Monday, September 25, 2006

Extended Abstracts

A non-theorist saw the upcoming STOC call for papers and asked about the following line.
Authors should submit an extended abstract (not a full paper).
and asked why we theorists only wanted extended abstracts and why she couldn't submit a full paper if it met the page limit.

An abstract just states the main results of a research paper. An extended abstract gives a little motivation, background and a hint of the proofs. The extended abstract serves not as a paper in itself but as an advertisement of the journal paper to come. If one had a full paper published in the proceeding than one couldn't publish the same results in a journal later on. That's the way conferences work.

Actually that's the way conferences work in all fields except computer science. In CS conferences play a more important role than journals; in most cases people judge the quality of a paper based more on the conference than the journal. Many conference papers never see a journal at all and many journal submissions are only slight variations of the conference proceedings version.

You really need to send something closer to a full paper to STOC if you hope to get accepted. You need to convince the committee that the results are new, important, correct and nontrivial. Most people submit a rough draft of a final paper, using the appendix if needed to put in their proofs.

Why do we keep the fiction of the extended abstract? Ethical and copyright rules prohibit us from publishing the same results in multiple places. We call the proceedings version an "extended abstract" so we can believe we don't violate these rules.

Non-theory conferences don't even pretend. The EC Conference, for example, solicits full papers though they do include the following note.

To accommodate the publishing traditions of different fields, authors may instead submit working papers that are under review or nearly ready for journal review. These submissions will be subject to review and considered for presentation at the conference but not for publication in the proceedings. These submissions need not conform to the conference paper format. Abstracts of accepted working papers will be included in the proceedings and must be coupled with a URL that points to the full paper and will be reliable for at least two years.

12 comments:

  1. In mathematics there is no such thing as a "conference paper". Such things would carry essentially no scientific merit, and there is no sense of urgency to push half-baked papers out the door to meet an artificial deadline.

    ReplyDelete
  2. Re: commnet #1, there are some exceptions, where announcements w/limited technical details _do_ appear in published, written, and highly respected form .It's just not very common and appreciated as in CS.

    ReplyDelete
  3. A puzzle: Name a well-known and respected theory conference the ratio of #(accepted papers)/#(PC members) is < 1/2.

    Disclaimer: I don't know the number of submissions, so I used # accepted papers.

    [Bonus point: Do it without looking again at lance's post.]

    ReplyDelete
  4. I must agree with Anon #1.
    It happens too often that an accepted paper to STOC/FOCS emerges as consisting of fatal errors.
    The problem is that, in order to meet the deadlines, conference papers are just not checked for correctness.

    I strongly believe that we, as a community, should be careful not to attribute these conferences too much credit, and should encourage all researchers to publish in journals.

    ReplyDelete
  5. This hype for conferences over journals is ridiculous, most conf papers are reviewed very poorly --nothing compares to journal reviews -- not to mention that when writing a scientific proposal to be evaluated by researchers from other disciplines, you're nothing more than a guy that hasn't been able to make any of his conf papers into a journal.

    ReplyDelete
  6. This hype for conferences over journals is ridiculous, most conf papers are reviewed very poorly --nothing compares to journal reviews

    Indeed, but the system as it is has very few incentives for spending the time to write up the journal version of the paper. The Goedel price comes to mind as one of those rare incentives, but chances of winning it are so remote that most people won't respond to the incentive.

    Over the years I've come to believe that allowing double counting of papers published in journals and conferences would be such an incentive, and to a certain extent the criteria for conferences and journals are not exactly the same. A conference paper tends to be ranked mostly on novelty and excitement, a journal paper on correctness.

    ReplyDelete
  7. What about arXiv preprints?
    Do CS conferences allow submissions of work that the author may have uploaded on arXiv (but unpublished elsewhere)?

    ReplyDelete
  8. Where did the fact norm that it's unethical to publish in multiple places come from?

    ReplyDelete
  9. What about arXiv preprints?
    Do CS conferences allow submissions of work that the author may have uploaded on arXiv (but unpublished elsewhere)?


    Yes. Both the ArXiv and similar sites such as ECCC. Frequently, authors will upload 'full' versions to ECCC of papers for which they have just submitted extended abstracts to a conference.

    The 'extended abstract' form is typically far more convenient for PC members to work from than a full paper would be. It puts the focus on explaining the results to the PC. Standard math style focusses on the steps of the proofs but not on why those steps are being taken. This makes them OK for line-by-line verification but not on understanding the arguments in the large. I think that having to write extended abstracts that necessarily focus on the bigger picture has been good for the community and has produced better final papers (and broad understanding) as a result.

    ReplyDelete
  10. Do anybody knows why the journal in CS work so extremely slowly? It is normal to wait 1 year to get just the review of your paper. If think the problem would be solved if one could send a paper to a good journal and have it published in about 3 months. Than there would be no need to use conferences as a way for fast publishing.

    I have a comparison with Physics, where a journal paper is usually published in 3 months and a rapid communication in 1 month. This has several advantages:
    - you do not have to stick to the strict deadlines,
    - conferences are much more interesting - usually contain a half of good quality invited presentations,
    - you do not have to have money to publish a paper fast - conferences are costly.

    ReplyDelete
  11. Do anybody knows why the journal in CS work so extremely slowly? It is normal to wait 1 year to get just the review of your paper.

    You are seeing the fast version. A decade ago two-three years wasn't unheard of. Going down to a year is progress, we need to move faster though, but realistically I doubt we'll get to times below six months to get your review...

    ReplyDelete
  12. In theoretical CS, the conference paper matters for the first 2-3 years after the result comes out; after only the journal version should matter. A conference paper that hasn't been reviewed in journal form by then should be thrown out for no credit :-)

    But: I agree with a previous commenter, the (lack of) promptness for journal reviews in CS theory is appalling!! It's impossible to get people to agree to review, and even when they say they will, they often don't or certainly don't in anything remotely approximating a timely fashion. Has it always been this bad in theoretical CS (it's much better in mathematics, for example).

    ReplyDelete