LINGUIST List 6.215

Tue 14 Feb 1995

Disc: Comparative Method

Editor for this issue: <>


Directory

  1. , Re: 6.168 Comparative Method, Speech Error
  2. Jacques Guy, Comparative method vs DNA

Message 1: Re: 6.168 Comparative Method, Speech Error

Date: Wed, 8 Feb 95 15:56:44 ESTRe: 6.168 Comparative Method, Speech Error
From: <amrphoebe.cs.wayne.edu>
Subject: Re: 6.168 Comparative Method, Speech Error

Paul Purdom makes some excellent points re DNA sequences and languages,
but it would not be fair to suggest that those of us interested in
language classification are necessarily totally ignorant of the work
in biology. Long discussions with some of the leading biologists in
the English-speaking world on this very point took place during the
Evolution conference at Cold Spring Harbor a few years back, and I
am afraid that the biologists there seemed much less sanguine about
assuming that linguists simply need to emulate them. There are major
differences having to do with the nature and quantity of the data,
as well as with the fact that we in linguists can enjoy the benefits of
the regularity (or even quasi-regularity) of sound change, which seems to
have no analogue over there. In brief, the linguistic work is much much
easier and much more qualitative, and possibly more conclusive.

Alexis MR
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue

Message 2: Comparative method vs DNA

Date: Mon, 13 Feb 1995 10:31:30 Comparative method vs DNA
From: Jacques Guy <j.guytrl.OZ.AU>
Subject: Comparative method vs DNA

It is true that linguist would do well to look at and
learn from the methods used in biology. But not blindly.
The fundamental question, which have never seen tackled
anywhere, is: is this or that classification algorithm,
this or that measure of distance, congruent with the
properties of the linguistic data, of its transmission?
 There is no value in applying methods from other
disciplines to language without asking and trying to
answer those questions first. I am fond of quoting
a 1965 article by Milke in which he dazzled his audience
by dissecting some Polynesian lexicostatistical data
in ways then unheard-of ("Experiments in Matrix
Reduction". Lingua 1965, 14:443-445). It was, in fact,
factorial analysis by the centroid method, designed
for the analysis of psychological data, with properties
strikingly different from linguistic. Later, 1974,
Hartigan, in his "Clustering Algorithms" applied
an algorithm called "reduced-mutation algorithm"
to amino-acid sequences from a few species (Chicken,
Kangaroo, and I forgot what else), remarking that
this method should work well on linguistic data
(supplied by Dyen). So there we are there, as
early as 1974, with linguistics using the very same
methods as biology. But I tried the reduced-mutation
algorithm on linguistic data and it gave the worst
results, by far, of all the methods I tried (in my
1980: "Experimental glottochronology: basic methods in
results"). It was also, by far, the slowest algorithm.
The main reason was that the chosen measurement of
distances had nothing to do with the properties of
language -- and I forget what the clustering algorithm
did. Thus there is no salvation in uncritically
espousing this or that method, this or that algorithm,
or measurement, just because it's done in biology,
and biology is a *true* science, so there. This
is magical behaviour, the comparative linguists'
equivalent of the Cargo Cult.
 To conclude, yet another sally on "mass comparison"
vs pair-wise comparison. When biologist reconstruct
the DNA from the evidence of N species, they always
rely on a matrix of the distances between those N
species, and it does not matter how those distances
are expessed, scalars, vectors, matrices, you cannot
build such a matrix without carrying out first
pairwise comparisons. "Mass comparisions" are nothing
but aggregates of pair-wise comparisons.
 But, as we say in French, autant souffler dans un
violon. Je pourrai m'arreter de souffler quand les
poules auront des dents...
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue