The Comparative Method and semantics

[Sean Crist]

In an earlier post, I stated that to be judged cognate, a pair of words in
related languages must both be derivable from the proto-language by
regular sound change.  Further, we have to have some plausible account by
which the meanings of the two words could have developed from some single
meaning in the proto-language.

Steven Long objected to the second criterion, stating that it should be
adequate for the phonology to match.  He correctly pointed out that there
are cases where accepted cognates have wildly different meanings.  This is
true.  What I'm saying, however, is that if we don't have a plausible
account for the semantic development, we shouldn't judge the words to be
cognate.  The reason is that there are other ways the situation could have
arisen.

In any etymological dictionary of even a very well-studied language, there
are many words labelled "of unknown origin".  No doubt, some of these have
cognates in other languages which we've simply missed so far; but it is
equally doubtless that some of these words were novel coinages, loans from
unattested prehistoric languages, etc.  And doubtless, there are cases
where a word existed in the proto-language and was lost in all its
daughters but this one (in which case we can't reconstruct it for the
parent language; those are the breaks). In all of these cases, we don't
know and often can't know which story is the right one.

Given that languages pick up new words, it's bound to happen now and then
that that pairs of words will chance to arise with phonological forms
which _could_ have arisen by regular sound change from some word in the
proto-language, but actually didn't.  What can tip us off to such cases is
a wild mismatch in the semantics. _If_ we have a plausible account for how
the meanings of the words could have come to mean what they do, we can
accept the words as cognate.  Bear in mind, tho, that it's better to miss
a real cognate than to include a false one, for reasons I'll go into in a
later post.

On Mon, 6 Sep 1999 X99Lynx at aol.com wrote:

> ...when it comes to what we call "semantics", every word has its
> own history...

I agree totally.

> So, really, the "meaning" of a word only makes phonological ancestry more or
> less likely, it doesn't define it.

> But the "exceptionless sound laws" support the idea that if two words are a
> phonologically the same, they have a high probability of common ancestry.

As I've pointed out, this is incorrect.  There are several other ways that
such pairs could arise.

> Unless you have the historical background to definitely eliminate the
> connection between 'river' and 'leather', your off-the-cuff impression is
> just not enough to settle the matter.

It's the other way around.  Until you have a reasonable account by which
'river' and 'leather' could both rise from some earlier meaning, you have
to hold that the words have not been demonstrated to be cognate.

> <<2) have meanings which can have plausibly developed from some meaning in
> the proto-language.>>

> And again this is backwards.  We can only guess at meanings in the
> proto-language by looking at meanings in later language.

That's why I said 'some meaning', not 'the meaning'.  If you suspect that
two words are cognate, but the meanings are substantially different, you
have to come up with some meaning for the word in the proto-language which
could have given rise to the attested meanings.

  \/ __ __    _\_     --Sean Crist  (kurisuto at unagi.cis.upenn.edu)
 ---  |  |    \ /     http://www.ling.upenn.edu/~kurisuto/
  _| ,| ,|   -----
  _| ,| ,|    [_]
   |  |  |    [_]