LINGUIST List 3.872

Sat 07 Nov 1992

Disc: Probabilistic Reasoning

Editor for this issue: <>

Directory

1. Don Ringe, probabilistic analysis of language comparison
2. Don Ringe, probabilities: six languages compared
3. , Probabilistic Reasoning in Linguistics

Dear Colleagues on the Linguist List-- I hope that those of you who are
interested in language comparison and probabilities will take a look at a
monograph by Don Ringe (that's me), called *On Calculating the Factor of Chance
in Language Comparison*, published this year as Vol. 82, part 1, of the
Transactions of the American Philosophical Society. The math in it is
extremely primitive, and that's part of the point: it doesn't take much
mathematical sophistication to cope with chance resemblances--but it *does*
take more than was recently exhibited in Scientific American. (I should also
warn you, if you are not already familiar with the problem, that fn. 57a of the
monograph is incorrect; I apologize for not catching that before it went to
press.) --Don Ringe

John Coleman points out that in their *SA* article Greenberg & Ruhlen compute
the probability that a certain type of peculiarity will appear in words
translating each other in six languages by chance. This makes their findings
look impressive at first. But their argument is valid only if *exactly* six
languages are being compared; as the number of languages being compared
increases, the probability that a given peculiarity in a word of given meaning
will show up in *some* six of the languages also rises dramatically. Here is
an example.
Suppose the probability of some phonological characteristic appearing in a
particular word in each language is .1 (which is realistically small for some
types of cases). If only six languages are being compared, the probability
that that characteristic will appear in "the same" word in all six is .1 to the
6th power, or .000001. But if twenty languages are being compared, the
probability that that characteristic will show up in a given word in some six
of the twenty is .1 to the 6th power (for the languages in which it appears),
multiplied by .9 to the 14th power (for the languages in which it does not
appear), multiplied by 38,760 (because there are 38,760 *different*
configurations in which six such words can be distributed over twenty different
languages--six "x's" in the twenty columns, so to speak), which comes to about
.008867 (if I've done the arithmetic correctly). Note that the latter
probability is more than *8,000* times as great as the former--and we're
talking about *only* twenty languages!! Greenberg and Ruhlen compared many
more.
Greenberg and Ruhlen certainly have *not* demonstrated that the similarities
they're finding could not be the result of chance. --Don Ringe

Bruce Nevin has asked that I post the full text of a short notice I wrote for
the current SSILA NEWSLETTER (11(2), October 1992, p.9) on Don Ringe's recent
monograph on the role of chance resemblances in comparative linguistics. I
join with Bruce in urging that everyone who is interested in the statistical
reasoning (or lack of it) behind claims that "(such-and-such a degree of
lexical resemblance) can be no accident" take a look at Don Ringe's very lucid
presentation of the statistical realities that must be dealt with.

 *********************

*On Calculating the Factor of Chance in Linguistic Comparison*. Donald A.
Ringe, Jr. Transactions of the American Philosophical Society 82, Part 1,
1992. 110 pp. $16.

This is a readable and accessible demonstration that chance resemblances
can be more frequent than some people think and that multilateral comparison
further increases their frequency. Using only "the elementary mathematics of
probabilities," R. shows that "a non-negligible number of fortuitous
similarities" can be found between every pair of languages.

R.'s argument may be briefly summarized:

Since the distribution of sounds in vocabulary lists is effectively random,
resemblances in sound between synonymous words in of different languages
arise by chance according to the general laws of probability. Investigation
of real-language examples shows that resemblances between the basic vocabu-
laries of languages commonly believed to be demonstrably related occur with
clearly greater-than-chance frequency (a fact unaffected by the use of longer
wordlists and/or word-comparisons which are not semantically exact), while
resemblances between languages not commonly believed to be related do not
occur with greater-than-chance frequency. Comparison of the vocabularies
of several languages at once yields a pervasive pattern of systematic
similarities which are the result of random chance, indicating that the
results of the multilateral comparison must be treated with extreme caution.
Since the burden of proof is always on those who claim to have demonstrated
a previously undemonstrated linguistic relationship, it is very surprising
that those who have recently tried to demonstrate connections between far-
flung language families have not even addressed the question of chance
resemblances. This omission calls their entire enterprise into question.

This will be unwelcome news to "Long Rangers," and will surely be challenged
by them. But R. makes a convincing case that "it is urgently necessary to
subject all controversial 'demonstrations' of language relationship to
investigation by the probabilistic method."

[Order from: American Philosophical Society, P.O. Box 40098, Philadelphia,
PA 19106 (tel: 215/444-3400). Postage free on prepaid orders.]

******************************************************************************
* THE SOCIETY FOR THE STUDY OF THE INDIGENOUS LANGUAGES OF THE AMERICAS *
* *
* Victor Golla Dept. of Ethnic Studies *
* Secretary-Treasurer Humboldt State University *
* Arcata, CA 95521 USA *
* *
******************************************************************************