Quoting Jörg Rhiemeier <joerg_rhiemeier@...>:
> Hallo!
>
> Andreas Johansson wrote:
>
> > [...]
> >
> > It can be read here:
> >
> >
>
http://www.psych.auckland.ac.nz/psych/research/Evolution/Gray&Atkinson2003.pdf
> >
> > Turns out I was misremembering - it shows Germanic+Italic as the sister of
> > Celtic. Next group out is Balto-Slavic.
> >
> > An odd feature is that it had Sinhalese+Romani as the outgroup of all other
> > Indic languages bar Kashmiri. It's very tempting to suspect a linguistic
> > equivalent of lateral gene transfer (ie. borrowing) at work here.
>
> I finally found the time to read it, and I noticed a number of problems.
>
>
> Second, and this is a much bigger problem: what Gray and Atkinson do
> is glottochronology. And glottochronology is no longer accepted
> by the vast majority of historical linguists. This is because
> glottochronology is based on the assumption that the rate of
> lexical replacement is constant across time and languages
> - in reality, it is neither. For example, languages in intense
> contact with other languages (e.g., Middle English) replace much
> more vocabulary than geographically isolated languages (e.g.,
> Icelandic). Gray and Atkinson address some "minor" problems with
> glottochronology which they claim to have overcome with some
> advanced mathematics they call a "Bayesian Markov chain Monte Carlo
> model"; but I don't see how that can remedy the problem that the
> basic assumption of glottochronology - that lexical replacement rate
> was constant - is false. It is still glottochronology.
That's not what tey use bayesian Markov chain Monte Carlo for. What they do is
replacing the assumption of a constant change rate with the assumption that
equal rates are the likeliest - if the likelihood penalty from varying rates
than the gain from achieving a more likely topology, the model will produce
unequal change rates in different branches.
(I'm not saying this works; I'm not in a position to judge. What I can say is
that if it does not work, the reason is not the mere fact that lexical
replacement rates are not constant.)
> Third, the age they assign to Proto-Indo-European is impossible.
> Any archaeologist will tell you that the wheel wasn't invented
> yet 8000 years ago. Yet, a PIE word for `wheel' is reconstructed
> with as much certainty as is possible in this discipline. And also
> words for `yoke', `wagon', `carry by wagon', etc. This means that
> Proto-Indo-European can hardly be older than 6000 years.
[snip]
If their time-depth is wrong, it doesn't necessarily mean that their topology is
wrong.
Andreas