At 6:10:33 AM on Tuesday, July 15, 2008, Patrick Ryan wrote:
> Michael Witzel has been kind to make available Pierre
> Bancel's critique of Ringe to members of MTLR.
> Anyone like Brian who supports Ringe should read:
> http://geocities.com/proto-language/Ringe-Bancel-Critique.pdf<http://geocities.com/proto-language/Ringe-Bancel-Critique.pdf>
Okay, now I've read it. Bancel is a statistical moron. I
quote:
If one throws two dices together, each of them giving
1 / 6 chance to get an ace, there are two chances out of
six (1 / 6 + 1 / 6) to get an ace at least, ...
Even students in baby probability and statistics courses,
the kind with no algebra prerequisite, learn not to make
*that* elementary blunder.
He also obviously does not understand even the (rather
basic) notation used in the formula that he misquotes from
p. 417 of the review, let alone its meaning.
Most important, he clearly doesn't understand Ringe's
argument, since his 'refutation' is quite beside the point.
He's correct when he points out that Ringe's numbers do not
produce probabilities; what he fails to understand is that
they were never intended to do so.
Bancel wants to look at the probability of getting a cognate
set (really a 'cognate set') when one picks a word at random
from each family; this is of course very small -- and
completely irrelevant, since even mass comparativists don't
pick words at random. When doing comparative work of any
kind, one actively looks for candidates for potential
cognate sets; the more there are, the easier it will be to
come up with some. Ringe's calculations demonstrate that
Greenberg's criteria are loose enough to guarantee the
existence of a very large number of 'cognate sets' even if
the languages are completely unrelated; thus, anyone looking
for them will find some.
On the linguistic side, Bancel clearly doesn't understand
why 'reaching down' is problematic and effectively increases
the number of families being compared.
(By the way, there are some mathematical errors in Ringe's
1999 CVC-roots paper, a fact that I've pointed out before
elsewhere, but they don't qualitatively affect his
conclusions, and they have nothing to do with the material
in his review of Greenberg.)
Brian