LINGUIST List 7.1654

Sat Nov 23 1996

Disc: Tonkawa, Zuni & Vocabulary Comparison

Editor for this issue: Ljuba Veselinova <>


  1., Tonkawa Zuni & Vocab Comparison

Message 1: Tonkawa Zuni & Vocab Comparison

Date: Wed, 13 Nov 1996 23:24:29 EST
From: <>
Subject: Tonkawa Zuni & Vocab Comparison

Tonkawa Zuni & Vocabulary Comparison

(Those wishing to discuss this topic during the AAA meetings in
San Francisco, please leave messages at the AAA message desk or

Alexis Manaster Ramer's (AMR's) recent IJAL article vol.62 #3
July, pp.264-288)

"Tonkawa and Zuni: Two test cases for the Greenberg

marks a major step forward from the fruitless debates of the
recent decade.

AMR understands what the issues are and what they are not, and
succeeds almost completely (for the "almost", please see below)
in using terminology which will help others understand how to
make progress.

His conclusion is exactly right, that "the work on larger
linguistic groupings cannot and need not wait for the last *i* to
be dotted and the last *t* to be crossed in the work on these
reconstructions. Both kinds of endeavor can and should proceed
side by side."

To AMR's admirable work in clarifying terminology and avoiding
the pitfalls of useless argument and assertion (not
demonstration) of particular points of view, I wish to add three

Point number 1.

separates A and B when it finds B is more like C than B is like
A. Or it can fail to separate them, when such a demonstration is

It is amazing how hard it is for many people to understand this.
Greenberg's method of vocabulary comparison has been used with
good results by many before him. As pointed out by AMR,
Strahlenberg used it SUCCESSFULLY already in 1730 on the
languages of Siberian Russia.

This method takes a set of languages GIVEN IN ADVANCE and
proceeds to discover which of of the languages ALREADY IN THE SET
are probably relatively more closely related or probably
relatively less closely related. This method can do nothing if
the set has in it a language which has no other relatives within
the set.

This is, according to AMR, perhaps the case of Zuni in
Greenberg's classification.

AMR's conclusion that "the real contribution of Greenberg's
(1987) work may yet turn out to be in his proposed
subclassification of Amerind", that is, of the set of languages
of the Americas which are not already fairly clearly part of the
long-acknowledged Na-Dene or Eskimo-Aleut families. That is
exactly as it should be, and the only possible conclusion given
the nature of the method.

AMR's treatment of Tonkawa (suggesting that it may belong as part
of an extended Na-Dene rather than as part of Amerind) is
extremely attractive to me, and exactly in the right spirit. I
hope it leads to great advances.

Where I think AMR has not totally adopted a wording which
reflects what vocabulary comparison really is, is where he talks
about falsifiability of claims which Greenberg's method cannot
make, namely, a claim that a language IS A MEMBER OF the set
which was the defining set for the problem. That is, a claim
that ZUNI is a non-member of Amerind (granting that we do not
group it with Na-Dene or Eskimo-Aleut) is not a claim against
Greenberg, because his problem took Zuni as a part of the set,
and cannot do anything but say which other members of the GIVEN
set it is perhaps closer to. When we use ANY language outside
the starting set, we have changed the problem. This is entirely
separate from the question whether Zuni is in fact part of
Amerind, I am only here talking about a question of logic.

The choice of set is the choice of the problem, nothing more.
Hopefully we choose more rather than less reasonable sets to work
on, but we cannot know in advance. For example, what set of
languages should be chosen to work on the possible Asian
relatives of Na-Dene using vocabulary comparison methods?
 Not purely American languages, by definition.

A very similar thing happened to me in a paper I presented years
ago to the CAIL at the AAA, on Number Words in North America.
Included in a set of Na-Dene languages, unknown to me, was one
language which was not a member of that family. The method of
vocabulary comparison, however, worked as it should. In creating
a spatial array to reflect the degrees of relatedness of the
several attestations of number words, this particular attestation
had to be placed at an outermost margin, less closely related to
the others or any of their subgroups than the others were to each
other. The method could not conclude "unrelated" by its very
nature, all it could conclude was "more distant".

