Informal is fine (was Re: AW: Increasing interest in the HPSG conference)

[Luis Casillas]

I have to side with John Nerbonne's comments, but I would (and will)
dare to say even more: most linguists, if they could sociologically
get get away without using ANY "formalism", would do so in favor of
something like Dixon's "Basic Linguistic Theory". In fact, a lot of
actual theoretical work in syntax already goes on in this level;
linguists write papers which "describe" something in terms of, say,
"argument linking", make a bunch of arguments, and then in the end
provide the "theory" in terms of a "formalism".

Maybe I've spent too much time around functionalists, cognitive
linguists, field linguists and the likes (or Arnold Zwicky, for that
matter), but I would say that (a) this is a really misguided view of
what's "theory" and what's "description" in linguistics, (b) that the
formalism-laden "theory" that linguists feel like they have to include
in the paper very seldom adds anything of value, and is just there
because of the dictates of convention (as enforced by, say, conference
abstract reviewers, or student advisors).  It is way too often the case
that an otherwise perfectly fine paper is defiled by other people's
demand that it use the formalism du jour, and the author ends up spending
two sleepless weeks doing a half-assed formalization of their theory
(which their torturers call "description") that adds NOTHING to the actual
paper.

GB/MP is informal quite simply because (a) it is dominant, (b) most
interesting stuff to be said about syntax doesn't require heavyweight
mathematical formalisms to say it, so (c) despite the mantra of "precise
formalisms allow you to test your theory's claims very precisely", most
linguists have very little to gain from them.  There is very little
that's precise about predictions such as "sentence X" is grammatical,
given the fact that everybody who makes them accepts that "grammatical"
doesn't mean "acceptable": linguistic theory provides for a number
of factors, for example performance, that override the predictions
of the theory.  And what counts as a performance factor among linguists,
in practice, is not at all a matter of precise theory, but rather,
something that's negotiated with their peers (that's the fancy social
science way of saying "what their peers will let them get away with").

Having said that, I would add: the major problem with GB/MP is not
that it is informal; rather, its problem is in part that it is TOO
formal.  All those trees, mathematical sounding language, and its
history (its roots in formal language theory) give the impression of a
mathematically precise character that's just not there, and as I argue,
that contributes very little to most linguists' work.  To say yet another
provocative thing, which I may not believe 100% but which I'll say
anyway: most linguists, because of the hegemonical status of natural
science as a judge of knowledge in our society, want to believe that
what they're doing is a form of natural science (like cognitive
psychology), and not of social science (like cultural anthropology);
thus, a lot of the formalism is really there to support a bid for the
prestige of mathematics and natural science in general.  (The short and
sweet term for this is "physics envy".)

To be fair, I should mention one sort of use case where the mathematical
precision of HPSG does in fact buy you something: you can write software
that enables you to implement and develop grammars, parse standardized
corpora with them, and thus get numeric measures of how one of these
grammars compares to the next.  Is this good?  Yes.  Should more people
do it?  I'm ready to agree.  Should this be the privileged way of doing
linguistic theory?  Not so fast.

I now brace for the flames ;).

--
Luis Casillas <casillas at stanford.edu>
Proud postmodern hack since, um, 2003 or so

"If an airplane can fly, why can't a computer think?
But if a submarine doesn't swim, how could a computer think?"
-- me