look like.
  3. It enhances fairness. Because this is a standard with a clear reasoning behind it -- you cannot have a longer submission than conference paper -- people are more likely both to follow and enforce the rule, avoiding potential unfairness.
I have heard of some disadvantages of this approach. Let me attempt to dispense with them.
  1. The format is too hard for the reviewers to read.

    My response: If this is the case, then perhaps the conference paper format itself should be changed -- after all, don't we expect many people to actually read the conference version? If the conference paper is packed tight for other reasons (the publishers charge by the page), then for submissions design as near an equivalent format as possible. If we find 10 double-column 10 point pages with style file A essentially equals 20 single-column 11 point pages with style file B, then clearly state that in the call and ask for the latter. (Luca Trevisan pointed out this is done for the Complexity Conference already.)

  2. Appendices are necessary when there are long proofs that won't fit in the paper.

    My response: If the proofs won't fit in the final conference paper, this is something a reviewer should see and know. The program committee can either allow appendices, with the knowledge they won't have room to appear, or allow pointers to more complete versions (TRs, arXiv preprints) that the reviewers can examine if they desire.

  3. By having different formats, we force authors to revisit and hopefully improve their paper.

    My response: Nice intentions, but don't people already want to make their published work as good as possible? This seems unnecessary, and not worth the price.

I ask all program committee chairs to please consider this modest proposal.

6 comments:

  1. Mild recommendations, very mild. The title is misleading...

    ReplyDelete
  2. I am one of those that hate to format (as an author) or read (as a reviewer or simply a reader) anything published in 10pt or smaller font, with two columns crammed into a letter-sized paper.

    I agree with Michael on one aspect though: it is pointless to have authors format their same work twice.

    The present proceedings format are a vestige of a bygone era: scientists depended on publishers more than publishers depended on scientists. The result is the type of proceedings formats that we see --- increasingly ridiculous by the year. Several conferences use a "standardized" ACM style where papers are printed in --- I kid you not --- nine point font.

    Just what is the point of even a "uniform format" anyway? Am I going to be upset if I open a conference proceedings and if papers are formatted in different styles? Am I going to not read a paper because it doesn't look like the others in the proceedings?

    My displaymath and eqnarrays need good spacing around to breathe, I can only take so many hyphenations per page.

    There was a good reason why -- despite the publishers' insane requirements -- program committees of previous years settled on certain formats: simple-to-format, easy-to-read, and succinct to state in the CFP.

    In my ideal world, conference submissions will simply not state any formatting constraints. Let the authors be the best judge of how their work will be presented. It is not in authors' best interests to format their paper in any way that hurts their chances --- horribly small fonts, ridiculous margins, inconsiderately organized abstract where the main ideas are buried in page 37. In response to a CFP of this sort, my ideal submission would be formatted to be easy on the eyes, and the ideas would be explained in a way that the authors feel would best explain the significance of their results (eg., a one-page abstract, followed by a two-page high-level summary including comparisons to related work, etc., followed by a four-page overview of the technical ideas, an eight-page clear-and-careful exposition of the work, followed by technicalities, proofs, etc.).

    Publishers, conference organizers, professional societies, and the scientific community need to seriously rethink this whole business of conference proceedings in an age where it probably makes more sense to take advantage of the digital medium (CDs, internet, etc.), and restrict the paper proceedings to contain two-page abstracts of the talks.

    Some conferences have taken formatting guidelines way too seriously: summary rejection of abstracts not in line with the guidelines (I once got a review from a non-theory conference that had the sentence "Does not follow line-spacing guidelines"; I had a paper slightly shorter than their required limit, so I stretched the linespacing ever so slightly so that it was more pleasant to (my) eyes.).

    I, for one, am extremely happy that theory conferences don't take themselves too seriously in these aspects of the review process. I hope that it continues to be as simple as it is today, perhaps even simpler, without any formatting guidelines, leaving the authors to choose their format, just like they choose the logical structure of their submission. If form ever comes in the way of content, I would trust the creator of the content to decide what is best.

    D. Sivakumar

    ReplyDelete
  3. I am sure formatting is an important issue. At the same time, I would love to hear comments from Lance and his readers about two other more serious conference-related problems:

    1. A randomly generated paper gets accepted at a junky conference. The problem is that the conference has been run for 9 years, and MANY people put papers appeared in this conference's proceedings in their CVs [just do a google search]. There are people from some pretty decent CS departments.

    Is CV-padding as easy in other disciplines like Math, Physics, etc.?

    2. The problem discussed by Graham Cormode, Artur Czumaj, v� Muthu Muthukrishnan in their hilarious article [PDF].

    ReplyDelete
  4. You might want to explain what a PC is, or at least expand the abbreviation the first time. To me it means "Personal Computer" or "Politically Correct."

    ReplyDelete
  5. Requiring that submitted versions look like final versions would only hurt the community unless we decide to relax standards for final versions!

    Do not suggest AAAI/IJCAI's rules as a standard we should emulate!
    The process of cutting a paper down to produce the 6 page two column 9 point versions that AAAI insists on is ridiculous. (One can buy 2 extra pages for the final version but one is not allowed to do this for the submitted version so that everyone is on a level playing field for the program committee.)

    I think that most of us find it easier to write a draft of a full paper making sure that all the details are correct. Cutting a longer paper down to fit the page limits is by far the most painful part and can take a couple of days (by which time the submission deadline has passed!) Allowing appendices makes this a triviality. Insisting on shorter versions would merely discourage people from bothering to write out all the details before they submit. As a PC member I'd rather have more of the details available if I want to see them (and greater likelihood that someone has worked them out).

    More than once I have paid an extra $100 or so for an extra page in the proceedings solely in order to save the effort of a day of work to cut a paper from 11 double-column pages to 10.

    One thing you didn't mention is how long submissions (and final papers) should be or what format they should be in. Should the limit be 6 pages, 10 pages, 15 pages? Double column? Single column?

    We do not control the final format for published versions of papers simply because the publishers insist on them (acm/ieee/llncs etc).
    These formats are very poorly suited for the kind of reading required for PC's. Maybe we can lobby publishers for better formats but, until we succeed, I'd like to keep the submissions and final formats separate.

    P.S. A HINT Trying to fit your paper into two column format? Forget those eqnarray*s and lefteqn! \usepackage{amsmath} and use align*.
    It is much easier to use, more flexible, and doesn't put quite so much extraneous white space on the lines.

    ReplyDelete
  6. I expaned the all the "PC"s (program committee). I wouldn't want to give anyone the impression that this weblog is politically correct.

    ReplyDelete