Alexis on Wald on Linguistic classification

Johanna Nichols johanna at uclink.berkeley.edu
Sat Feb 21 17:19:39 UTC 1998


----------------------------Original message----------------------------
Alexis Manaster Ramer writes:
 
>I could not disagree more.  The "biggie" involves one of the
>biggest myths in linguistics.  Although I hope that a paper
>with my name among several others on it will some day appear
>detailing this whole sordid mess, for now I will ask those of
>you who trust my competence in the relevant areas (i.e.,
>baby mathematics and history of linguistics) that (a) the
>vast majority of publications claiming some such limit on
>time depth contain not even a shred of the required documentation
>that such limit exists or that it is what they claim it is,
>(b) that the works that tried to do the math required to
>demonstrate such a limit (the earliest of which that I know of
>was believe it or not by Swadesh!!!!!!!!!!!!!!!!) are mathematically
>invalid     [...]
 
(a)  Documentation is not required.  Here's the argument:  Of the language
families that are both demonstrated and reconstructed (or reconstructable,
i.e. having regular correspondences and cognates), none are older than
about 6000 years.  In a very few cases the fortunate combination of a
distinctive and durable grammatical signature enables us to demonstrate
relatedness farther back than we can reconstruct; the clearest case is
Afroasiatic.  This is why I usually use language like "the diagnostic
evidence usually fades out after about 6000 years" and "we can reach back
some 6000 years and occasionally somewhat farther, perhaps to 10,000
years".
 
(b)  The idea of trying to prove this mathematically strikes me as
misguided; it's just an empirical observation.  If proven and reconstructed
(or reconstructable) families distinctly older than 6000 years start
showing up I'll change my estimate accordingly, and I assume others will
too.
 
I know that works exist in which correspondences and/or reconstructions are
proposed for families supposedly much older than 6000 years, but I haven't
seen any demonstration that the resemblances fall outside the expected
chance range.  Demonstrating relatedness means showing that the
resemblances are highly unlikely to be due to chance (or to borrowing or
universals, though all linguists know how to avoid these) and highly likely
to be due to common descent.
 
This view (which I believe is widespread) does not carry the burden of
proof.  The burden of proof is on Alexis:  if you maintain there are
genetic groupings that are much older than 6000 years, proven, and
reconstructable, please identify some and show what proves their
relatedness.
 
Johanna Nichols



More information about the Histling mailing list