LINGUIST List 9.757

Thu May 21 1998

Disc: Limits on Knowledge in Linguistics

Editor for this issue: Martin Jacobsen <>


  1. F. K. Lehman (Chit Hlaing), Re: 9.740, Disc: Limits on Knowledge in Linguistics
  2. Fidelholtz James L., Re: 9.740, Disc: Limits on Knowledge in Linguistics

Message 1: Re: 9.740, Disc: Limits on Knowledge in Linguistics

Date: Tue, 19 May 1998 09:52:52 -0500
From: F. K. Lehman (Chit Hlaing) <>
Subject: Re: 9.740, Disc: Limits on Knowledge in Linguistics

Alexis Manaster-Ramer makes an important point (actually several). In
fact, as it appears to me at least, there are at least two sources of
very dfferent sorts for the idea that there is, or ought to be a sort
of upper limit on the capacity of at least available comparative
methods to demonstrate more or less unambiguously (notice the
deliberately weak phrasing here) genetic relationships.

	On the one hand, as AMN says, there are some
order-of-magnitude numbers associated with this sort of claim and
those numbers are in fact calculated in the framework of
glottochronology. Such stuff is, of course, dubious both because of
the dubious character of glottochronology itself and because by now no
one seems sure what these numbers are or ought to be (and, yes,
Nichols indeed waffles on this matter egregiously). In this connection
the question is not so much whether such stuff is still taught (it is)
but rather whether glottochronology is still being taught. Yes, in
some places it is, namely, in the cntext of anthropological
linguistics (a hybrid sometimes uncharitably caricatured as being
practiced by people who can excuse themselves by noting that they are
not actually either linguists or anthropologists), and the reason, so
help me!, is that, though it may be exploded, 'we have nothing better'
for trying to calculate depths of divergence. That is to say that some
historical anthropologists would like to be able to refer to some
independent linguistic evidence for their estimates of intergroup
relationships, and they seem to feel that they can rely on very
someone else's definite numerical claims, apparently based upon
'rigorous' calculations, without having to know much about the field
in which those calulations have been made

	However, the other source is actually not one that can
plausibly produce this sort of upper limits at all. Rather it is the
proposition, possibly not very controversial, that beyond some point,
and especially given the 'thinness' of the corpus of evidence and
generalised or formulaic character of reconstructions for supposedly
very remote relationships, it simply becomes increasingly hard to
distinguish between similarities due to genetic relationships and
those due to convergences based on the aforesaid universals (e.g.,
possible typological generalisations). Any upper limits associated
with such an argument cannot possibly, however, be given as even
ball-park definite numbers (give or take the odd few millenia, as AMR
says), but only as a rate of increasing difficulty with an
<underline>arbitrarily great</underline> upper bound. I suspect that
at least some people involved in this sort of work may not understand
that 'limits' of this sort are not to be associated even in principle
with definite numbers.

	For current practical purposes, of course, and given the fact
that our methods can always use refining and the fact that our
understanding of, say, phonetic/phonological universals is extremely
gross and uncertain, there may indeed be a <italic>pro
tempore</italic> more definite upper limit (God knows what it is!),
but it is not fixed in principle. The matter gets very sticky, of
course, at this point because the farther back one attempts to extend
one's comparative methods, the more it becomes possible to claim
<underline>either</underline> that universals are themselves due to
very distant genetic relationships <underline>or</underline> the
contrary, and at this juncture the arguments seem to become circular
and confusing. In fact (here AMR is better qualified than I am,
certainly) the question arises whether, in the context of such numbers
and methods, the arguments as between universals and monogenetic
origins of language are computationally resolvable.

F. K. Lehman
Department of Anthropology
University of Illinois at Urbana-Champaign
109 Davenport Hall
607 South Mathews Avenue
Urbana, IL 61801
U. S. A.
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue

Message 2: Re: 9.740, Disc: Limits on Knowledge in Linguistics

Date: Wed, 20 May 1998 11:29:08 -0500 (CDT)
From: Fidelholtz James L. <>
Subject: Re: 9.740, Disc: Limits on Knowledge in Linguistics

	I have published in a couple of articles the claim that
vocabulary considerations limit us in principle to more or less 20,000
years in time depth. The calculations are simple. Glottochronology
supposes (on the basis of test cases like Vulgar Latin and others) a
rate of change around 17 or 18% loss per millennium. A simple series
of multiplications (.83 to the twentieth power) will give a residue
well under 4% (2.4 to 1.8%) after 20 millennia. Others (I can't at
the moment remember who) have calculated, using some fairly
noncontroversial assumptions about the phonetic structure of languages
(maybe it was Greenberg), that any two random unrelated languages are
likely to score about 4% on the Swadesh test just by pure accident,
with that then being the lower limit for showing a relationship other
than chance. Therefore, after 20,000 years of separation, any two
languages are likely to show less than a 4% similarity in basic
vocabulary, and therefore not be provably related via
glottochronology. Of course, there still remains the good old
comparative method, which, depending on what the residue is, might
just be able to provide some evidence, and syntactic or morphological
evidence, which is supposedly (reference: general knowledge osmosed in
the linguistic community) more resistant to change might also allow us
to push the limen back. Also, if one takes different figures from the
17% or 18% which is generally accepted, the dates would need to be
correspondingly changed (eg a 15% rate would give almost exactly 4%
after 20,000 years).
	In one of the aforementioned articles, I provide evidence
supporting the now generally accepted age of modern human language as
a probable minimum of 100,000 years (I don't think anyone anymore
accepts a figure under 40,000 years, and almost everybody would
probably push the minimum to at least 50,000 years), which figure
could go back even as far as 300,000 years ago or more. The clearcut
conclusion is that the theoretical limit on Swadesh's method ALONE is
about 16,000 years for nonchance resemblances. Considering the
advances (believe it or not) in studies on human language origins over
the last decade or so, I certainly would never eliminate the
possibility of going very far back in reconstructions, via some
methods yet to be thought of, or maybe even just inventive
applications of methods already though of. It does seem unlikely to
me, however, that, even if the hypothesis of the monogenesis of
language is correct, we will ever be able to prove it, just because of
the huge time depth and the necessary imprecision of the instruments
we have for reconstruction and language comparison.

James L. Fidelholtz			e-mail:
Maestri'a en Ciencias del Lenguaje
Instituto de Ciencias Sociales y Humanidades
Beneme'rita Universidad Auto'noma de Puebla, ME'XICO
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue