During the review process for ICALP track B, a colleague wrote to me saying that he was planning to submit a paper longer than 70 pages to the conference, but eventually decided not to do so. While he was considering submitting the paper, this colleague was naturally wondering about the implications of submitting such a long paper to a conference with a 12-page limit on submissions. This academic told me that he thought that it would not be unreasonable to reject a very long paper submitted to ICALP without even reading it. Such a decision could be backed up by a notification notice stating that "the paper is unverifiable given the available time and resources" or even "considering the paper is pointless since only 12 pages will be published."
I can understand very well why this colleague decided not to submit the paper to ICALP. Shrinking a 70+-page paper to 12 pages is a major effort, and one has to wonder whether a conference is the right outlet for such a lengthy piece of work. Indeed, one may argue that our conference publication structure does not lend itself to the publication of (very) long papers. However, I know of at least three exceptions (listed here in chronological order).
- The ICALP 1990 paper in which Daniel Krob presented his equational axiomatizations of equality of regular expressions was 14 pages long, but the journal paper Complete Systems of B-Rational Identities (solving two problems posed by John Horton Conway in his monograph Regular Algebra and Finite Machines) that appeared in TCS in 1991 was 137 pages long.
- The ICALP paper in which Sénizergues introduced the decidability of DPDA equivalence is only 10 pages long, but the journal paper is 166 pages long.
- The short paper by Martin Grohe at http://www2.informatik.hu-berlin.de/~grohe/pub/gro10.pdf was published in LICS 2010, but the full work takes a book that is still under development. See http://www2.informatik.hu-berlin.de/~grohe/pub/cap/index.html.
Now, one might ask what the purpose of those conference papers is. In some sense, my colleague was right in saying that they are "unverifiable" (given the length and time constraints imposed by conferences). I doubt that the conference reviewers went through all the details of the full versions of those papers, when they were available. Even a journal reviewer would probably not do so.
I guess that the answer is simply that the conference versions announce the results and "mark the territory" by saying "I did it". However, IMHO, the results only stand after the interested community has not found any serious errors in the full versions of the papers for a long time.
A separate issue is that of writing a conference paper based on a long full paper, which does justice to the main results and techniques presented by the authors in the full version in all the gory details. IMHO, conference papers reporting on very long and technical developments are "trailers" for the full version of the story, which is told elsewhere in all its glory. As a movie trailer, the conference paper serves the purpose of enticing potential readers to check out the full version by motivating the work, putting it in context, stating the achieved results, discussing their importance and giving a high-level sketch of the techniques and of the tools involved in the proof. (I am thinking here, for example, of the above-mentioned paper Grohe published at last year´s LICS, where he did precisely what I wrote above and, to my mind, did it well.) Of course, this is easier said than done..
