LINGUIST List 5.1468

Sun 18 Dec 1994

Disc: Comparative method

Editor for this issue: <>


  1. Ecological Linguistics,Anderson,PRT, Comparative Method

Message 1: Comparative Method

Date: 15 Dec 94 22:36 GMT
From: Ecological Linguistics,Anderson,PRT <>
Subject: Comparative Method

These are notes
after the recent discussions of Karl Teeter, AMRamer, Sally Thomason,
after the AAA meetings sessions on distant language reconstructions,
and also in response to part of Mr. Poser's message of 14 December.

First, most of the discussion of Karl Teeter I can agree with wholeheartedly
except in the use of any of his criteria as absolutes. Even grammatical things
can be borrowed (as Sally pointed out). People do in practice ***correctly***
accept a language family such as Uto-Aztecan as essentially proven, even when
no grammar has been written for the proto-language, or when only parts of a
grammar are written. On the one hand, nobody would be disturbed much if it
turned out that one of the assumed Uto-Aztecan languages was originally from
another family, with truly massive borrowing from Uto-Aztecan. The overall
hypothesis of a Uto-Aztecan family would not be shaken. On the other hand,
asking the question how much of the grammar of the proto-language has to be
written reveals the non-absoluteness of the criterion. Only a tiny bit? That
might be subject to one of the special cases in which some morphology and other
grammar were borrowed (Sally's general type of example). So, again, Karl is
right in general, but wrong if any single criterion is taken as an absolute
(unless circularly many criteria are combined by feats of legalese into a
single criterion).

I would like to modify AMRamer's statement in exactly the same spirit:

)we canNOT demand a detailed morphological reconstruction
)UNTIL the languages are accepted as related.

This is too strong.
It is not quite the same as Ramer's

)But surely he does not mean that a comparative grammar is a
)prerequisite to a reconstruction; it is PART of a reconstruction.

In practice, an extensive comparative grammar is indeed written as part of a
reconstruction (Ramer's second wording above, not his first above), but it is
not likely to be written unless ***BOTH*** of the following two conditions are

1) The languages are in fact related (makes the grammar easier to write)
2) Enough scholars believe that the languages might be related and so put
effort into establishing their links.

It is some unwarranted discouragement of the second which needs to be dealt
with in our field. Ramer is right that some linguists do discourage attempts
to prove what has not already been proven. No risk-taking, in other words.
Good researchers, attempting to write a grammar of Altaic for example, will
report all of their results, both for genetic relatedness, for borrowing, and
for a host of other questions they will not even have thought of when they
started their research. Some researchers evaluate only the hypothesis they
started with.

Very often the differences of viewpoint amount to nothing more than an
elevation of what one does oneself into the "true" or "real" work of the field,
instead of recognizing that it takes a number of different contributions. I
will illustrate from Karl Teeter's recent messages, with no malice intended,
because I am absolutely sure he intends none himself.

In modifying his wording to take account of AMRamer's point that of course
people do properly classify languages on the basis of phonological
correspondences, Teeter writes on 10 December:

"since it is clear that everybody's first approximation to linguistic history
begins with such classification. What I say is just that you cannot
RECONSTRUCT languages on this basis."

(Please note the words "first approximation" in this. That you cannot
reconstruct languages on the bases of classification alone seems to me a

Later in the same message, Teeter reverts to the more absolute statement:

"On the contrary, my contention (not my invention), is that the only way to
establish that languages are related is to write a grammar of the proto-
langauge and show how it developed into different later grammars."

(Please note the words "the only way" and "establish". Using these words does
not change the fact that Uto-Aztecan is ***correctly*** accepted as a proven
family without Teeter's criteria being satisfied.)

On 8 December, Teeter wrote:

)"Systematic correspondences of sounds in the vocabularies" may prove
)a connection between languages, which is certainly an interesting
)first step, but there the real work of comparative grammar starts:
 [Then Teeter mentions four possible explanations, only one genetic]
)Until one can exclude the first three factors, one has proven
)nothing at all regarding genetic relationship.

Teeter discounts the enormous work of discovering likely language families in
the first place, trivializes that as not the "real work" and establishment of
systematic sound correspondences as proving "nothing at all" regarding genetic
relationship. On the contrary, that does prove a grouping as a legitimate
candidate for genetic relationship, and often the nature of the sound
correspondences found will also have made one or more of the alternatives less

The work of comparative-historical grammarians is "real work". So is the work
of those who spend enormous hard-working hours sifting potential cognates to
discover potential sound correspondences. Do Teeter and others really have no
knowledge of how much work that takes?
At no stage is something ever completely proven in an absolute sense (not even
after a comparative grammar is written, because of the potential for
undiscovered problems of the kind noted by Sally Thomas). All stages of the
process contribute to the end result. All stages are equally "the real work".
Proof is always incremental, not nothing, not complete and absolute.

