Dixon's "The rise and fall of languages"

Sat May 9 16:36:01 UTC 1998

----------------------------Original message----------------------------

I have just read R.M.W. Dixon's "The Rise and Fall of Languages"
(C.U.P. 1997), and I have found it deeply disturbing.  For two
completely different reasons.

The book (or essay) itself carries two separate messages.  One is an
attempt at a new theory of comparative linguistics.  The other is a
passionate plea for linguists to do what they should do, before it's
too late.

Thousands of languages will be irretrievably lost in the next hundred
years or so, and there is nothing we can do about it.  Dixon conveys
this disturbing message forcefully, while at the same time urging
us ('people who call themseleves linguists') to literally drop
everything and record the languages that are still extant NOW.

        The most important task in linguistics today  -- indeed, the
        only really important task -- is to get out in the field and
        describe languages, while this can still be done.  [Other
        things] can wait; that will always be possible.  Linguistic
        description must be undertaken now.

And:

        If this work is not done soon it can nver be done.  Future
        generations will then look back at the people who call
        themselves 'linguists' at the close of the twentieth and
        beginning of the twenty-first century, with bewilderment and
        disdain.

I don't know how this message could be conveyed more forcefully.  It
is with a sense of personal shame, as someone who unlike prof. Dixon
has never done fieldwork, that instead of turning to the Amazon
jungle I now turn to the other aspect of Dixon's book.

The new model that Dixon proposes, the "punctuated equilibrium
model", aims to integrate two distinct views on the development of
langage: the "family tree model", and various models based on areal
diffusion of linguistic pehenomena.

Dixon makes four basic assumptions:

1) Languages are always in a state of change
2) The rate of change is not constant and not predictable
3) Core vocabulary is not (universally) more resistent to change than

   "non-core" vocabulary
4) In the normal course of linguistic evolution, each language has a
   single parent

Assumption (4) denies the validity of the "rhizotic" models of
'glotto-genesis', at least "in the normal course of linguistic
evolution".  The existence of exceptions or apparent exceptions, like
mixed languages (Ma'a and Copper Island Aleut are reviewed) and maybe
creoles is acknowledged.

Assumptions (2) and (3) are aimed at lexicostatistics and
glottochronology as methods of linguistic classification and dating.
Unlike assumption (2), assumption (3) is surprising:

        This [greater stability of core vocabulary] does appear to
        hold for the languages of Europe [..] and of many other parts
        of the world.  But it does not apply everywhere [..]  In
        Australia, for instance, similar percentages of shared
        vocabulary are obtained by comparing 100 or 200 or 400 or
        2,000 lexemes, from adjacent languages.

Dixon reports evidence from New Guinea suggestive of the same
phenomenon.

Unlike, for instance, Malcolm Ross in his essay "Social networks and
kinds of speech-community event" (in: "Archaeology and Language I",
Roger Blench & Matthew Spriggs, eds., Routledge 1997), Dixon does not
give an exhaustive catalog of inguistic phenomena that are at odds
with the traditional model of the "family tree".  We are told that
"the family tree model was developed for -- and is eminently
appropiate to -- the Indo-European (IE) language family", that in
many parts of the world (Africa, Australia) no proto-languages have
been reconstructed [although in the case of Africa this is formulated
as "no attempt has been made" (p. 33), leaving us to wonder if the
fault is the comparative method's or the Africanists'], that
proto-languages [unlike real languages] are often too tidy and too
regular [?], that there is no linguistical method to assign dates to
proto-languages, that sub-grouping is often a problem, and that not
all sound-changes are recoverable (Appendix).

One argument that I hadn't seen before is given on pp. 29-30.  It is
based on the assumptions that PIE is 6,000 years old, and human
language 17 times as old as that, 100,000 years.  At the IE rate of
language split (100 IE lgs. in 6,000 years), that would have given
10^34 languages today (or even 10^60 at the Austronesian rate), when
in fact we only have 5,000.

