LINGUIST List 9.889

Tue Jun 16 1998

Disc: Limits on Knowledge in Linguistics

Editor for this issue: Martin Jacobsen <martylinguistlist.org>

Directory

1. H. M. Hubey, Re: 9.879, Disc: Limits on Knowledge in Linguistics
2. starhawaii, Re: 9.879, Disc: Limits on Knowledge in Linguistics

Message 1: Re: 9.879, Disc: Limits on Knowledge in Linguistics

>From: manasterumich.edu:
> 
>Some weeks ago, Larry Trask challenged my reference to
>Don Ringe as someone who appears to be claiming that linguistic
>classification has already reached limits beyond which it cannot go.
>The following quotation seems to me bear that out very well, however:
>"Investigation of real-language examples shows that resemblances
>between the basic vocabularies of languages commonly believed to be
>related occur with greater-than-chance frequency, while resemblances
>between the basic vocabularies of languages not commonly believed to
>be demonstrably related do not occur with greater-than-chance
>frequency" (Ringe 1992:80)

The main problem with Ringe's book is that there is a gap between what
heactually does and what he apparently thinks he has done. First he
leaves halfway what he started to do. Then he assumes that he has done
something which he has not. Finally, he attacks his own results on the
mistaken belief that he has done something wrong or that there is
something seriously wrong with mathematics (in this case probability
theory).

The last part is a much more serious error (at least to me) because it
is at the root of all the difficulties with his comprehension of
randomness and probability theory. All statements in probability
theory are of the probabilistic type. That is because everything in
the world is that way except mathematics. All statements of math
follow from axioms via the use of logic. Logic itself is an axiomatic
system so all statements of math are a part of some axiomatic (formal)
system. Formal language theory is an example of such a formal system.

Even in physics (and the related areas in engineering) the equations
are of a probabilistic (and inductive nature). It just so happens that
there is so little variation in results obtained in laboratory
experiments that we can give deterministic laws (which are the
equivalent of axioms of math systems) as good (actually often
excellent) approximations of reality. In real world applications there
is often lots of noise/randomness and it must be taken care of using
probability theory.

After Ringe goes thru the whole process of explaining why his method
is better, he is forced to backtrack when he gets better-than-chance
correspondences

between Turkish and English. But that is circular. If we compute that
chance correspondences between English and Taiwanese should be 10
words and instead get 200 either we have made errors in the
computation or there is something to these correspondences that is not
due to chance. If the claim is that the method is good, then we have
to accept the results. If we do not want to accept the results then we
have to find the weak points in the method, but we can't have both.

> To be sure, this (to my mind, astounding and completely unjustified)
> claim seems to be contradicted by other things Ringe says, but I
> cannot see how this passage can be read to mean anything other than
> what I said, namely, that any resemblances which may be found among
> languages not ALREADY classified will be due to chance and hence
> will not be usable as a basis for FURTHER classification. If Ringe
> had said, as he should have, that his (as it happens, mathematically
> incompetent, see the review by Baxter and myself in Diachronica)
> investigation of a trivial number of unrepresentative examples shows
> (once the mathematical blunders are corrected) that the situation he
> claims is still at best an enormous overgeneralization of what is
> true in even those few cases, then that would be another matter.
> But in the absence of a quantifier like "some" or "a few", I can
> only understand his claim to be a universal one.

Before I followed thru on what Ringe did, I expected that he would
tosomething else from what he wrote so there is obviously a problem of
what he thinks he is doing/done and what he has actually
done. Secondly after I looked at his computations I saw that they are
OK. But there is no reason why he should not have continued and done
the same calculations for the second phonemes and the third phonemes
and used those also. Thirdly, there are two correspondances between
Turkish-English which are much beyond chance. That means the prob that
they are due to chance has to be multiplied (i.e. logical AND) so the
prob gets even smaller. He does not show this explicitly
anywhere. Fourth, I have an alternative explanation which, in fact, is
for proto-world, not against. It can be seen in several examples, in
Tuna's book that Sumerian m becomes Turkic k. But m and be are often
mixed up. Furthermore we have yet Hunnic mola > kala (Common
Turkic). So we can already see very easily why English b seems to
correspond to Turkic k. This is what I referrred to above. If Ringe
first says his method is great and then does not believe its results,
it is rather silly. Why contradict yourself in one work?

