LINGUIST List 5.1500

Wed 21 Dec 1994

Disc: Comparative method, Polarization & reviews

Editor for this issue: <>

Directory

1. Ecological Linguistics,Anderson,PRT, getting past a polarized field
2. Ecological Linguistics,Anderson,PRT, Compar: IndoEur as the model?
3. Ecological Linguistics,Anderson,PRT, Evidence against Greenberg?
4. Ecological Linguistics,Anderson,PRT, Compar outsider / specialist

Good reviews are difficult to get in a polarized field.

I would love to be working in a field which was a mixture of enthusiasm of
discovery and carefulness in handling data.

Without all the blasted politics.

Not to take any position at all on the recent controversy concerning a review
of Illyc-Svityc's Nostratic work in *Language*

but to point out the difficulty of getting reviews in a polarized field,

the first review of Greenberg's Language in the Americas

by Campbell, omitted to treat the chapter on detailed analysis of method in
statistics of vocabulary comparisons at different time depths, thereby to my
mind illegitimately giving the impression that Greenberg was naive and had not
considered this. (Completely separate from whether Greenberg's considerations
on techniques are valid, I of course suspect some are, some are not.)

Sally Thomason was able to get a second review, from Matisoff.
This one used the term "megalolinguistics", from which the invited inference is
obvious to anyone native to the English language: "megalomania". This is ad
hominem, and inappropriate to a scholarly journal.

What can one do in such a climate?

We have still not had a review by someone who is actually interested in all
possible outcomes of the investigation, without any axe to grind, correcting
errors, estimating how much influence that would have on the conclusions, etc.
etc. It may be that it will be a long time before we can really get such an
evaluation, simply because as a field we do not yet have enough information on
how various measures of potential relatedness we may use are actually affected
by increases in time depth etc.

For those who wish to contribute to our increase in knowledge in this sphere,
hence our ability to even attempt an objective evaluation of such hypotheses
and techniques as Greenberg's, I posted earlier the main outlines of at least
one plan of investigation, namely, making explicit rather than implicit the
"typology of historical changes" in semantics, phonetics, etc.

I would love to gradually develop a growing core consensus,
of people
and of techniques
(I use "techniques" rather than "methods" because "techniques" is positive and
lacks the overtones of religious purity often attached to "proper method")

of people who, without knowing in advance the conclusions to be drawn,
at least agree that all or most of a list of possible investigations would
contribute to our techniques and ability to evaluate claims
of deep or distant relationships.

The "typologies of historical changes" which are my own personal interest and
where I have had something to contribute can of necessity be only one part of
those.

Lloyd Anderson

Indo-European as the prime model for comparative-historical linguistics?

Certainly insofar as techniques have been developed there, and there are older
written documents helping to verify the results of those techniques.

However, as pointed out by others, the particular structure of Indo-European
languages must not be permitted to bias our historical techniques against
languages with different kinds of structure.

I'd like to see more respect and more modesty in all directions on this issue.

Lloyd Anderson

The initial replies to Geoffrey Pullum quite properly said that the question
might be misplaced and not really be about what it seemed to be about:

What counts as the true null hypothesis?
Multiple variables so nothing can be definitively disproven (or I would add,
proven);
Turning the discussion to how people in the past have actually changed their
views, as separating Tai from Sino-Tibetan.

Let me instead give some indications of what does and does not for me count as
evidence against some particular claims. I am being very precise in my wording
here, using "count as evidence against" rather than "disprove" because it is a
fallacy to believe that evidence "disproves" or "proves" hypotheses directly.
That would be contrary to everything we know about the history of real science,
as opposed to abstracted philosophy of science.

First an example which does not at first hearing seem to me to be highly
relevant or a strong argument, then one which might be more relevant.

Bob Rankin mentioned to me at the AAA an as-yet-unpublished paper by Lyle
Campbell from the Colorado conference, in which Rankin says Lyle showed that
taking Uralic and Amerind, that Greenberg's method would subgroup part of
Uralic with part of Amerind. (I have no independent knowledge of Lyle's paper
besides this comment from Rankin.) That seems counterintuitive, though not all
connections across the Pacific of that kind must necessarily be
counterintuitive. Indeed, if we believe in multiple migrations and had
infinitely powerful tools to discover true language families, we should expect
to perhaps find multiple connections of different parts of Amerind.

There are three considerations which make this not particularly relevant to me.

1) As in statistics, if the assumptions of the technique are violated, the
results of using a technique may be less than useful. Greenberg's technique
assumes (and does not try to prove) that all of the languages under examination
are in fact related at some level. The only conclusions from his method are
subgroupings, i.e. splits, not groupings, i.e. assertions of relatedness. That
is, ***if*** all of these are related, then group A are more closely related
within themselves than they are with B.

2) I assume (perhaps wrongly, by the way), that the distance between Uralic
and Amerind (if indeed it is a single group) is of such an order of magnitude
greater than distances within Amerind, that assuming they are related is
tantamount to a violation of the assumption behind the technique. In other
words, Greenberg's technique like any other must have time limits reducing its
effectiveness past certain depths. I am assuming only that there are
techniques (Greenberg's, improved upon and with fewer errors, might be one or
might not be) which can go deeper than the traditional techniques of the
comparative method as practiced in the past. It is our goal to increase the
power of our techniques, after all.

3) I know that Campbell is so intent on discrediting Greenberg that his review
in Language improperly failed to really recognize even the existence of a major
chapter of Greenberg's book, that part dealing with analysis of rates of change
and other techniques in long distance comparison rather than with the
particulars of Amerind. So I cannot trust his conclusions, but would have to
go over the data of the Uralic and Amerind languages myself to see whether
Campbell had, knowingly or not, selected data to discredit Greenberg.

Now I turn to an example which may count as evidence against Greenberg's
hypothesis.

The findings of the DNA analysis reported at this past AAA, again according to
Rankin who attended, came down firmly on the side of there being two subgroups
within Amerind (aside from the Eskimo and the Athabaskans). If Greenberg's
method fails to find these two major subgroups, that is for me more relevant to
arguing against Greenberg than the report I have of Campbell's paper, because
it is presumably on a lesser time depth, and either Greenberg's method or his
assumption that Amerind are all related at a time depth his techniques can
handle must be wrong. Of course that is also assuming that to the degree
relevant for this discussion, the genetic groupings do match the true language
family groupings, which as we know need not be the case because of language
shift. I say this merely for completeness, because I can simultaneously say
that such a finding (if it does not match Greenberg's subdivisions within
Amerind) does for me count as evidence against, and also say that like any
evidence it cannot ***by itself*** disprove Greenberg's hypotheses.

Since I have as yet no reason to suspect the authors of the DNA paper as having
an axe to grind about Amerind linguistics, that is also one reason I will tend
to treat this argument as more telling. I do not yet even know what the two-
part division of Amerind was that the DNA analysis suggested!

Does Geoffrey Pullum really propose that the null hypothesis should be that
Muskogean is not related to anything else in Amerind? Perhaps I am misreading
his words. I find such a suggestion extremely implausible, and my null
hypothesis would be the reverse, assuming it was related to something in the
Americas but without specifying what. However, when faced by the claim that
Muskogean is most closely related to some particular Amerindian family, I would
take the null hypothesis to be "not related".

Lloyd Anderson

Readers would do well to compare Larry Trask's message on errors made by
outsiders (and why only specialists should be allowed to engage in comparison)

[but of course they often refuse to engage in it]

with Benji Wald's discussion a couple of days ago of the attitudes to the
Greenberg reclassification of African languages. It was the well entrenched
prestigious elites that resisted the longest, according to Wald's summary if I
understood him.

That is typical in all history of science. I do not see why it is any differnt
here. It is also the case that elites resist things that are wrong quite
strongly, though not always. Elites can betray their own highest values too.

So the fact that "outsiders" make errors is simply a given fact of science, and
it will always be so viewed by the specialist insiders. The strong language
Trask used merely manifests attitude, not facts.

[Alice Faber has properly responded in part to this, I find since writing
this.]

If correction of Greenberg's errors leads to loss of his hypotheses, so be it,
but let's discuss this based on facts, not on attitudes against outsiders by
specialists.

Comparison and its techniques is a specialty too, just as much as a particular
language or language family can be a specialty.

By the way, Benji Wald's mention of the following I find very interesting:

"the selective circularity method which G's method is best equipped to attack"

I have not seen any specialist historical/comparative linguists pointing out
the value of G's method precisely in countering hypotheses of relationship of
particular pairs of languages or language families (even with support of a
comparative grammar!), when those comparisons were based on the limited
knowledge of the specialist in one or two of those being compared who did not
have an equivalent level of knowledge of other languages within sufficient
geographic and genetic distance to reveal the non-genetic relation of the
particular two being compared.

The notion of "specialist" is not usually an empirical notion, it is usually a
political one. On the other hand, notions like "careless" and "dilettante" can
be applied to all, or not, equally. A specialist in one language who chances
to see comparisons with one other, however much their knowledge of the one, is
still a "dilettante" if they are not well grounded in historical / comparative
techniques and the errors to which these like all techniques are subject, as
well as other candidates for the comparison, and "careless" if they do not
investigate other possible candidates, by exactly the same uses of the terms.

And despite all of that, a "careless dilettante" may happen to have observed
true cognate sets and have hit upon a correct relation of two languages.

Once again, let's get rid of all this politics and deal with the data!

Lloyd Anderson