        If we adopted a more modest rate of splitting, supposing that
        a language is likely to split into two languages every 6,000
        years, we would expect proto-Human to have given rose to2^17,

        or about 130,000, modern languages.  If we allow for a number
        of languages becoming extinct along the way, this would be a
        more reasonable number. [..] The lesson from these
        calculations is that language split and expansion on the
        scale that is put forward for the IE and Austronesian
        families is highly unusual.

Hold it right there.  Of course the expansions of IE and AN are
unusual. But do the numbers constitute an argument against the family
tree model?  Do they support the notion that "Language development
during the past 100,000 and more years has involved long periods of
equilibrium, with only the occasional punctuation"?  Dixon correctly
notes that the IE expansion in Australia and the Americas has been at
the expense of a 1,000 languages dead or dying.  We cannot just
"allow for a number of languages becoming extinct", we have to
include the extinctions in the mathematical model.  If that's not
done, the argument comes dangerously close to rejecting the "family
tree model" for rabbits because of the absurd rate of growth forecast
by the Fibonacci series.

Introducing the notion of "punctuated equilibrium" into linguistics
is an interesting idea.  I certainly agree with Dixon that certain
phenomena of language change (transitions between head and dependent
marking, certain kinds of phonological and lexical change, etc. can
and do take place in short periods of time), and that even the origin
of language itself might be adequately explained as a sudden
punctuation.

On the other hand, I'm deeply troubled by some of the unexplained
assumptions in Dixon's model.  Most importantly, Dixon's assertion
that a state of equilibrium is unobservable:

        European scientists have only ever been able to observe a
        time of punctuation since, wherever Europeans go (with their
        weapons and religions and writing), they effect a punctuation
        in the existing state of equilibrium

and:

        No equilibrium situation ever has been or ever could be
        observed by a scientist (although it can be readily
        reconstructed for Australia, and probably also for other
        parts of the world)

I'm also troubled by the notion of a "punctuation" that keeps going
on for "the last 2,000 years" (p. 4).

But let's allow these premises and see what difference they make in
practical terms for historical linguistics.

Dixon states the family tree model is only valid for periods of
punctuation, when a "proto-language" spreads out and diversifies into
a number of daughter languages.  After some time, equilibrium is
restored, and, within an area of diffusion, languages are repeatedly
said to "converge on a common prototype", regardless of their genetic
origin.  It is well known that phonology, lexicon and grammatical
categories are readily diffused.  Grammatical forms (morphemes) are
much more resistant, "but during periods of equilibrium there was
time-a-plenty (perhaps tens of millennia) and then grammatical forms
certainly were borrowed".  Language families "are slowly blurred" (p.
71).  "In time, the convergence will obscure the original genetic
relationships" (p. 96).  "Family membership ceases to be a useful
concept" (p. 99).  But, perhaps surprsingly, the languages in a
linguistic area in equilibrium do not merge (p. 71).  And most
surprisngly of all, when a new episode of punctuation ensues, and new
language families are created, Dixon warns that "the language family
may have emanated not from a single language, but from a small areal
group of distinct languages, with similar structures and forms".
This obviously undermines the very concept of a "genetic group", in
spite of Dixon's basic Assumption #4 that "each language has a single
parent".

In even more practical terms, Dixon reviews some concrete examples:
Austronesian (a classical case where the family tree model works
adequately in general), Australia and the Americas.

There is considerable controversy at the moment about the date of
earliest human settlement of the American continent.  There is
surprisingly little evidence for settlement before the Clovis horizon
of 12,000 BP, although a slightly earlier date (perhaps 16,000 or
20,000 BP) is now beginning to be accepted in archaeological circles.

The great linguistic diversity of the Americas, however, is a major
problem for linguists, and leaves only two options open: either to
accept the archaeological dates and to hell with the linguistics (my
assessment of Greenberg's "Language in the Americas"), or to posit a
more reasonable date for the initial settlement and to hell with
archaeology (my assessment of Johanna Nichols' argument for a
time-depth of c. 35,000 years, which I share).  Dixon, armed with his
model of punctuated equilibrium, sees no problem:

        [Nichols' argument].  I take a viewpoint that is
        diametrically opposed.  The fact that so many language
        families are recognisable indicates a relatively recent
        series of language splits, quite compatible with a
        12,000-20,000-year period.  Give the languages in the
        Americas another 20,000 years and the diffusional patterns
        that are now emerging would become far more pervasive.

Counterintuitive to say the least.

But, given the authors admission that "It was largely in order to
adequately account for the linguistic situation in Australia that I
had recourse to the idea of P.EQ. as a model for language
development", we should turn to Australia next.

The c. 260 languages of Australia show many similarities, and "have
been said to comprise a single language family".  One supposed
(sub-)family, Pama-Nyungan, covers about 85% of the continent.
However, neither for Pama-Nyungan (apparently [p. 91] a
lexicostatistic construct), nor for Australian as a whole have family
trees or proto-languages been succesfully set up.  "It is possible to
establish low-level subgroups in Australia -- groups of from two to a
dozen or so languages that appear to have a close genetic
relationship", but the usefulness of the family tree concept as far
as Australia goes is apparently so low, that Dixon does not even
bother to mention the number of distinct genetic groups.

        It is likely that the whole of Australia was populated within
        a few thousand years of the first colonisation, about 50,000
        BP.  Many scholars believe that all Australian languages
        belong to one linguistic family.  Assuming this hypothesis
        there are two alternative scenarios:

        (i) Proto-Australian, the putative ancestor of all the modern
        languages, was spoken by some of the first people in
        Australia, about 50,000 years ago.

        (ii) Modern Australian languages are descended from a proto-
        langugae that was spoken much more recently -- say, 10,000 or
        5,000 years ago.

Hypothesis (ii) is dismissed, because we do not find pockets or
substrates of non-Australian languages, and there is no plausible
mechanism (like agriculture) to explain the expansion.  "We are left
with scenario (i)":  rapid expansion (punctuation) throughout the
whole continent, followed by essentially 50,000 years of equilibrium
until the English invasion, during which time the Asutralian
languages gradually converged to "a common prototype".  This is, as
Dixon states, "the only model able to explain the relationships
between languages in Australia" (p. 68).

Is it?  One might object that New Guinea, settled at the same time as
Australia, and united to it by a land-bridge until c. 10,000 years
ago, despite some areal characteristics, and despite an attested
punctuation with the arrival of Austronesian-speakers maybe 4,000
years ago, shows a bewildering linguistic diversity, parallelled
nowhere on earth.  Dixon blames this on the mountainous character of
New Guinea (and, similarly, the Caucasus).  And then, how tranquil
was the Australian equilibrium during those 50,000 years?  On p. 92,
Dixon suggest that the low-level genetic groupings that he recognizes
in Australia arose during the last of possibly several cycles of
contraction of the population to the coast and main rivers during
periods of drought, followed by expansion when conditions improved.
Repeated episodes of contraction and expansion may well have blurred
the genetic affiliations of the Australian languages by the processes
described by Ross (loc.cit.), such as language/"linkage" fusion.
Elsewhere (note, p.76, we are told that the population of Tasmania,
part of Australia until 10,000 years ago, did not have axes,
spear-throwers, boomerangs or dingoes.  It seems to me that the
introduction of those items, after 10,000 BP, might have caused quite
a bit of punctuation (and incidentally makes a good candidate for
the mechanism of expansion required by scenario (ii) above).

Finally, Ross (op.cit, p. 244), mentions an important socio-cultural
factor, not mentioned by Dixon at all, which seems to be of the
utmost importance to explain the "blurring" of sound-correspondences,
and the consequent difficulty in establishing family trees and
proto-languages.  It bears repeating here.  The case has been
described for a group of related languages in southern New Caledonia:

        Because the south New Caledonian languages were closely
        related, there were once regular sound correspondences
        between their vocabularies.  Where speakers regularly used
        two or more lects, they had an intuitive grasp of some of
        these correspondences and used them to convert the
        phonological shapes of words from one lect to another.
        However, the speakers' intuitive correspondences and the real
        correspondences resulting from historical change often
        differed from each other.