Finally, one of the tenets of probability theory and statistics is
that more the better. The smaller the sample greater the variance and
hence greater the uncertainty in the result. So there is no need to
stick to the Swadesh-100 if statistics is employed. Since the whole
basis of historical linguistics is chance vs not-chance and since we
do have the power of mathematics behind computing chance occurrences
the bedrock of historical linguistics and its whole expression should
be finding results which are not due to chance. The problem of
geneticity vs borrowing is a much more complex phenomena. It cannot be
resolved except via additional assumptions. It is those assumptions
that have to be clarified and questioned. But we do have the means at
our disposal to compute probabilities of chance occurrences. But it
is important not to get stuck in circularities.

Those wishing to employ probability theory should also pay careful
attention to the assumptions behind the formulas and make clear
statements. As such Ringe's complaints about weaknesses of the
multilateral comparison vs binary comparison are unclear. As long as
probability theory is used correctly, neither method will give a
result better or worse than the other because the basis of both is
probability theory and it will give the same result for both. The
problems are in the use (or misuse) of the theory.

> Reference:
>
> Ringe, Donald A., Jr. 1992. On calculating the factor of chance in
> language comparison. Philadelphia: The American Philosophical Society.
> [Transactions of the American Philosophical Society, 82(1).]

- 
M. Hubey
Email: hubeyhMontclair.edu Backup:hubeyhalpha.montclair.edu
WWW Page: http://www.csam.montclair.edu/Faculty/Hubey.html

Message 2: Re: 9.879, Disc: Limits on Knowledge in Linguistics

> From: manasterumich.edu
> 
> Some weeks ago, Larry Trask challenged my reference to Don Ringe as
> someone who appears to be claiming that linguistic classification has
> already reached limits beyond which it cannot go. The following
> quotation seems to me bear that out very well, however:
> 
> "Investigation of real-language examples shows that resemblances
> between the basic vocabularies of languages commonly believed to be
> related occur with greater-than-chance frequency, while resemblances
> between the basic vocabularies of languages not commonly believed to
> be demonstrably related do not occur with greater-than-chance
> frequency" (Ringe 1992:80)
> 
> especially since he goes on to argue that the results are not
> "significantly different" if one looks beyond basic vocabulary or
> otherwise changes the method of comparison which he himself had
> employed.
> 
> To be sure, this (to my mind, astounding and completely unjustified)
> claim seems to be contradicted by other things Ringe says, but I
> cannot see how this passage can be read to mean anything other than
> what I said, namely, that any resemblances which may be found among
> languages not ALREADY classified will be due to chance and hence will
> not be usable as a basis for FURTHER classification. ...

I do not see a problem with the universe of discourse expanding to
admit later, not already classified languages.


> If Ringe had
> said, as he should have, that his (as it happens, mathematically
> incompetent, see the review by Baxter and myself in Diachronica)
> investigation of a trivial number of unrepresentative examples shows
> (once the mathematical blunders are corrected) that the situation he
> claims is still at best an enormous overgeneralization of what is true
> in even those few cases, then that would be another matter. But in
> the absence of a quantifier like "some" or "a few", I can only
> understand his claim to be a universal one.

I have not seen the review you speak of so I can't comment
specifically on that. The objection you have would seem to center
around all possible universes of discourse: past, present, future and
so on. Ringe would seem (from your above statement...I am new to the
list) to be speaking about a well defined present universe of
discourse/languages. This would be quite apart from all possible
universes (of languages).
 
Kevin Johnson

Philosophy Major
University of South Florida