03-12-03 11:20, S.Kalyanaraman wrote:
> That is right, Pete. The methodology is flawed ab initio. The sample
> is not only small but is a biased, non-random sample from the
> lexical repertoire of the languages involved. One can't draw
> conclusions about the population (language relationships, in this
> case) based on such a sample.
The sample is not that small (2449 cognate sets in 87 languages), and
the algorithm used is sophisticated enough to deal with some of the
obvious pitfalls of glottochronology. The authors were careful to
estimate some possible sources of error, perform consistency tests and
calculate the probability values supporting each grouping. They used
Dyen's database of 200-word Swadesh lists, which is not the worst choice
one could make, since the Swadesh items are typically more resistant to
borrowing than less basic vocabulary (if we chose completely random
words, English would appear to be a Romance language; technical and
cultural vocabulary diffuses more readily than anything else). Words
judged to be borrowings by Dyen _et al._ were excluded from the
analysis. One could argue that since the Swadesh list was intended to
include "universally basic" terms, it would be more profitable to use a
modified list for studies of a language family whose ancestor is known
to have been used by a neolithic society of farmers and pastoralists,
and its basic vocabulary included some characteristically neolithic
terminology. The exclusive concentration on lexical "characters" is a
serious shortcoming from the point of view of a linguist, and as far as
I'm concerned allows one to treat the analysis only as a preliminary
investigation, but even so the results look promising.
> The error gets further compounded by using the questionable premises
> of semantic or phonetic distances sought to be gleaned through
> linguostatistics and glottochronology.
I've no idea what you are talking about.
> The biggest mistake occurs from trying to extrapolate the recent
> perceived relationships into the time periods as deep as the 7th
> millennium Before Present.
The "perceived relationships" are products historical processes. They
are not "extrapolated" back into the past; they just preserve some
information about the origin and history of genetic groupings.
Piotr