I turn next to Mr. Poser's message, and to other information gleaned at the AAA

Perhaps Mr. Poser will be surprised that I am enthusiastically in favor of
anyone correcting any errors in any claims of language relationship or language
structure, including Mr. Poser's mention of Kimball correcting errors in
Muskogean, even or especially if that means that a Greenberg claim about
pronouns in Amerind is weakened. It is actually claims about morphology where
I expect Greenberg is least likely to have succeeded in contributing something.
I thank Mr. Poser for the bibliography I can check against my lists of
corrections and for this note on Muskogean pronouns. I also agree with Mr.
Poser that the criticism of errors in data does not rely on the mere authority
of the critic. (That was not my point about appeals to authority, so one of
Mr. Poser's paragraphs was not directly relevant. My earlier point about
appeals to authority remains.)

I did not indicate (as Mr. Poser's message suggests) that critics complained of
errors when they could not back them up. What I did say was that they claimed
Greenberg's errors made the method worthless, without bothering to test whether
correction of the errors would actually lead to a change in his conclusions,
and without promptly providing the data so others could carry out such a test.
)From conversations, I would judge some are still reluctant to face this test,
the one required by a part of their claims. (Even if their claims on this
point prove wrong in some degree, because the conclusions of Greenberg's method
mostly remain the same even after errors of those kinds are corrected, it still
will not follow that Greenberg's methods produce valid results on a regular
basis. Please notice how careful it is important to be with notions of proof
for or against anything.)

In conversation with Bob Rankin at the recent AAA meetings I indicated I was
glad Greenberg had made the attempt at morphology, with his 3rd-person
alternation between /y/ and /t/ as a putative relic irregularity, but I was
equally likely to end up believing that he had discovered a new typological
fact, a preference for these unmarked segments and recurring conditions under
which they might alternate in more or less the same way, which might indicate
that the same phenomenon could arise by chance repeatedly. We are always
dealing simultaneously with possibilities of genetic relationship or convergent

As a general warning about the danger of throwing out hypotheses too early
because they are "obviously" cases of chance lookalikes or convergent
evolution, an article "Common pathways of illumination" by Stephen Jay Gould of
Harvard, in Natural History magazine, December 1994 pp.10-20, discusses the
origins of eyes in different phyla of the animal kingdom. He states that this
had been a classic case in biology, used to show convergent evolution of
originally unrelated organs to serve the same function. However, the DNA
specialists have gotten their hands on this one, and apparently it is ***the
SAME DNA*** which is regulating important parts of the production of eyes in
these various phyla, barring of course a few changes in a few amino acid codes.
This DNA is therefore genetically inherited (and we are talking as far back as
the common node on the family tree of Drosophila fruit flies, Squids, and
Humans!). If this line of argument holds up, we have quite a revolution in
thinking on our hands. (Such thinking can of course go far overboard too.)

"Chance lookalikes" certainly do exist massively. Sound symbolism is a
typological basis for their recurrence. The too-easy use of "chance
lookalikes" to reject comparisons can however also be just like the too-easy
use of "substrate" and many other ways of purporting to explain phenomena, but
in reality merely naming them without doing the hard work of really explaining

As another factual contribution to the discussion of Greenberg's errors, Bob
Rankin clarified for me at the AAA meetings that he had looked at Greenberg's
original notebooks, and that apparently what happened is that Greenberg had a
single flexible flap in his notebook bearing the language names, and sheets
which he matched up against that to enter data for particular lexical items.
This is a mechanism subject to errors of the two pieces of paper slipping
vertically relative to each other, and in fact there were rather a large number
of such errors. It was errors of rows rather than of columns (as I loosely had
assumed without ever bothering to ask if rows or columns).

Mr. Poser originally asked on 9th November whether

"limitations [of the comparative method] had been and were being used to
justify resistance to proposals of emote relationships..."
Please take careful note of this wording, as Mr. Poser changed the wording in
his message of 14 December, when he asserts no example has been given.

I gave Mr. Poser an eyewitness account satisfying the wording he used. I stand
by that eyewitness account. (I do not like naming names ever in these matters,
but Mr. Poser's assertion could be answered only by an eyewitness account.)

None of Mr. Poser's supposed rebuttal in his recent message is at all relevant
to that claim, though it is relevant to another claim, an absurd one, that I
did not make. (As in his previous message, Mr. Poser mixes several different
wordings of what are quite radically different hypotheses. Any extended
discussion of these variants would not be relevant to any of our central
points, so I omit them.)

To be specific about that absurd claim I did *not* make, I quite agree with Mr.
Poser that some of the same people also criticize Greenberg's hypotheses based
on the factual data. I have never denied that, and in fact took pains to refer
to other good work by the same people, and have always emphasized the
importance of having corrections of data.

Mr. Poser's conclusion was:

)It thus appears, as I thought, that there are no real examples of perceived
)limitations of the comparative method being used as the basis for rejecting
)proposals of genetic affiliation.

