No Proto-Celtic?

Larry Trask larryt at
Mon Apr 30 14:19:16 UTC 2001

Steve Long writes:

[on my description of the views of Trubetzkoy et al. on IE as "crazy"]

> Oh, come now, Larry.  I challenge you to describe how the current outcome
> would be different if you assumed there was no PIE and the known IE
> languages arose out of convergence.  How would the situation be
> different?  What signs should be present in specific languages that are
> absent now?  Would you expect to find different classes of noun and verb
> morphology?  Would some words be n-stems and others not?  Different words
> for fire, eat and horse, perhaps?  What would tell you exactly that
> convergence had happened?  Or would there really be any way to tell?

Yes; certainly.

If the IE languages had really arisen by some process of 'crystallization'
or language mixing, then we would *not* expect to find the tightly
interlocking patterns linking these languages that we do find.  Instead, we
would expect to find only shared elements and miscellaneous resemblances --
without pattern, without system, without rhyme or reason.  In other words,
we would expect to find something similar to what we see with Tlingit and
Eyak-Athabaskan: commonalities without system.  And that is precisely what
we do not find.

> I don't think you'll be able to do this.  But I hope you won't avoid the
> challenge and at least try.  Because unless you can describe a different
> expected outcome, convergence becomes just as likely an explanation as a
> single parent.

No; not at all.  The null hypothesis is this:

The IE languages are not descended by divergence from a single common

This null hypothesis is spectacularly falsified by the data, by the
extensive and elaborate systematic patterns linking all of the IE
languages.  A convergence scenario predicts nothing more than shared
elements and resemblances.  This is not what we find, and so the
convergence scenario is falsified.

> The comparative method doesn't allow for convergence.  It assumes there
> was no convergence.

This is far too strong.  The comparative method only *addresses*
divergence, sure.  But it is not inconsistent with convergence; it doesn't
deny convergence; and it certainly does not assume an absence of
Convergence merely has to be handled outside the comparative method -- as
indeed it routinely is.

For example, Latin <lingua> 'tongue' fails to satisfy the usual IE
correspondences, and so it is regarded as a loan word from another Italic

Take another example.  Martin Bernal, in 'Black Athena', argues that a
number of Greek words and names must have been taken from Semitic or
Egyptian, merely because they happen to resemble something in those
languages.  But Jasanoff and Nussbaum, in their devastating critique of
Bernal's linguistic efforts, dispose of this argument in a number of cases.
They do this by showing that the Greek items in question are *predicted by
rule*, to the last phonetic detail, by the comparative IE evidence.
Convergence predicts only resemblance, while the systematic correspondences
among the IE languages predict exact forms.

This argument from systematicity is what makes PIE valid.  The systematic
nature of the correspondences forces us to reject weaker conclusions, such
as all those involving no more than contact and convergence.

> *PIE is a creation of the comparative method.

No, it's not.  I'm afraid this is a fantasy.

The comparative method is not a piece of abracadabra which can be applied
to any arbitrary languages at all and still produce a "reconstruction".
The method can only be applied at all in one circumstance: when we find
systematic correspondences among languages.

It is a common misconception among non-linguists that comparative
linguistics deals in resemblances.  It does not.  The comparative method
deals in *patterns*, and only in patterns, and resemblances have nothing to
do with it.

If we find some languages linked by an extensive network of systematic
patterns, then we can apply the comparative method successfully and obtain
a reconstruction -- even if there are no resemblances at all.  In very
great contrast, if we find that certain languages exhibit numerous
resemblances, but no systematic patterns, then the comparative method
cannot be applied, and no reconstruction can be obtained.  This is the case
with Tlingit and Eyak-Athabaskan, for which no reconstruction is possible.

In short, the method can only be applied at all when the systematic
correspondences objectively exist, and not otherwise.

> It
> may in fact reflect historical fact.  I suspect it does.  But I don't
> believe you have any way of proving that.

Sure, I have: the cases in which reconstruction can be performed, and those
in which it cannot.  That's a pretty big difference.

