Polar Models vs. real-world Intermediates

[ECOLING at aol.com]

I would much prefer that we were working on displaying
data of the difficult intermediate cases,
rather than fending off the theoretical pronouncements
of Trask that in essence deny the existence of real-world
intermediates.

The debate, as so often with Trask,
seems to be based mostly on definitions,
his polar oppositions instead of the reality of gradient intermediates.

For Altaic,
for Germanic within IE,
and for English descended from a partial fusion of Older English
     with a substantial morphological component from Norman French,
for Indo-Aryan (vocabulary and some morphology from PIE,
     much syntax and other morphology from Dravidian area
     especially at the contact fusion zones such as Konkani and Marathi,
     if I remember rightly),

the simpler polar oppositions are inadequate.
These are, despite Trask's denials, real-world cases,
whose intermediate status is most probably *known* at this point,
(though like all presentations of facts, not absolutely certain the best,
and subject to future revision).

In a message dated 3/1/2000 12:01:34 AM, Trask writes:

>All parties to the Altaic debate agree that there are just two
>possibilities: either Turkic, Mongolic and Tungusic are genuinely related
>within a single larger Altaic family, or they are not, and the obvious
>connections among them are merely the result of prolonged contact among
>unrelated families.  The parties differ as to which view they believe is
>the true one, but nobody I know of is arguing for any third view.

The statement above would be true if our conclusions were limited to
the two extremes, limited to the pure models.
But they need not be so limited,
not if our conclusions are to have mostly empirical content,
rather than mostly definitional non-empirical content.

Once we recognize that our models are not the same thing as reality,
and that there are intermediates and fuzzy edges in the real world,
it becomes possible to recognize in a particular case
that the truest answer we can give *may be*
that the real world was *intermediate* between two polar models.

The reality for the Altaic grouping
*might* have been that there were once independent languages,
or parts of their areas,
which were in such intimate and intense contact that they fused
into a single language continuum, describable imperfectly either
as a family tree (with strong areal differences) or
as a dialect network (with strong polar differentiations and overlapping).
Either of those phrasings clearly describes an intermediate.

Trask must I think reject at least a part of what is in the
immediately preceding paragraph, having to do with languages
fusing, because he writes:

>Middle English did not "inherit" these words at all:
>it *borrowed* them from Norman French.
>This is just the way the terms are used in linguistics.

Trask seems unwilling to recognize that the degrees of language
fusion can differ, and that there is no sharp boundary for the
appropriate use of the word "inherit".
Is there some unbiased view in which we can clearly say
that the descendants of the Norman conquerers did *not*
inherit language material from their ancestors, they only borrowed it?
At each stage of the process, they probably believed they
were inheriting, borrowing from English bit by bit.
Despite the recognized role of Norman French morphology in this
history? Despite the fact that modern English (and Middle English)
*use* parts of French word-formation morphology?
That morphology sits very deeply in the structure of modern English,
so deeply that "inherited" seems indeed an appropriate word
to describe the social situation, the psychological situation of
individual speakers, and the linguistic situation.

None of the above implies that the status of Old English and
Norman French are equal in the fusion.  But there was some kind
of fusion, more than simply borrowing of a few items of vocabulary.

That is why it is indeed dealing in "extreme" cases to compare
the borrowings of English from Indic, Chinese, and other sources.
It is a failure to recognize the gradient intermediate nature of the real
linguistic worlds that would class these as the same as the Norman
French component in the history of English.
The clarity of the simple models in *one* case
(polarly opposed "genetic" vs. "borrowing" in the case of
Chinese loanwords in English)
*can in no way* demonstrate any similar clarity in another case,
such as the interaction between Old English and Norman French.
The situations are so different in degree that the simple polar models
do not work for the case of English, which both
(a) emerged from the unequal fusion of the two, and
(b) emerged from the borrowing by one from the other.
BOTH statements (a,b) are true in a gradient real world,
and the denials of EITHER of them are false.
Saying so does not imply the absence of polar cases,
merely that polar cases *are* polar cases,
not fully appropriate to describing interemediates.

*

Noting the possibility of intermediates and demonstrating their
applicability to some particular cases of course decides
no empirical question  in any other particular cases.
Each case must be examined in its own right.

Whether a more intermediate model better applies to Altaic
is of course still an open question.
It may simply be that there has been insufficient analysis,
and that it may someday be discovered that one of the simpler,
more polar models actually does work quite well
(with certain extraneous factors then identified and segmented off).

Or it may be that the debate is an artificial one, sharpened by the
oversimplified models being used as the polar opposites in the debate.
Altaic, so far as we can reconstruct it from available data,
may be a dialect network which can be best analyzed using
neither pure tree nor pure dialect network models.
We may be unable to go deeper than that,
in which case we should state just that as our most responsible conclusion,
and claiming either polar model would be a less responsible conclusion.
BOTH polar parties to the current debates about Altaic
(the only two alternatives Trask considers) may be wrong.

Obviously, neither polar model is yet satisfying for Altaic,
which is why the debate continues!

Saying that the ultimate answer may be an intermediate
in no way says that people should stop investigating lines of
reasoning and sets of evidence which might more strongly support
a model closer to one of the polar models.
But the failure of either pole to satisfy in the case of Altaic
does suggest that we should use some frameworks
for displaying our data and results which recognize the possibility
of an intermediate as the closest to the truth.

