Austronesian tree in _Nature_

[Michael Cysouw]

----------------------------Original message----------------------------
Dear Histling,

In a recent volume of _Nature_ (Vol. 405, 29 june 2000), there are two
articles on a quantitative study of the family-tree of the Austronesian
languages. There is one 'letter to nature' by R.D.Gray & F.M.Jordan from
the department of Psychology in Auckland, entitled 'Language trees support
the express-train sequence of Austronesian expansion' (pp. 1052-1055) and a
'news and views' comment by Rebecca L. Cann, from the department of
Genetics and Molecular Biology in Honolulu (pp. 1008-1009).

After a first jump of joy that _Nature_ would publish articles on
historical linguistics, my first caution was raised by the affiliation of
the authors, none of whom seems to be a linguist. And indeed, in my opinion
(please note that I am not a specialist in this field!), the articles are
rather tendentious towards historical linguistics. The authors are very
eager to proclaim that their quantitative methods, which are taken from
biology (cf the resent discussion on cladistic classification), are
important, or even better than the methods used by linguistists:

'so, although linguists routinely use the "comparative method" to construct
language family trees from discrete lexical, morphological and phonological
data, they do not use an explicit optimality criterion to select the best
tree, nor do they typically use an efficient computer algorithm to search
for the best tree form the discrete data.' (p. 1052)

The authors used 'an efficient computer algorith' on the unpublished data
from Blusts's Austronesian Comparative Dictionary to build a language-tree
of the Austronesian languages. As far as I can see, nothing new results
form their analysis. There is a rather nice congruence between their tree
and the tree as I knew it from the literature.

'We found that the topology of the language tree was highly compatible with
the express-train model.' (p. 1052)

So, the method seems to work. Yet, a main error they seem to make is that
they think that linguistic data are 'discrete', that there is a clear
yes/no answer to the question whether a part of words is cognate or not
(see also the methods-section, p. 1054).

In general, I grew rather angry after reading the articles. The method is a
nice addition to historical linguistics, but there is nothing really new.
So, it seems to be possible to publish an article in _Nature_ just by using
the right computer programm and forget that many years of research that has
been performed in linguistics to be able to perform these analyses. I even
started to wonder whether _Nature_ has had the article reviewed by a real
historical linguist.

Of course, it is a good thing to have some historical linguistic research
published in _Nature_, and maybe this is the best possible outcome for
something to get published in such a journal. Still, I would have liked to
see some more credit to the linguistic effort.

Michael Cysouw
University of Nijmegen