[on my astonishment about some "convergence" views of Celtic]

> First of all, a "trade language" as the term is used by a Sheratt or a
> Whittier, refers to how the language spread, not to its genetic
> characteristic.  "Lingua franca" is a term often applied to "genetic"
> languages, e.g., "North Sea" Germanic.

I'm afraid I don't know these two writers.  But I don't recognize the
distinction made here, and I can't agree that a lingua franca has to be
recognizably a single genetic language.  See the definition in any standard
linguistic dictionary, such as Crystal or Matthews.  The original Lingua
Franca, in the Mediterranean, was, according to all standard sources, a

Anyway, Barry Cunliffe, who I quoted in particular, uses 'trade language'
and 'lingua franca' pretty much interchangeably in his book -- reasonably
enough, I think.  This is just about the only point on which I don't take
issue with him.

> Secondly, what exactly would you look for, if Celtic was the result of
> some kind of a convergence?  Would it be like Haitian Creole, where
> according to Michel DeGraff at MIT, virtually all affixes have cognates
> in French affixes? Or rather would it be like English, where most affixes
>  are of non-Germanic origins?  What would you expect Celtic to look like
> if the convergence hypothesis were correct?  Especially after 5000 years.

OK.  First of all, we now have a number of examples of *individual*
languages with non-genetic origins.  These are quite varied in their
structures, and I'm not sure that many solid generalizations can be drawn.

Second, I am not sure that Haitian Creole represents a particularly good
example of a non-genetic language, since it is so strongly based on French.

Third, and most importantly, we appear to have *no single example* of the
kind of thing that is being proposed here for Celtic: a *group* of distinct
and seemingly closely related languages which have come into existence
through convergence, without a single common ancestor.  Steve, can you cite
a convincing example of such a thing?  I can't, and I don't think there is

Fourth, I have already explained what I think a "convergence" origin for
the Celtic languages ought to give us: a group of languages sharing common
elements and exhibiting a number of miscellaneous resemblances, but linked
by no systematic correspondences -- making reconstruction of a common
ancestor impossible.  But this is not what we find, not at all: instead, we
find numerous and detailed systematic correspondences, and we find no great
obstacles to reconstructing Proto-Celtic.  The convergence scenario is
therefore falsified.

> Is there an example of an IE language you can point to that shows what
> convergence would be like after 5000 years?

No.  I know of no substantially recorded IE language which cannot be
derived from PIE in the usual way.  As I remarked in an earlier posting, IE
languages are dreadful candidates for convergence origins -- even if some
other languages are excellent candidates.

> And therefore shows why we
> can be sure Celtic did not undergo convergence 5000 years ago?

This is not quite the proposal on the table.  The proposal of Renfrew,
Cunliffe and others is that the individual Celtic languages arose by
convergence from diverse origins over a period of 4000 years or more --
from about 5000 BC to around 1000 BC or later.

Of course, one might put forward a quite different proposal: that
Proto-Celtic arose by some convergence process, and then gave birth to the
individual Celtic languages in the familiar genetic way.  But this is not
the proposal on the table, and anyway the tables of systematic
correspondences linking Proto-Celtic both to PIE and to the daughter
languages are enough to falsify this view, too.

> Even if as you have said the whole idea of a language like Celtic is just
> a "reification."

Clarification: Celtic is not a language but a family.  The single language
in question is Proto-Celtic -- whose former existence I regard as proved on
the basis of the linguistic evidence.  Proto-Celtic is not a reification
dreamed up to satisfy somebody's ideology about how languages ought to
behave: it is a conclusion forced upon us by the evidence.

> Finally, the "highly regular phonological and morphological prehistory of
> Proto-Celtic (vis-a-vis PIE)" is the only kind of prehistory the
> comparative method would yield.  Of course, Celtic is going to show
> regular development out of *PIE.  *PIE is nothing but a construct based
> on Celtic and other IE languages.  It has no choice but to show regular
> developments and a Proto-Celtic.  *PIE is nothing but Celtic (along with
> other languages) reconstructed back to an assumed parent.

