[Corpora-List] Suggested Track for Studying Computational Linguistics

[John F. Sowa]

Bogdan,

I would generalize that point:

BB> Your example about Barbara Partee reminded me
 > what it is frequently said:
 >
 > -- linguists often become the best programmers and
 > computer scientists, given the examples of Larry Wall
 > (the inventor of Perl) and Noam Chomsky (for his
 > contribution to formal language theory).

It is also said that musicians and chess players
become good programmers -- and probably for similar
reasons:  linguistics, music, chess, and mathematics
depend on recognizing very complex structures.

But the cause and effect relations are not clear:
Is it nurture or nature?  Does the intense discipline
required to master one of those fields carry over
to the others?  Or do they all depend on a similar
kind of native talent?  Or does the study of more
than one type of pattern help develop more general
ways of looking at and analyzing patterns?  Or is
it some combination of all of the above?

At the present stage of neural science, the best we
can say is "Nobody knows".  Therefore, people who study
such topics resort to statistics to get approximate
answers, even though they have no deeper model.

That is also what I would reply to Mark Line's comment
that statistical NLP is not linguistics:  When you have
a complex problem in an area that is not well understood,
statistics can help discover quantitative relationships,
but, by itself, it cannot provide a detailed explanation.

One major difference between the fields of NLP and
theoretical linguistics is that a linguist can pick and
choose what problem to analyze and write about.  But
an NLP system has to deal with whatever language anyone
throws at it.  Therefore, some use of statistics can help
cover areas for which no satisfactory models exist.

John Sowa