Re: [tied] Re: searching for common words for all today's languages

From: Brian M. Scott
Message: 43324
Date: 2006-02-09

At 8:10:19 AM on Tuesday, February 7, 2006, tgpedersen

> --- In, "Brian M. Scott"
> <BMScott@...> wrote:

>> At 6:06:39 AM on Monday, February 6, 2006, tgpedersen wrote:

>>> --- In, "Brian M. Scott"
>>> <BMScott@> wrote:

>> [...]

>>>> Salmons did a pretty thorough job on Ruhlen's alleged
>>>> *tik word; some of the criticisms are specific to that
>>>> word, but the (rather serious) methodological criticisms
>>>> carry over to the rest of the list.

>>> He probably did, I haven't read it. AfaIk no one proposed
>>> so far that the alleged cognates were loans, much less
>>> refuted that idea.

>> Actually, many of the methodological objections to the idea
>> that they're cognate are also damaging to the idea that
>> they're loans:

>> * unconvincing semantics;
>> * absurdly generous criteria for phonological matches;
>> * the fact that he ignores time depth.

>> In short, it's not clear that there's anything there to be
>> explained in the first place, as cognates *or* as loans.

> Applying the above objections to the idea is a categorial
> mistake, they must apply to the individual members of the
> individual proposed sets (at least wrt the first two,

Not a category mistake in the way that I meant it: these
arguments cast significant doubt on the idea that the **tik
words, for instance, are a loan-group. The unconvincing
semantics are every bit as much of a problem for loans as
for cognates; the absurdly generous criteria for matches are
somewhat less of a problem. It is of course still possible
that *some* of the items belong together as loans, just as
it's possible that *some* of them are cognate, but each pair
or plausible small subgroup would have to be investigated on
its own merits.

> I'm not sure what the third means?).

Ruhlen uses attested forms in particular languages,
including isolates; shallow reconstructions in tiny
families; remote reconstructions in large and well-studied
families; and reconstructions in highly speculative
macro-families like Amerind. Given a form, he's perfectly
willing to accept as a match either an attested form in a
language X or a reconstructed form in a language ancestral
to X; this obviously greatly increases his chances of
finding a match. Whether or not the extra matches allowed
by this approach are more likely to be false positives --
and I think that they are -- it seems clear that true
positives among them will be harder to identify as such.

> Thus they can be used to reject individual members of sets
> or individual sets of cognates, not the theory. If you
> have such objections for eg. the "aqua" set, let's hear
> them.

As far as I'm concerned, that's the wrong question: the
burden of demonstration is on those who wish to claim a
relationship, by descent or by loan.