larryt at cogs.susx.ac.uk
Thu Jan 28 20:22:13 UTC 1999
On Wed, 27 Jan 1999, H. Mark Hubey wrote:
[snip of comparisons of English with English, German, Russian]
> In all of these it is precisely the correlation of phonetic distances
> and semantic distances that are use to posit family relationships.
No; not at all.
To begin with, you can *posit* a family relationship on the basis of
anything you like, and people do this all the time. Positing is cheap,
easy and generally not very interesting. The hard bit is to
*demonstrate* a relationship beyond reasonable doubt. And this can't be
done merely by appealing to phonological resemblances: it has to be done
in a more principled manner.
> The conclusion is that precisely the opposite of what Larry Trask says
> is true, namely that historical linguistics is about nothing else except
> the computation of correlation of phonetic and semantic distances.
This is not what I said. I never spoke of the computation of anything.
The classical methods of demonstrating a linguistic relationship do not
involve anything that could reasonably be called `computation'.
> Of course, the word "resemblance" like "similarity" is nothing more
> than the inverse of distance.
> What Larry means by "rigorous statistics" is not clear to me.
I mean the expression to apply to something which does not at present
exist. Though vigorous efforts are being made in some quarters, at
present we do not have a reliable and generally accepted statistical
technique which can identify otherwise obscure genetic relationships.
I hope we will have such techniques one day, but we don't have them now.
> After all, the present families posited are nothing more than
> heuristic(heuristic) computations of the distances (phonetic and
> semantic) and a heuristic correlation of the said distances.
I think you are confusing the identification of families with the
problem of subgrouping the languages they contain. What is described
here is of some relevance to subgrouping, but not to the identification
of families. In fact, it is perfectly possible to prove a language
family without being able to provide an acceptable family tree
(subgrouping). Semitic is a classic example of this, but there are many
others, such as Austronesian, Austro-Asiatic, and, for that matter,
> And probability theory (and statistics) is nothing more than the
> same common sense except with explicit computation.
Hardly. Probability theory is a triumph over common sense, not an
elaboration of it. As has been demonstrated countless times, human
estimates of probability are woefully inadequate.
Most of you will already know that old chestnut, the shared-birthday
problem. How many arbitrary people do you have to assemble in one room
before the probability that two of them celebrate their birthday on the
same date becomes greater than 50%? Ask this of *anybody* who is
untrained in probability, and you will get a large answer: typically 183
or something even bigger. But the correct answer is 23. Most people,
relying on their common sense, will refuse to believe this until it's
explained to them, and some will refuse to believe it even then.
And I won't even get started again on Marilyn and the Goats. ;-)
> However, where heuristics fails, explicit computation (i.e. prob
> theory and statistics) does not fail. It's time to move to the next
> stage of historical linguistics.
After you, Mr. Hubey. ;-)
Actually, probability has not failed in comparative linguistics. But,
so far, at least, it has not proved possible to develop probabilistic
approaches to identifying families which work. If you'd like to propose
an explicit procedure which *does* work, there are plenty of linguistics
journals which will be happy to publish it.
University of Sussex
Brighton BN1 9QH
larryt at cogs.susx.ac.uk
More information about the Histling