LINGUIST List 3.865

Fri 06 Nov 1992

Disc: Probabilistic Reasoning

Editor for this issue: <>

Directory

1. Henry Kucera, Re: 3.859 Probabilistic Reasoning in Linguistics
2. "Bruce E. Nevin", Probabilistic Reasoning
3. Ann Taylor, Re: 3.859 Probabilistic Reasoning in Linguistics
4. Scott C DeLancey, 3.859 Probabilistic Reasoning in Linguistics (fwd)

Message 1: Re: 3.859 Probabilistic Reasoning in Linguistics

 Am I missing something? There is a vast amount of literature, mostly from the
1950's (e.g.Swadesh) about lexicostatistics also known as glottochronology. The
 issues of "basic vocabulary" and of the significance of statistical comparison
of various languages was much discussed. As a curiosity: There is an MA thesis
 (by someone named Murphy, I think) that shows (on Sanscrit and Modern English)
 that IE separation took place about 3500 BC, not a bad conclusion.

The fact that there can be isolated accidental resemblances between two
languages is already mentioned in Bloomfield's Language (he cites the word for
'eye' in Modern Greek and Malay, if I remember correctly).

Is something else being discussed here or are we no longer reading things
written in the 1950's?

 Henry Kucera

Reference:

 Ringe, Donald A., Jr. 1992. On calculating the factor of chance in
 linguistic comparison. Transactions of the American
 Philosophical Society 82, Part 1. 110 pp. $16.

Briefly reviewed by Victor Golla in the October issue of the SSILA
Newsletter (The Society for the Study of the Indigenous Languages of the
Americas _Newsletter_ XI:3 pp. 9-10.) Victor, I know you're among us,
maybe you could be persuaded to post the text of your review?

A sentence from the review:

 This will be unwelcome news to "Long Rangers," and will surely be
 challenged by them.

Is the SciAm article perhaps just such a challenge, or is it only
(perhaps belated, as Ringe suggests) attention to the issue of chance
resemblances? I have not read it.

Bob Oswalt has also done a great deal of work on this topic. A long
piece he wrote 20 years ago, he has told me, was turned down at that
time by _Language_. I have suggested resuscitating it. Unfortunately,
Bob does not have email access.

	Bruce Nevin
	bnbbn.com

Message 3: Re: 3.859 Probabilistic Reasoning in Linguistics

For people interested in the probability of chance matchings when comparing
languages, there is a recent monograph ('91 or '92) published by the American
Philosophical Society called On Calculating the Factor of Chance in Language
Comparison (or something very similar) by Donald A. Ringe, Jr. which addresses
these issues in a detailed and systematic way.

John Coleman's reponse to Manaster-Ramer's criticism of Greenberg and
Ruhlen's statistical argument is doesn't really save it. He says:

>Alexis Manaster-Ramer misrepresents Greenberg and Ruhlen ...
>Greenberg and Ruhlen estimate the probability of similar words with
>with a similar meaning occurring in SIX languages (not two), chosen
>from a longer list of similar words with a slightly wider range of
>meanings.

But G&R do in fact say, in reference to the match of 3 consonants in the
words for 'throat' in Halkomelem and 'swallow' in Tfalati:

	the chances of an accidental match are (1/13)(2/13)/(4/13)=
	0.0036412391

Which is just plain baloney no matter how you slice it. Maybe they really
know better, but that is what they said, and it's not true -- and apart
from the linguistic legitimacy of the claim, there's something alarming
about the sight of *Scientific* American printing a sentence which makes
such an elementary error in statistical reasoning. And there's an equally
egregious error immediately after; in calculating the odds of the six-way
match that they present, G&R simply multiply out this spurious 0.004
probability for each of six families. But this is correct only if the
six languages compared are the only ones in the relevant data set. In
fact each representative language is chosen from dozens or scores of
possible candidates within its family, and no evidence is presented to
show that the form is prevalent or reconstructible within any of the
families. Thus the odds of finding a match are considerably improved
beyond what G&R claim. Possibily the statistical argument, correctly
done, would still hold--but the statistical claim made in the SA article
is nevertheless simply wrong.
	That said, of course, Coleman is correct in pointing out that

>And G&R claim to have done likewise for over a hundred
>other "basic vocabulary" words. Which makes their argument
>much stronger than Alexis characterizes it as being.

though as many people have pointed out, the ultimate strength of the
argument is dependent on the validity of the "over a hundred" other sets.

Scott DeLancey