LINGUIST List 5.1116

Thu 13 Oct 1994

Disc: Comparative Method

Editor for this issue: <>

Directory

1. Karl Teeter, Depth of Comaprative Method
2. , Comparative Method
3. Stephen P Spackman, Re: 5.1111 Limits of the comparative method

It is certainly true that when something is gone completely there is no
reconstructing it, so SOMETHING of a language must be left. But the
comparative method is not, as suggested, primarily as matter of lookign
at similarities among vocabulary items, but of writing a grammar of a
putative protolanguage, just like writing a grammar of a regular
langauge based on work in the field. In both cases thee must be
something left, but we cannot determine in advance how much we need, so
no real meaning to ceilings. Yours, Karl (=Karl V. Teeter, Professor of
LInguistics, Emeritus, Harvard University)

In response to Alice Faber, Jacques Guy, and Matthew Dryer:
I would like to clarify some issues:
(a) It is not the same thing to say (as Jacques does) that there
may be some in principle limit on the CM although we have no idea
what it is, and saying (as Nichols, Bender, Kaufman, etc. have done)
that there is a very shallow time limit of somewhere between 7000
and 10000 years. I am not against the idea that there might be
an in principle limit, and I certainly do agree that there is a limiit
due to the fact that there are only so many languages on Earth. (The
main argument against a limit is that, if you increase the number of
languages, you increase the time depth that can be reached--but, as
I just said, this argument only goes so far because we run out of
languages).
(b) It is, as I have said before, also not the same thing to say
that the CM can do no more than it already has simply because
it has not done any more and to say that there is some independent,
non-circular argument about its limitations. In the discussion so
far, I have yet to see argument of the later kind, but it is certainly
not inconceivable that such an argument could be made.
(c) It is not the same thing to observe (as Alice seems to) that
over time individual languages will tend to lose most of the
original vocabulary as it is to claim that over time NO TWO of
the languages in a family will tend to lose the original vocabulary.
It can be shown, on purely probabilistic grounds, that, given
a large enough family, the chances are very good that much of the
original vocabulary (certainly enough to establish that it IS a
valid family) is likely to survive in at least 2 languages for
much longer than 10 thousand years, although at that time depth
no ONE language is likely to preserve much of the original
vocabulary.
(d) It seems very clear to me that the reason for the claims
about a time ceiling on the CM is simply that some people want
to have a simple way of justifying not merely questioning or
rejecting but basically refusing to even seriously consider
certain hypotheses, such as the Amerind or the Nostratic. While
this does not apply to any of the people I am responding to here,
I think we would not be discussing the issue if that were not
the case. Seen in this light, the distinctions I made in (a)-(c))
above become more than academic, since we are talking about whether
it is legitimate to advance what seems to me a completely unsupported
claim of a cut-off point at 7000-10000 years as a way of precluding
serious discussion of whether either of these hypotheses (or any
part of them) might be correct.
So, while agreeing with much of what Matthew, Alice, and Jacques
have to say, I would like to focus this a bit, by simply asking
not whether they (or anybody) believes that there might possibly
be a time ceiling of either unknown or unknowable magnitude but
rather that there actually is--and has been demonstrated to be--
a ceiling shallow enough to make either Nostratic or Amerind
in principle non-empirical hypotheses. For, if the ceiling is
at 40000 years or is of unknowable magnitude, then for the
purpose of debating the possible validity of such hypotheses, it
will have no significance.
Alexis Manaster Ramer

Message 3: Re: 5.1111 Limits of the comparative method

In order to develop confidence about deep reconstructions, it may be
necessary to take the bull of uncertainty by its horns and start
treating reconstruction as an *experimental* undertaking, using
statistical methods and explicit controls to quantify the errors and
validate the results. It's probably not quite attainable at present,
but one procedure that would permit this would be to require (1) that
all alignment and reconstruction be done according to FULLY formal rules
(formal enough that they can be applied by a computer - no on-line
judgments allowed, and with detected exceptions taken to degrade the
reliability of the result) and (2) that experimental "runs" be done on
groups containing controls, in this case languages believed NOT to fall
within the relevant groupings. Semantic alignment should be part of the
controlled process, of course, which means that the bases for comparison
will need to be bigger (we might need to have BOTH "wolf" and "dog" in
ALL our bases, for instance :-), but that's ultimately all to the good.
Aside from geographically removed natural languages, one useful control
"language" is arrived at by shuffling together data from all of the
experimental languages. This should of course NOT be assignable a
position within the reconstruction, since there is no CONSISTENT way in
which it could be related to all of them. It should appear unrelated
when the time depth exceeds some lower limit.
Passing this shuffled-lexicon test should distinguish viable
reconstructions from the fallacies of optimism quite sharply, since it
is designed specifically to pander to the latter.
stephen p spackman +49 681 302 5288(o) 5282(sec) stephenacm.org
 dfki +1.24 / stuhlsatzenhausweg 3 / D-66123 saarbruecken / germany
http://cl-www.dfki.uni-sb.de/~spackman finger:spackmandfki.uni-sb.de