Since I gave an eyewitness account of such an example of the perceived
limitations being used as ONE basis for rejecting proposals of genetic
affiliation, Mr. Poser can only maintain his original assertion by converting
it into a different assertion as he has here, namely that there is no case of a
person using those perceived limitations as "the" (read "the only") basis for
rejecting such proposals. I will simply repeat that in the case to which I was
eyewitness, the tone of the presentation was quite clear that the absurdity of
the time depth was sufficient ***in and of itself*** to rule out the legitimacy
of attempting such distant comparison.
(A critique of errors in Greenberg's data is not directly relevant to this
point, even if engaged in by the same person.)

Apparently Mr. Poser wants his allies to be persons who not only do good
work, but also do nothing wrong. But we cannot posit such a division between
the "holy" and the "unclean", like the caste distinctions of traditional India.
No one has the right to take such a position vis a vis other good-faith
researchers, however much they may disagree with data, results, or methods.

I gather from some conversations at the AAA that some of the people involved
have become more moderate since the earlier years in these matters, at least in
their public statements. That is certainly all to the good. There are also
small beginnings of developments of method which may turn out to be relevant to
our current limitations.

I am perplexed by Mr. Poser's discussion of the supposed rule of using only
three-consonant matches never merely two-consonant matches. Although I used
Siouxan-Yuchi, because some Amerindianists are considering this, the point does
not depend in the slightest on whether one believes these two particular nodes
are related or not. Perhaps someone can propose another alternative (I would
suggest looking for one in Tibeto-Burman, where perhaps only one consonant, one
vowel, and tone would be available, and yet the genetic relations are in some
cases secure.) I agree with Mr. Poser that it is better to have three
consonants than two, and precisely because it helps to avoid chance
resemblances. As I stated in the previous message. That means that the only
difference between us is that Mr. Poser will consider treating the preferred
method (3 consonants) as the only permitted one:

)If (probably contrary to fact), matches of three consonants are
)necessary to exclude chance, ... I do not see why we should be
)unwilling, in that case, to conclude either that they are not
)related or that, if they are, the relationship is not demonstrable.

The problem here is the notion that such a particular rule *could even
conceivably* be "necessary to exclude chance". No rule is necessary to exclude
chance. A large number of procedures and methods can ***help*** to exclude
chance. Except in the most difficult cases of all, probably most single rules
can be violated, and there will still be enough other ways of excluding chance
that a good result can be achieved.

It is this fossilization in the response to expansions of the methods available
as part of "The Comparative Method" which is damaging to increasing the rigor
as well as the power of our field.

It is encouraging to see sessions on distant language relationships. I firmly
believe some of these sessions would not be occurring nor would there be so
much work on studying methods, if Greenberg had not published that book. Some
of the growth of interdisciplinary cooperation and communication was also in
the process of happening anyway, and Greenberg just accidentally published at
this time. (I am not trying to credit him with causing all of this interest.)

I particularly liked John Colarusso's contribution to the Eurasian session,
because he outlined his views on why distant language comparisons are very
difficult because of the progressive loss of data, yet he is not in the
business of criticizing people who attempt these. He is in the business of
himself contributing to a deepening of both linguistic and mythological
comparisons as much as he can in areas which are still to a great extent
uncharted. Johanna Nichols also provoked much thought with her methods, and
ways of integrating more information on peripheries vs. centers of innovation
and residue of repeated waves of innovation visible even at the peripheries, as
a pattern to look for in projecting a homeland backwards in time.

I am somewhat skeptical about very distant language comparisons which involve
only identities of sound correspondence, precisely because it is so easy for
this approach to select a very old but still relatively more recent layer of
borrowings (more recent than the node on the family tree we may be struggling
to find a way to reach). As Greenberg pointed out, an increase in the number
of conditioning contexts over time, once it approaches the number of lexical
items available for comparison, leads to the result that there are virtually no
recurrences of exactly the same sound correspondences. Please see a separate
message on how we can sharpen our tools to deal with this, titled
"Typology of Historical Change". This means that I start out somewhat
skeptical about Alan Bomhard's Nostratic and similar comparisons, precisely
because he uses only or preferentially sound correspondences of identity. But
that may be my personal bias, and I may end up granting that such a strictness
of sound correspondences can actually work over great time depths. I cannot
presume to know.

There was one session I did not attend but report here from the abstract, in
which researchers reported results of mitochondrial DNA studies of the various
Native American populations. As expected, Eskimo and Athabaskan were separable
(one variety each?). The remainder of Native American populations shared four
varieties of mitochondrial DNA, either without subgrouping, or perhaps in two
major subgroups. Bob Rankin did attend and says that the authors in their
presentation had come down on the side of two major subgroups. Population
genetics need not match language directly, but the data is still at least
interesting and obliquely relevant...

Lloyd Anderson
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue