February 03, 2004

Attributional abduction again

Cosma Shalizi, always interesting, discusses a fascinating PNAS paper on emergent distributed computation in plants. (If you're a sociolinguist, or can play that role, you might read Cosma's summary and ask yourself whether there are "domains" and "particles" in patterns of linguistic variation and change, and if so, what computations they might be performing...) At the end of his piece, Cosma says

For another take, see this news piece in Nature ..., in which the usually-reliable Philip Ball (or his copy editor?) manages (1) to confuse the emergent particles with the basic cells of a CA, and (2) to say that Wolfram was the first to show cellular automata can "mimic computers", something established by John von Neumann and Stanislaw Ulam before Wolfram was born.

This is an all-too-familiar example of the problem of attributional abduction. A piece of traditional journalism says something very implausible, obviously wrong, or completely nonsensical. Is it there because the journalist was misled by a source? because the journalist misunderstood or misremembered something independently? or because the piece was subverted by an editor, accidentally in the course of hasty re-writing, or on purpose due to conceptual confusion or some independent agenda? In this case, Cosma has enough experience with the particular journalist to suspect an editor -- though the "copy editor" is probably not the most likely culprit in the editorial chain at Nature, whatever sins copy editors may sometimes be guilty of.

This case is also a good example of how useful weblogs as a form can sometimes be, in helping those outside a narrow subdiscipline sort out new scientific research. This is simultaneously despite and because of the lack of an editorial process. I certainly don't trust Cosma to get everything right all the time -- he'd be the first to tell me not to -- but I put a lot of credence in what he has to say on a wide variety of topics where experience has taught me that he knows a lot and has an interesting perspective. In this case, his blog gives me a clearer and more reliable summary of the Peak et al. PNAS paper than I can get from Nature. (Though I should say that the Nature piece seem pretty good overall, even if it sacrifices accuracy to simplicity in the two respects that he mentions). If he gets something badly wrong, or even if someone else thinks he has, he will most likely post an update or a link to the outside discussion; or at worst, if I care enough, I'll find out about other perspectives by looking at the pages that link to him.

We've made a few analogous contributions of our own, for example Bill Poser's evaluation of the Gray and Atkinson paper on dating Indo-European, or my discussion of the Fitch and Hauser work on monkey "grammar learning." As a result, some people have started looking to our site for reactions to newly-reported results in the sciences of language, just as I look to Cosma Shalizi (among others) for help in understanding new work in what one might call "natural distributed computation."

There's certainly a lot of garbage out there on the web. But a surprising amount of it is in the digital pages of reputable publications. And the low barriers to informal publication and re-publication, combined with (up to now) trustworthy information about authorship of such material, and the still-emerging mechanisms for establishing and navigating cross-links, combine to produce (the beginnings of) a dynamic, distributed information source that can be more reliable than the major outlets of science journalism are.

Posted by Mark Liberman at February 3, 2004 07:20 AM