Alan Selman comments on Lance's blog about copyrights on conference and journal papers. Essentially, his point is that the common practice in many areas of computer science (but not all; see below) of submitting the same paper verbatim to both a conference and a journal is legally dangerous and therefore to be avoided.
Assuming that the publisher of the conference proceedings holds copyright of the papers that appears in the proceedings, the paper that one submits to a refereed journal must be substantially different from the one that appears in a conference proceeding in order to avoid copyright infringement. [...] The essential point is that the journal version must be essentially different from the conference version.
If Prof. Selman's argument had focused on self-plagiarism or resume-padding, I might have bought it. But the copyright argument just doesn't hold any water with me, for one simple reason—publishers shouldn't own the copyrights in the first place! Authors, not publishers, should be the final arbiters of the legal use of their written work. There should be no legal prohibition against my publishing my own papers as many times in as many different media as I like. Of course, I'd still have to convince all those editors, but that's a different problem.
(Once again, I'm tempted to submit my next conference paper with “This paper is in the public domain.” plastered on the bottom of the first page, just to see what ACM or SIAM or IEEE does.)
The argument is further weakened by the existence of short, self-contained, high-quality papers. See, for example, Ben-Or's classic paper on lower bounds for algebraic decision trees, or Seidel's paper on shortest paths. Should the authors of such short papers cripple them further for conference publication, or artificially inflate them for journal publication? Neither choice serves either the author or the research community.
I don't have any problem with the “common and acceptable practice” of sending virtually verbatim conference papers to journals, provided that paper appears as a single entry on the author's CV. I think good computer science papers should be published twice: once at a conference to make the result known to the community, and once in a journal to benefit from rigorous refereeing. But listing the same work in five different places on your CV is simply dishonest, even if it was published in five different contexts with five different levels of detail.
An anonymous commenter asks:
Don't most journals anyway have the requirement that the journal version be substantially different from the conference version?
In my experience, the answer to this question is no. At least in the journals where I publish, neither the editors nor publishers seem to care about duplicating conference papers, if they even notice.
However, I have been told by some systems colleagues that in their communities, republishing results in a journal that previously appeared in a conference is strictly forbidden; once a paper is published, it's published. And simply providing more details or running more experiments with the same system isn't good enough; journal papers must reflect new research. This makes it much more difficult for systems researchers to publish journal papers, since to have any impact they must present their work at prestigious conferences and publish it in the proceedings.
Most (all?) academic fields outside computer science treat conferences in a way that avoids this problem entirely—conference papers just don't count as publications. In many ways, this seems more sane than the computer science standard. Unfortunately for computer scientists who prefer sanity, it's not the computer science standard. For better or worse, computer science conference publications really do matter. (It's not entirely clear why computer science evolved this way; Lance thinks it has something to do with airplanes.)
Needless to say, the differing social standards for conference and journal publication, even within computer science, play absolute havoc with the tenure process.