LINGUIST List 2.102

Friday, 29 Mar 1991

Disc: Functionalism and Computational Linguistics

Editor for this issue: <>

Directory

1. Frederick Newmeyer, formal and functional approaches
2. Vicki Fromkin, MT
3. , Computational Linguists vs. 'Real' Linguists
4. , Locality

I knew that I couldn't stay out of this debate for long!

Scott Delancey writes correctly that the formalist and functionalist 
research programs are different, but implies incorrectly that they 
are incompatible. Why should they be? As I see it, the task of the 
formalist is to characterize the structural possibilities of language, 
both universal and language-particular. The task of the functionalist 
is to elucidate the principles (largely nonspecific to language) 
governing how those structures are employed in actual discourses. 
Thus each will find different data relevant to their concerns. For 
example, the formalist will show little interest in the fact that 
referential-indefinite NP's rarely occur as subjects in English 
(Givon), since our grammars ALLOW us this possibility. To the 
functionalist, the rarity of such subjects in English and non- 
existence in many languages is a fact demanding a functional 
explanation. The functionalist, on the other hand, will show little 
interest in the fact that speakers judge low text count sentences 
like 'This is the paper that I filed after reading' acceptable and 'I 
filed the paper after reading' unacceptable. To the formalist, 
however, these facts are central, since they point to knowledge 
without teaching/observing and from there to abstract structural 
principles.

I realize that there are functionalists who argue it is incorrect to 
try to characterize structures independently of their functions (eg 
Hopper's 'emergent grammar'), and I dare not hog the space in one 
message to try to rebut them. But the most frequently voiced 
argument for this position, namely that functional factors shape the 
form of grammars, is not a threat to autonomy. It's quite true, I'm 
sure, that (functional) pressure on the parser explains why in V-O 
languages heavy constituents tend to appear at the right (Hawkins). I 
suspect that Tomlin's functional explanation for why certain word 
order types predominate is also largely right. But how grammatical 
properties were shaped is quite a different matter from whether 
grammar is a formal system governed by 'internal' principles. The 
formalist position entails the characterizability of grammatical 
systems by means of an elegant set of principles and that this 
system is actually used by the speaker and hearer. I feel that the 
evidence supporting the former is overwhelming and that supporting 
the latter is very strong.

I can't resist another chess analogy. Maybe it is the case that there 
is a functional explanation for why the pieces can move as they do. 
Maybe its inventor(s) worked out the most optimal set of moves to 
make chess as satisfying as possible. Maybe there were religious or 
political motives in having bishops move diagonally. I have no idea. 
But whatever, it has no bearing on whether the layout of the board, 
the pieces, and the moves form a structural system.

One last point. Delancey writes (citing Givon) that the functionalist 
program, unlike the formalist, is subject to disconfirmation. As long 
as one can make unconstrained appeal to 'competing motivations', I 
doubt the the functionalist program is disconfirmable (Croft makes 
a similar point). If you have a functional principle to explain A and 
another to explain not-A, and can appeal to either at will, then what 
is explained? So, in what seems to me to be a particularly notorious 
example, Haiman, in a 1983 paper in Language, has two iconic 
functional principles, one which says that the linguistic distance 
between expressions corresponds to the conceptual distance 
between them, and the other which says that the linguistic 
separateness of an expression corresponds to the conceptual 
independence of the object or event which it represents. At times 
they conflict with each other. He chooses the principle that gives 
the right results for the particular example, saying that in that case 
it 'wins out' over the other. 

It's true that there is functionalist work (eg Du Bois on split 
ergativity) that recognizes this problem and tries to deal with it, 
but in my reading of the functionalist literature this happens very 
rarely indeed. As long as there are zillions of potential functional 
causes out there, and no independent means for weighting their 
relative importance, the functionalist program is on very shaky 
ground.

Fritz Newmeyer
University of Washington

TO: AMR -- I think you and I better call it quits. We obviously do not
understand each other, and seem to answer questions that are not raised
by the other party. Oh well.