The result of this "Volkskomparativismus" is that the
sound-correspondences are FUBAR, a complete mess.

        Similar situations to the one described for New Caledonia
        also occur in parts of Australia.  In traditional aboriginal
        Australia each person belonged to an exogamous patriclan and
        spoke its emblematic patrilect.  However, aboriginal
        Australians moved around hunting and gathering in bands whose
        members belonged to different patriclans.  A number of
        patrilects, often quite closely related to each other, were
        typically represented in a band, and band members spoke their
        own and other members' patrilects.  Their vocabularies seem
        to have been affected in much the same way as those of the
        south New Caledonian lects.

While this picture is not in itself incompatible with Dixon's state
of equilibrium, it offers the advantage of actually explaining why
and how some of the languages might have converged in a way that is
not readily tractable by traditional comparativist means.

Finally, a word of caution to linguists.  As Dixon states in his
introduction, "many groups of linguists may be offended by what is
said about their area of specialisation".  Africanists may not like
Dixon's assessment of their field on pp. 32-33.  If there are still
some lexicostatisticians/glottochronologists around (and Dixon
chastises on Australian member of the species) they won't like pp.
35-37.  "Armchair typologists" and linguists *talking* about
"endangered languages" are dealt with elsewhere.  The heaviest
criticisms, however, are reserved for two sub-species: the
formalists, and the Nostraticists.

Formal theories (Chomskyan or not) have a "typical half-life [of] six
to ten years", and "few formalists do attempt to write comprehensive
grammars of languages (which is just as well [..])", although some of
them do from time to time consult the descriptive gammars written by
"real linguists".  They are "like a group of 'surgeons', none of whom
has ever actually performed an operation, giving courses of lectures
on the principles of surgery".

Nostraticists (or 'Nostraticists' as Dixon calls them), especially of
the Russian kind "openly boasted (and still boast) that they are
cleverer than anyone who has come before" [reference?], their
theories are "palpable poppycock", and "they have put forward the
idea that the main thing to be considered when formulating a genetic
connection between two languages is lexemes" [no reference].  "There
is no reputable linguist [defined as "anyone who teaches the subject
at a leading university in the USA or in a EEC nation"], anywhere in
the world, who accepts the claims of Greenberg and the
Nostraticists".  If we accept Dixon's punk-eek model, "there could be
no tempation to perpetrate anything such as 'Nostratic'".

        These 'Nostraticists' purport to work in terms of the
        comparative method, by assembling cognate sets.  However,
        they achieve their results only by allowing excessive
        phonological and semantic leeway.  In the 'reconstructions',
        scarcely any vowels are specified (given just as V), N is
        often employed for an unspecified nasal, and so on.

Regrettably, Prof. Dixon, with this last remark, makes it painfully
clear that he has never so much as set eyes on Illich-Svitych's
Nostratic dictionary...

Apart from the unfounded accusations, the only reasoned critique
against the "Nostratic fallacy" in Dixon's book is the following:

        It is not sensible policy to try and compare the original
        proto-languages of language families, and attempt to
        reconstruct a  proto-proto-tableau.  Firstly, we have only an
        approximate idea of what a proto-language was like. Secondly,
        it may not have been one language, but instead a group of
        languages.  And thirdly, proto-languages or
        proto-linguistic-situations are likely to be the product of
        diffusional convergence, at the end of a period of
        equilibrium, rather than languages which result from a
        family-tree-type expansion and split.

Of course, to agree with objections (2) and (3), one has to buy into
Dixon's punctuated quilibrium model, which I'm not prepared to do
right now.  As to the first objection, well, let's do away with
archaeology and paleontology as well, then.

=======================
Miguel Carrasquer Vidal
mcv at wxs.nl
Amsterdam