Trask writes:

>Anyway, I can see absolutely no point in invoking hypothetical "substrates".
>If we have no hard evidence for the reality, and the nature, of a particular
>substrate in a particular case, then there is no point in raising the issue.

Certainly there is, if the point is to accumulate and display data
in such a way that evidence for a substrate or for "language area" sharing
or similar intermediate models can be recognized more easily when
it might be staring us in the face.  Excluding such possibilities
from the beginning is an unjustified bias against them.

*

One, but only one, possible outcome in such a case can be the
recognition of elements common between on the one hand
(a) one portion of a dialect network of a fused language area "F", or
(b) one polar focus of a fused language area "F"
(whichever view one takes of a fused language area)

and on the other hand a portion of a neighboring fused language area "G",
which suggests that fused language area "G" has as one of its components
something which is intermediate between being the effects of a dialect area
and the effects of a component language in fused language area "G".

(This is not what I think is most typically meant by "substrate",
but whether one uses that particular term or another
is not my concern here, the substantive facts are.)

In such a case, one might consider the possibility that there was once
a language "L", a part of which was fused into area "F"
and another part fused into area "G".
Absent such (or of course other external evidence which might support
one of the more polar models, clear genetics or clear borrowing),
we may simply remain with a "language area" as the best statement
of our results.

***
Trask writes:

[LA]
>>  To recognize that this is a possible situation
>>  for a proto-language,
>>  we must handle vocabulary and morphological
>>  distributions across *portions* of the dialect network of any
>>  proto-language in *at least* the frameworks of the following

[LT]
>Sorry, but this is putting the cart before the horse.
>
>We cannot identify any possible cases of variation within a proto-language
>before we have first identified the proto-language itself.

Unless the above is definitionally tautologous (and therefore without
empirical content, describing the status of a particular model not
the status of a real-world case),
I must reply that this is simply on its face wrong.

Much of the work on Altaic has consisted in doing both of these things
simultaneously, accumlating vocabulary distributions
as areal phenomena, and developing family-tree analyses.
Neither needs to precede the other.

Variation *within* a proto-language which is *not* demonstrably
linked to the genetic family-tree structure of the proto-language,
and which is *not* simply a demonstrable result of waves of borrowings,
may be discovered or recognized
*at the same time* as one is establishing a genetic tree or a dialect network
or other model as partially fitting that particular proto-language.

The two can proceed quite together, and in fact may reinforce each other.
The fact of some *partial* shared genetic origin must of course be assumed for
such a discourse to make sense.  But one need not know the answer
to any one part of it with complete assuredness at any one time.
Even a "final" conclusion, in the sense of the best a set of techniques
can do given a set of data available, may very well need to rest with
a conclusion that there was a proto-language, but
(a) what we can reconstruct from the evidence
     is not a uniform proto-language, and
(b) perhaps there never was a uniform proto-language.

Mr. Trask agrees that proto-languages need not be completely uniform.
I think by that, he is agreeing with the point being made here,
once we add modesty about the limitations of our techniques and conclusions.

Asking that the data relevant to studies of historical relationships
be assembled and visually displayed so far as possible
*independent of any conclusions one may wish to draw from such data*
is merely responsible common sense,
it is more conservative and careful than what Mr. Trask appears to be
advocating, which is to draw one kind of conclusion, then only to consider
questions (even some which might undermine that first conclusion)
within the limits imposed by that conclusion.

One important difference does come in just that, the ability to
carefully consider new data as evidence,
or even to consider previously known data from new perspectives, etc.
If we cannot see the data *except* restricted by a particular model,
especially a polarized model,
then the data which might lead to an eventual demonstration
that another model is required, also or in stead of a prevailing one,
simply cannot be gathered.  Each individual piece of data is ruled
out of order by the assumptions of the model if it is taken as representing
the reality with complete adequacy and no strain whatsoever.
So the fact that a number of pieces of data do converge on an alternative
synthesis will not be seen.
A classic case of Kuhnian resistance to changes of view in science.

***

Trask's definitional approach, referring to discussion of
English having more links to French than to Italian:

[3rd party]
>>  >They're not my terms; I didn't invent them -- they are standard
>>  >in historical linguistics textbooks.  But I agree with the
>>  >premise -- except that I wouldn't say "equally related"; I would
>>  >say "related at the same level."
>
[LA]
>>  I of course agree.
>
[LT]
>And so do I.
>
[LA]
>>  Perhaps our common reluctance to use the phrase
>>  "equally related" here is that it has a portion of its ordinary-language
>>  meaning,
>
[LT]
>Well, it shouldn't.  [snip]

That is not an adequate response.

Mr. Trask can of course try to tell others what they "should"
mean by terms, in his view,
but that is quite different from what I was doing,
describing what they *do* mean by terms.
*Including* those who use *genetically related* in a sense distinct
from borrowing, and who sometimes recognize
a more subtle 3rd distinction, such intimate contact that neither
simple genetic inheritance nor borrowing seem correct terms,
rather something in between the two.

If we are dealing only with simplistic models,
then of course a model can rule out fuzzy edges and rule out intermediate
cases, but that makes any such model less widely applicable in a real world
which actually contains such intermediates.
There is no justifiable burden-of-proof, by which we can somehow
magically know that one model is more true of a given
real-world case than another model *independent* of any actual evidence
(both models here assumed to be valid for some known languages).

***

Lloyd Anderson
Ecological Linguistics
PO Box 15156
Washington, DC 20003