Doing historical linguistics (part 3)

[Larry Trask]

----------------------------Original message----------------------------
On Sat, 7 Nov 1998, H.M.Hubey wrote:
 
[LT]
 
> > OK.  Here's a problem from my field.  The four major regional variants
> > of the Basque word for `ear' are as follows:
> >
> >         <beharri>
> >         <begarri>
> >         <belarri>
> >         <biarri>
> >
> > So: what's the proto-form?  And what "simple rule" should be invoked to
> > discover it?
 
> As for the "regular sound change rule" as applied to Basque ear,
> it is plainly possible to see that the algorithm is a "partial
> Caesar cipher". A Caesar cipher (the easiest cipher to crack) consists
> of changing every letter (sound) to another one by shifting them
> by an integer n mod M. In the case of linguistics it is "partial"
> because only some sounds get changed.
 
> In the case above if we make the equivalence l=h=g=* we can write the
> set
> as  {be*arri,biarri}. Now we equate *=0=# and e=i=@ and obtain b@#rri,
> where @=ei and #=l=h=g=0.
 
> Now we go back to the original problem of language universals, and
> whether h>g is more common than g > h etc. This is why I asked the
> question of how protoforms are constructed in the first place.
 
> How, if it is not clear which is more common (e.g. h>g or g>h) then
> do linguists (i) choose one, AND, (ii) at the same time claim that
> it is a science.
 
Neither h > g nor g > h strikes me as a very plausible change, at least
not as a single step.  If we allow these to be multi-step changes, then
I would expect g > h to be more frequent than the reverse.
 
However, none of this is relevant, because your partial Caesar cipher is
of no obvious relevance to my question.  Not one of the four attested
forms is a plausible ancestor for the other three, and the original form
must have been different from all of them.  In fact, our preferred
reconstruction is *<berarri>, which accounts perfectly for all four of
the recorded variants.  But, of course, the probable correctness of this
reconstruction is only obvious if you are intimately acquainted with the
facts of Basque.  Without such knowledge, *<berarri> looks no more
plausible than, say, *<bedarri> or *<befarri> -- both of which are quite
impossible.
 
> I will refrain from telling you what my experience with (some)
> linguists is :-) However I could easily tell everyone what I
> know of your competencies as displayed already on other lists, but
> I will also let that pass and let you demonstrate it on your own
> the same way you demonstrated it on other lists (re: context-free
> grammars vs contex-sensitive grammars, definition of language, etc).
 
Oooooh, Mr. Hubey.
 
[Note of explanation: Some time ago, on another list, Mark asked me to
explain what a context-free grammar was.  I did my best, but apparently
I was unsuccessful.]
 
> All you have to do is answer the question which you have dodged for
> over a year. I will repeat it:
>
> How, if it is not clear which is more common (e.g. h>g or g>h) then
> do linguists (i) choose one, AND, (ii) at the same time claim that
> it is a science.
 
Mr. Hubey, I'm afraid this question is unintelligible.  It's rather like
asking "What is the commonest word for `dog' on the planet?" and
expecting the answer to be of some relevance to doing historical
linguistics.
 
Anyway, this business of "choosing" between h > g and g > h is something
you have invented for yourself.  We don't do historical linguistics by
choosing from a Chinese-restaurant menu.
 
[LT]
 
> > Mr. Hubey, are you suggesting that one need not spend years studying
> > Sumerian in order to know Sumerian?
 
> In order to produce a list of cognates all you need is a dictionary.
 
No, Mr. Hubey, a thousand times no.  This is a common misconception
among non-linguists, but it is grossly false.  To produce a list of
cognates you need far more than a dictionary.  To start with, you need
to know what you're doing.  Trying to compile lists of cognates with
nothing but dictionaries -- a common but forlorn activity -- is about as
sensible as trying to construct a theory of music by counting the
occurrences of all the notes in the entire repertoire of operas.
 
 
Larry Trask
COGS
University of Sussex
Brighton BN1 9QH
UK
 
larryt at cogs.susx.ac.uk