TO: John Goldsmith. Of course you are right. I am not arguing against
present people trying to work on MT or even people in the past who did
but rather at the attitude displayed by many of the engineers who thought it
was all a very simple job because language was simple -- despite the
attempts on the part of linguists and others like Bar Hillel who was a
philosopher and not a card-carrying-linguist to say, look, guys, it's
a rough problem -- even beyond the difficulties of lexical ambiguity.
How many 'meanings' were assigned to TIME FLIES LIKE AN ARROW? I forget
you may remember, John. and why FRUIT FLIES LIKE A BANANA has fewer
meanings.
Despite the fact that this discussion makes it crystal clear (I was always
worried about anything which was said to be crystal clear) that Kuhn was
right regarding the difficulty of persuading someone of an opposing view
in science and that in order for one view to triumph all the antis have
to die off. I wouldn't go quite so far but it does dramatize the fact
that we cling tenaciously to our passionately held beliefs. Maybe that
is a good thing. It would be awful if we switched our scientific allegiances
each time a new argument was presented. Nothing would get done. Vicki

There have been a few disparaging remarks in this forum about the competence of
computational linguists, as opposed to so-called theoretical linguists. I
certainly want to endorse John Goldsmith's remarks about MT, a field that
has been widely misunderstood and unjustly battered. (I had occasion to visit
a DARPA official last year, and he expressed their intention, at least, to
acknowledge the suitability of MT for funding once again.) But there are two
subjects I would like to comment on: the conflicting goals between NLP
research and theoretical linguistics, and the reputation of Mr. Schank.

 1) What mainstream linguists need to understand is that computational
 linguists, for the most part, must deal with issues that relate
 to producing and interpreting language. Linguistic theory, for the
 most part, has not matured to the point where it has very interesting
 or useful things to say about these issues. More specifically, 
 linguists tend not to concern themselves with how speakers and 
 listeners RESOLVE ambiguity or handle linguistic ill-formedness.
 On the other hand, mainstream linguists seem to have a better grasp
 of what the potential ambiguities are in text or speech than do
 many computationalists.

 2) Schank's school of thought is much maligned, not just in the mainstream
 linguistic community, but in the computational community as well. I
 think that some of the criticism is deserved, but we should not let it
 dim our appreciation of his positive contributions. The overriding
 theme of his work is sound: to understand language is not to transduce
 a linguistic structure, but to relate it to a train of thought 
 (represented by "scripts", sets of "goals", etc.). It is not necessary
 for all aspects of linguistic structure to be perceived perfectly in
 order for language understanding to be carried off. (Notice that this
 claim does not apply to language production, which requires robust
 grammatical knowledge to implement.) I don't believe that Schank has
 ever claimed that *no* grammatical knowledge is needed to understand
 text. I believe that he has been a minimalist rather than a nihilist
 in this respect. In Chomsky-like fashion, he has taken a rather extreme
 position and stuck with it to see how far he could go. And, like 
 Chomsky, he has gotten more mileage out of his ideas than most of his
 critics wanted or expected. And, as with Chomsky, we may wish to chide 
 him from time to time for not stopping to ask directions along the
 way. (Formal linguists should please excuse the metaphor. :-)

 -Rick Wojcik 

David Pesetsky very kindly points out that I said the opposite
of what I meant in my last posting on the cognitive/functional/
modular/local issue. By omitting a 'not' and putting in an
inadvertent 'even', I appeared to 
say that I attribute to him (and others) the idea that people
who do not work on locality principles should not be taken
seriously as syntacticians. I did originally think that that
was his (their) view. But, of course, the point of my last
posting was to accept the reassurance that this is not the case.
The offending para should have read (and, boy, do I have egg
on my face for this typo!):

(4) I stand corrected specifically on the issue of locality. From
now, no one need to labor under the mistaken assumption (as I did
for so long) that people may NOT work on SYNTACTIC THEORY and BE
TAKEN SERIOUSLY if they do not account for locality phenomena
as their first order of business. I am so sorry about the misunderstanding.
But you don't know the relief I feel now that it has been corrected.

[End Linguist List, Vol. 2, No. 102]