Point number 2.

The usefulness of the method of vocabulary comparison will depend
to a great degree on the sophistication of the judgements of
"similarity" in sound and meaning. It is my judgement that a
major lack in Greenberg's 1987 treatement is to not explain more
extensively what he meant by reasonable similarity....

In this area, AMR's treatment is at odds with what I believe to
be fairly commonly understood craft knowledge of likely sound
correspondence. AMR's idea of reasonable sound correspondences
are, at least to my understanding, much looser than those used by

This concerns both AMR's statement of general principles (p.269)
and his treatment of Zuni (pp.279-282). In the following, I use
quotations for italicized forms, to avoid the confusion of the
email use of asterisks for italics, where linguists use the
asterisk to mark proto-forms. I also use the symbol "^" to
represent the hachek of palatalized consonants such as "c^" or

On page 269, AMR comments that "Greenberg's proposed Penutian
cognates of Zuni "ate" include forms like Koasati "ic^ikc^i",
which seems no more distant than PIE *akwaa. Likewise, his
proposed Penutian cognates of Zuni "towo" include forms like
Rumsen Costanoan "tio", which seems no better nor worse than PIE

These assertions are both quite contrary to my sense of craft

The Koasati form "ic^ikc^i" contains the palatal consonant "c^"
Zuni "ate" contains a palatal vowel "e" next to a consonant "t"
which might become "c^" by palatalization. PIE *akwaa contains a
consonant "k" which might become "c^" by palatalization, but it
is not adjacent to a palatal vowel which might cause that
palatalization, and indeed just the reverse, is adjacent to a "w"
which might very well serve to prevent just such palatalization,
as in the history of Romance languages ("ci" now pronounced /si/,
but "qui" still /ki/ even though the protective /w/ has been lost
). So the two comparisons are not at all equal on phonological

Similarly, among "towo", "tio", and "ter", the first two are much
closer, as there is a match of one vowel, and the intervening
semivowel "w" can easily drop. A greater change would be the
loss of "r", which is quite likely to not be AS weak a consonant
as a semivowel between vowels. If we had to choose one of the
three as intermediate, it would be "tio" with palatal vowel "i"
possibly matching the "e" of "ter" it would not be "towo", which
is nearly at an opposite extreme from palatality. (Perhaps some
kind of an "r" could have yielded a darkening of the end of a
vowel, cognate with the "o" of "tio"???)

In AMR's treatment of Zuni, this problem surfaces very regularly.
AMR has 32 proposed cognates (Greenberg's or his own) between
Zuni and Amerind (Penutian) or Proto-Indo-European. These fall
into the following groupings by my standards for what is a closer
phonetic similarity (and sometimes a closer semantic one). I add
some comments in the appendix on particular items, along the
lines of the comments just above.

In cases where one or the other seemed conceivably close enough,
I conclude that

Amerind "cognate" closer in
##1,2,3,4,5,10,(12,17),18,19,20,21,22,28,29,30,32 Either seemed
conceivable at about the same level of similarity in ##6,7,8,9,11

Indo-European "cognate" closer in #14

In cases where neither seemed particularly close, I conclude that

Amerind is at least better in #25
Either or neither: #31
Neither: #13,23,24
Indo-European is at least better in ##26,27
(I excluded #15 for semantic reasons.)

The conclusion to me is overwhelming (and I did not try to make
it so) that Zuni has more conceivable distant cognacies with
Amerind (Penutian) than with Indo-European, quite the opposite of
what AMR concluded.

This shows how sensitive the discussions of "method" are to what
we consider to be our craft knowledge of closer similarities in
sounds (or for that matter meanings).

Point number 3.

