LINGUIST List 2.118

Saturday, 6 Apr 1991

Disc: Functionalism and MT

Editor for this issue: <>

Directory

1. Frederick Newmeyer, formal and functional
2. Guido Vanden Wyngaerd, Re: Functionalism
3. , Re: Vicki Fromkin on MT

Scott Delancey writes that my bulletin board contribution exhibits a 
deep misunderstanding of the work that functionalists are engaged 
in. But the only concrete example he gives to illustrate this lack of 
understanding is my supposed implication that functionalists 
believe 'that languages do not have structure'. I NEVER intended to 
imply such a thing and, rereading my piece, I do not see how such an 
implication could be drawn from it. Of course we all agree that 
languages have structure. The question is whether grammars are 
STRUCTURAL SYSTEMS, with their own internal principles shaping 
them. This I take to be the main issue separating generative 
grammar from the wing of functionalism that Delancey represents.

Among such principles, I have in mind, of course, Subjacency, the 
ECP, Binding, the Case Filter, and so on. Why do they exist? In fact, I 
am perfectly happy to posit a 'functional' genesis for them: I assume 
that they arose to facilitate parsing. In effect, they help keep track 
of what's what and what's where. But over time, they have become so 
thoroughly grammaticized that their relation to parsing is indirect 
at best. (I.e. many if not most ECP violations pose no particular 
parsing problems.) This grammaticization was driven by what I see 
as an innate human drive to impose structure and to maximize and 
extend structural patterning (and thereby to wrench form away from 
function). So my position hardly entails sinking into 'vitalism', I 
would submit!

In other words, I do believe that there is a level at which formal 
principles admit to functional explanation. It's just not at the level 
of synchronic grammatical analysis.

Fritz Newmeyer
University of Washington

John Aske writes:
>And since Fritz brings up parasitic gaps, why not use this as a
>case in point too. I must admit my lack of expertise in the
>matter ~[...! Given the functional/semantic similarity
>between these clauses and coordinate clauses, one is not
>surprised that zero-anaphora would come to be extended to these
>special cases. I don't see how one can start analyzing
>'parasitic gaps' from any other perspective.
I suggest you acquire some expertise on the matter before making
such sweeping statements as the above. What formal research has
revealed is that the occurrence of "null anaphora", as you call it,
is subject to certain formal constraints, eg there has to be
a "real gap" licensing the parasitic gap, the real gap must not
c-command the parasitic gap, etc. As far as the analogy with
coordination is concerned, an analysis of pg's in terms of
coordination has in fact been proposed by R Huybregts and H van
Riemsdijk in a NELS paper. The problem with it is that pg's
typically occur in clauses with subordinating conjunctions,
not coordinating ones.

Scott Delancey <DELANCEYoregon.uoregon.edu> writes:
> My understanding of the autonomous position it that it
>assumes (and I use the word advisedly) a) that the principles which
>determine linguistic structure are autonomous, and b) that this
>is because those principles reflect the structure of an innate
>linguistic capacity which is distinct from other cognitive systems,
>i.e. that language is the way it is because it is represented in
>a neurological distinct system.
Opponents of the autonomous position tend to confuse an issue of
principle with a working hypothesis. The issue of principle is that
the points under (a) and (b) are empirical questions, *not*
a priori assumption. The working
strategy is that of adopting the assumptions (a) and (b) and
see where they lead. Rather than by making comparisons with biology,
mathematics, chess, etc., the non-autonomist position would be better
served by presenting empirical evidence against the points (a) and (b),
eg by showing that, say parasitic gaps, are better analysed in terms
of a given functional/cognitive principle than a formal one. I would be
more than interested in such an explanation, but I am not aware of any.
In contrast to Scott, I believe a formal
explanation is better than no explanation.
-Guido Vanden Wyngaerd

This refers to Vicki Fromkin's historical reminiscences on
machine translation and how, if I understand her correctly,
it was all a waste of time and money for 20 years, and,
she seems to add, speech recognition is now in the same situation.

Oh dear, oh dear, where does one start to someone who has been
asleep for so looooong!? Let me just settle for MT (though I have seen 
many effective and impressive demonstrations of speech recognition in
the last few years). 

Let me put it this way: standard US Government MT programs for 
Russian-English do millions of words of MT a month and appear to
have thousands of satisfied 
customers. The EEC now has memoranda roughly translated by MT
between English and French on a daily basis and the scale of
that usage is increasing. I have just toured Japan as part of an 
NSF/DARPA team inspecting MT R&D. We visited about 20 systems, gave them
unseen texts etc., and about 6 are really pretty good. The EEC
is using the Fujitsu system for the translation of thousands
of abstracts a month.

None of this is pipe dreams, just boring technology, stamina,
money spent etc. Not much of it uses linguistics, rather more is AI 
of a sort. Sorry if this sounds like a commercial but what
can one do in the face of tired old stories and memories
combined with total unareness of what is going on out 
there?

Yorick Wilks

[End Linguist List, Vol. 2, No. 118]