concerning the algorithmizability of historical linguistics

[Steven Schaufele]

----------------------------Original message----------------------------
H.M.Hubey wrote:
 
> I also have very small doubts that I can write a program that can take
> n words from language A, and m from language B and write a program that
> can change one set into the other (at least enough of them to dumbfound
> the skeptics and force them to having N new looks into the "comparative
> method") using only regular sound changes. The only problem is that I
> don't have the time or energy to put into such a useless demonstration.
> It will probably be done by someone (like Starostin) or someone else
> who is upset at the way linguists spurn statistics and math. After all,
> it the only thing that will make people sit up and take notice there is
> no better way than to demolish their toy :-)
 
This is precisely where i feel we are taking leave of scientific
responsibility.  If you can write a program that can take n words --
*any* n words -- from language A -- *any* A -- and m words -- *any* m
words -- from language B -- *any* B -- and change one set into the
other, how is such a result falsifiable?  Such a program sounds to me
like one that is guaranteed to find evidence of affiliation no matter
what data are fed to it.  And i'm quite sure that is not desirable.
Should we ever discover extraterrestrial, or otherwise non-human,
language, such an approach as Mark seems to be suggesting here would
surely be able to `prove' that such a language is nevertheless
affiliated somehow with any given human language.  Which would
definitely be a major step in the wrong direction, since at least IMHO
the discovery of such a language ought precisely to provide us with an
opportunity to get beyond studying merely human language (presumably a
manifestation of species-specific characteristics of the human brain,
etc.) and onto the path of examining the nature of Language itself in
the abstract.
 
I'm concerned that Mark's approach to developing AI programs for
historical linguistics might also lend itself to developing programs
that would quite happily `demonstrate' that, e.g., fish, ichthyosaurs,
and whales are all more closely related to each other than any of them
are to salamanders, turtles, or cows, or that dragonflies, eagles, and
bats are more closely related to each other than any of them are to
spiders, penguins, or shrews.
 
As, e.g., Larry and Miguel have noted, AI systems can be extremely
helpful in doing a lot of the comparative work our science is based on,
and perhaps even draw our attention to relationships and affiliations we
might not otherwise have considered.  But the results of such analyses
still need to be subjected to expert peer review.  Maybe a certain
amount of that sort of thing can be automated too, but i very much doubt
that all of it can be, at least in the rather elementary, simple-minded
manner of reducing it to an algorithm.  There's a lot of judgment
involved, which judgment needs to be based on a lot of experience which
is definitely not algorithmizable.  To what extent that kind of
experience can be taught to a computer, or a computer programmed to
learn it as a living, flesh-and-blood historical linguist does, i
confess i have no idea.
 
I am also bothered, as is Larry, by Mark's continued (apparent) tendency
to equate scientific research in a field such as (historical and
comparative) linguistics with artistic endeavours such as painting
pictures, composing music, or playing chess.  A freshly-painted picture,
a new piece of music, a game of chess is a new creation and derives some
of whatever value it has from that novelty; the same is true, of course,
of a freshly-crafted, previously-unused and -unheard-of sentence.  Or a
mathematical theorem, for that matter.  A reconstructed protoform, on
the other hand, is a hypothesis concerning something (a linguistic
expression) that is supposed to have really existed in the objective
universe at some point in the past.  Its validity ultimately stands or
falls on how closely it approximates objective, historical reality, not
on its solipsistic elegance.
 
Best,
Steven
--
Steven Schaufele, Ph.D., Asst. Prof. of Linguistics, English Department
 
Soochow University, Waishuanghsi Campus, Taipei 11102, Taiwan, ROC
 
(886)(02)2881-9471 ext. 6504     fcosw5 at mail.scu.edu.tw
 
http://www.prairienet.org/~fcosws/homepage.html
 
 
 
        ***O syntagmata linguarum liberemini humanarum!***
 
        ***Nihil vestris privari nisi obicibus potestis!***