Actually, as I pointed out long ago in the semantic realm, we are
ideally not concerned with "similarity" so much as with
"historical derivability". And we cannot escape the occurrence
of accidental combinations of circumstances yielding apparently
"crazy rules" which do not fit our craft knowledge. If these are
relatively more rare, vocabulary comparisons can yield better
results at deep time depths. (For the distinction between
"similarity" and "historical derivability" please see my article
"The Perfect as a Universal and as a Language-Specific
Category...", pp.227-264 in Tense-Aspect Between Semantics and
Pragmatics, ed. by Paul J. Hopper. John Benjamins Pub.)

So a real need in our field is an EXPLICIT formulation of what
constitute better candidates for closeness of sound and meaning
derivabilities, explicit so that they can be debated, corrected
when there is overwhelming statistical evidence, and limits of
validity established (dependent perhaps on sound-system

For the phonological field, there was long ago a much-neglected
article by Gene Schramm, "Distinctive Oppositions in Distantly
Related Languages", which considers the featural but not
segmental correspondences of oppositions in Afro-Asiatic pronoun
systems. That is to say, the features correspond more or less,
but in one language a given matched feature shows up in a
consonant, in another related language it shows up in an adjacent
vowel instead. It was at least to Gene Schreamm unpredictable
which place the feature would show up. Schramm's article should
be held up as a model of the kinds of correspondences of (true)
cognates which we actually do find, EVEN WHEN as in Schramm's
example, "recurring sound correspondences" DO NOT WORK.

In attempting to probe to greater time depths, or across kinds of
sound changes which more tend to conceal relationships, it is
exactly the kind of analysis Schramm did which we need. That is,
here is a definition of a topic for research:

Find all cases of vocabulary correspondences between languages
which are KNOWN to be related, yet which do not fit regular sound
 Rank order these if possible by degree of closeness of language
 relation. Then study the kinds of irregular sound pairings we
 obtain from the cognate
sets. Are these of the same general nature as the language
relationships become more distant? That is, can we inductively
derive patterns such as:

(a) As distance increases, does the number of phonological
features smoothly increase
 in which a pair of cognate segments disobey regular sound
(b) As distance increases, are the deviations from regular sound
correspondence of
 different types? That is, more vowel heights or roundings or
 More consonant manners or places?
(c) As distance increases, does the position become gradually less
predictable at
 which corresponding phonological features will manifest
 (That is, as in Schramm's examples, the feature may show up in a
 preceding or in a following segment, or even two segments
(d) Other or more specific patterns than these?

I hope this has been helpful.

Sincerely, Lloyd Anderson

Appendix: detailed comments on the Amerind (Penutian) - Zuni -
Indo-European "cognate" sets

1,2,5,6,18,20,21,28,29 have an extra consonant in the skeleton of the
 Indo-European cognate, making it relatively less qualified
 Of these, 6 has better meanings for IE?, 21 has exact matches

3 has closer meanings in Zuni and Amerind

4 was discussed above, is a case in which consonant-vowel
relations are considered,
 more like the cases discussed by Gene Schramm
29 may be of the same kind, with the darkness feature showing up in the
 of the Zuni form, and in the vowel /u/ rather than /a/ of the Wappo

Multiple more exact sound matches Zuni with Amerind:
10 has a noisier second consonant and a round first vowel
12 exact /mok.../ (although meaning might say PIE)
19 three consonants match and /u/ vs. /o/ (meaning also favors Amerind not
22 /tun/
30 the /k/ of PIE *leuk- makes it less similar than the matching /m/'s,
 unless there is a hint somewhere that the Zuni or Amerind
forms also
 had a velar final in the root, later lost.
32 the /ti/ of the Zuni matches the obstruent /c/ of the Amerind better than
it does
 the /l/ of the PIE; the root grade /ple-/ makes IE less good;
 and the Zuni /o/ is closer to the Amerind /a/ than to the PIE
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue