The Comparative Method and "semantics"

X99Lynx at aol.com X99Lynx at aol.com
Sat Sep 18 06:08:13 UTC 1999


I wrote that "if two words are a phonologically the same [within the workings
of the sound rules, of course], they have a high probability of common
ancestry"

[kurisuto at unagi.cis.upenn.edu] wrote that's "incorrect."  Because "There are
several other ways that such pairs could arise."

I wrote:
That doesn't follow.  The fact there are other ways they could arise only
means that the probability is not 100%.  It doesn't mean the probability is
not high.

On 9/16/1999 10:54:47 PM, mclasutt at brigham.net writes:
<<But you haven't demonstrated that the probability is even high.>>

#1 - But let me point out that Sean Crist's statement about other ways pairs
could form does not disprove it.  Thats important as a starting point because
it brings up the issue of probabilities.  (Or do you think Sean's statement
does disprove high probability" so in your opinion there's no need to go on?
Just to check on what you think would make the high probability statement
false.)

#2 - You write:
<<Take, for example, the English [ber] 'bear (n)', 'bare', 'bear (vb)'.
According to
your criteria, these three should have a high probability of being part of
the same cognate set.  They aren't. >>

Do you think the examples you mention disprove the high probability claim for
phonetic match-ups (within the sound laws of course)?  You realize of course
that a few examples could represent extreme deviations from the norm
statistically.  Is there any research that yields the percentage of cognation
that can be expected among words whose etymology can be documented?

BTW I was trying to restate the "exceptionlessness" of the sound laws.  This
is often called a working assumption and an effective one.  If it is
effective than presumably it has a "relatively" high probability of revealing
cognation.  If phonetic matching doesn't give one a much better than
fifty-fifty chance of cognation than that certainly does put the onus on
"semantics."  Which brings me to the next point:

#3 -   You write:
<<the English [ber] 'bear (n)', 'bare', 'bear (vb)'.  According to
your criteria, these three should have a high probability of being part of
the same cognate set.  They aren't.>>

You say they aren't.  May we examine that on "semantic" grounds?  How sure
are you of that?  What are your grounds?

You write:
<<The first is from PIE *bheros 'brown' (I'm using the PIE forms in the
Oxford Dictionary of English Etymology); the second is from PIE *bhosos
'uncovered'; and the third is from PIE *bher- 'carry, give birth'.>>

You've used reconstructions as your source.  Are those reconstructions based
on other sound matches?  But let's say you didn't have those reconstructions.
 How would you go about showing that <<[ber] 'bear (n)', 'bare', 'bear
(vb)'>> are not cognate?

Most importantly, how would you estimate the chance of error (reconstructing
on your own) in your finding that the three forms are not cognate?  Would you
allow for any unknown history concerning these words?  I don't have the old
OED with me, but my guess is that the latest of these words is not documented
before 1000 AD.  Would say 5000 years of undocumented meanings (in languages
undergoing ceaseless change) would be enough for you to concede even a small
degree of uncertainty?  (Neither do I have the original forms these words
were first documented in, but do those early forms show phonetic matches?  If
they don't, shouldn't you be using those first documented forms to determine
phonetic relatedness?)

Based on the above, how much uncertainty statistically would you estimate?
How would you calculate your uncertainty in standard deviations?

#4 - You write
<<Without considering semantics, we're lost. >>

In the wake of the Ringe quote Sean Crist gave us, my question is whether
that is such a bad state to be in.  Especially with regard to semantics.
It's not that we can't guess at meaning, but that we should admit that we are
guessing.  (For example, I have some reason to think that PIE color terms are
anachronisms, that the ancients described objects with only incidental
reference to what we now call color.  *bheros 'brown' should apply to every
other brown animal from here to Anatolia, along with everything else vaguely
brown.  And that makes bear = brown seem even weaker.  But I really have no
way of putting a number on my uncertainty.  But bear = brown will do until
something else comes along.)

<<Only by looking at semantics do we know which of the potential IE
sister forms are really cognate with each of these three.>>

Or are misled into thinking we "know" their relationship.  If "bear" was
originally an analogy to some other animal or a reference to its pelt as
covering or if brown derived from bear or whatever, perhaps 3000 years
separates us from those meaning and I question too much certainty about such
matters.

<<If you're going to make the claim that phonetics alone can
establish cognate sets with a "high probability", then prove it.  Conduct an
experiment in which you do just that and then show us the probabilities so
we can see your evidence of "high probability".  Don't just argue from
logic, use the Baconian model and perform the experiment.>>

I'd love to accept the challenge, but I'm afraid that it demands some serious
resources.

But let me ask you this..  If we gathered all the word pairs we could find in
a language who had historically documented common ancestors (no
reconstructions for obvious reasons), wouldn't you expect to find a high
degree of phonetic similarity?  And if we gathered all the homonyms with
historically documented non-cognancy, wouldn't you expect to find a much
lower frequency?   Or do you expect that documented cognancy would show no
particular correlation with phonetic  similarity?

Regards,
Steve Long



More information about the Indo-european mailing list