Steve, I am appalled.  As I explained above, if a group of languages do not
have a single common ancestor, then the comparative method cannot find one.
The method cannot find correspondences where none exist, and it cannot
create a common ancestor where none exists.  This is fantasy.  Where are
you getting this stuff from?

If you doubt me, then let me return to my earlier example.  *Why* has no
one been able to apply the comparative method to Tlingit and
Eyak-Athabaskan, and *why* has it not been possible to reconstruct a common
ancestor -- in spite of the numerous and undoubted common elements in these
languages?  Answer, please.

> If you sincerely were looking for evidence of convergence, you mention the
> obvious place to look, in such "picturesque novelties as the initial
> consonant mutations -- not to mention the Old Irish verbal system."  It
> wouldn't be in the consistencies that Celtic would show evidence of
> convergence,  it would be in the "innovations."

Er -- what?  I'm afraid this makes no sense to me.

Innovations are the grist for the family-tree model of genetic descent.
Convergence stresses diffusion, not innovations.  Innovations are a problem
for convergence views, since, in the kind of case under discussion, these
views require practically every innovation to diffuse across the *entire*
area covered by the converging speech varieties -- and we know that
innovations do not normally do that.

> But if we are dealing
> with possibly hundreds of EXTINCT European languages, IE or otherwise,
> how would you be able to identify other "genetic" influences?  Perhaps
> one of them had initial consonant mutation.

There is no way on earth we can take into account any features of languages
which are extinct and unrecorded, and trying to do so constitutes
fantasyland stuff in most cases.

Anyway, there is absolutely no need to appeal to hypothetical substrate
languages to account for the Celtic mutations: within our standard view,
that the Celtic languages are descended by divergence from Proto-Celtic,
and that Proto-Celtic is descended by divergence from PIE, the Celtic
mutations can be *wholly* explained by appealing to the reconstructed
Our explanation is entirely humdrum, and it is only the *later* development
of the mutations which makes them look mysterious.

> Calling the idea of convergence "dead and buried" will I suspect at some
> time in the future become nothing more than wishful thinking.

This is not what I said.  I expressly recognized that convergence is real.
What I denied was that IE languages, or Celtic languages, result from

> Calling such views "crazy" is a little uncalled for.

Steve, I didn't call convergence ideas in general "crazy".  I only applied
this adjective to the convergence proposals made by Trubetzkoy and others
for IE, and to the convergence proposals made by Cunliffe and others for
Celtic.  Please try to quote me accurately.

> I'm sure for example you
> wouldn't use words like "crazy" to describe the views of Boas or
> Trubetzkoy on the HistLing list.

Oh, I wouldn't hesitate, if I thought the label was appropriate.  But
Trubetzkoy's view of IE is the only one of his views known to me that I
would describe as crazy, while I know of no views of Boas that I would
describe as crazy.  Boas was unusually sensible, and his views were often
ahead of his time.

In fact, I almost sent my original posting to HISTLING, but I finally
decided that the IE-list was more appropriate, since my query was about IE
languages specifically, and not about convergence models in general.

One final comment.  Steve apparently suspects that I oppose convergence
ideas in general.  But I don't.  If my earlier posting didn't make this
clear, let me quote part of the entry for 'convergence' in my recent
dictionary, coming after a brief mention of some long-recognized types of

...Until recently, historical linguistics paid little attention to the
possibility of any further types of convergence, and divergence...was seen
as the primary phenomenon under investigation.  In the last few years,
however, convergence has begun to be taken far more seriously; the
examination of convergence phenomena is now seen as a major and growing
part of the field, and some linguists are suggesting that individual
languages can actually arise out of convergence...[I]t seems safe to say
that convergence will be a major theme in historical linguistics in the
years to come."

That clear enough?  I hope it's enough to show that I take convergence very
seriously indeed, and that I reject convergence accounts of IE and Celtic,
not out of ideology, but merely out of respect for the evidence.

Larry Trask
University of Sussex
Brighton BN1 9QH

larryt at

Tel: (01273)-678693 (from UK); +44-1273-678693 (from abroad)
Fax: (01273)-671320 (from UK); +44-1273-671320 (from abroad)

More information about the Indo-european mailing list