Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!wuarchive!udel!princeton!phoenix!eliot
From: eliot@phoenix.Princeton.EDU (Eliot Handelman)
Newsgroups: comp.ai
Subject: Re: AI & Derrida (was: Re: Speech Act Interpretation:... (Unisys AI Seminar))
Message-ID: <10791@phoenix.Princeton.EDU>
Date: 10 Oct 89 23:27:47 GMT
References: <11627@burdvax.PRC.Unisys.COM> <10714@phoenix.Princeton.EDU> <10744@phoenix.Princeton.EDU> <11577@eerie.acsu.Buffalo.EDU> <10780@phoenix.Princeton.EDU> <11597@eerie.acsu.Buffalo.EDU>
Reply-To: eliot@phoenix.Princeton.EDU (Eliot Handelman)
Organization: Princeton University, NJ
Lines: 69

In article <11597@eerie.acsu.Buffalo.EDU> lammens@sybil.cs.Buffalo.EDU.UUCP (Joe Lammens) writes:
;In article <10780@phoenix.Princeton.EDU> eliot@phoenix.Princeton.EDU (Eliot Handelman) writes:
;>In article <11577@eerie.acsu.Buffalo.EDU> lammens@sybil.cs.Buffalo.EDU.UUCP (Joe Lammens) writes:
;>
;>;Even the people who are going on and on about not being able to
;>;deduce the writer's intentions from a text, how a text is a world in
;>;itself with a million meanings and how it is full of contradictions,
;>;do not apply their own `theories' to their own writings on the
;>;subject, since that would imply that their writings are meaningless or
;>;at least not usable as vehicles for talking about their beloved
;>;subject.

;>As a matter of fact you're wrong, which is why deconstruction offers no
;>reading method, at least one that can be couched in propositional terms. 
;>The implications of this idea would, of course, be disasterous for
;>those factions of AI that do feel that the discernment of meaning and
;>intention is subject to rule decomposition, if it were to be taken
;>seriously. Eventually, I strongly suspect, it will be. 


;Now how can I be wrong in stating something like that, given that your
;own paradigm states that you cannot infer/abduct (nice lingo :-)) any
;singular meaning from what I wrote, let alone what I intended to
;write? 

Look, "deconstructing the deconstructionists" is as old as the hills. The
trouble is that it predates the deconstructionists and so is already a part
of their discourse. It's like trying to defraud Freud by subjecting his
writings to freudian analysis. You can do it, but what does it show? Either
an awareness or a lack of awareness that the text was subject to decomposition
in this way. Deconstructionist texts are partly informed -- or, as a 
deconstructionist would say, "constructed" -- by this awareness, and that's 
what makes them so complicated. To deconstruct a text is not to show that
a text has contradictions and is therefore meaningless. What a 
deconstructionist tries to do is to restructure the priorities of a text by 
analyzing the contradictory "awarenesses" and intentions which participate
in its construction. A simple priority, for example, is that "a text exists 
in order to convey the intention of its author." This model is good enough
for Unix shell commands. When I say "ls" my immediate intention is to make
"ls" do whatever it's defined as doing. But the model is not good enough for
natural languages, because "conveying intentions" or "understanding" is not 
necessarily a priority of natural language. The kind of language that is
described by these priorities is really a command language, as when, for
example, the Feldwebel says "Raus!" and the POW obeys. It isn't really
natural language at all. It's a kind of glorified Unix shell language.
Natural language is partly constructed through an awareness that the recipient
of the message WILL interpret and restructure the text according to his
particular priorities. Natural language does not assume that the relationship
of speaker to recipient is that of master to slave. And whatever assumptions
are made of the recipient are partly determinants of the shape of natural
language. In order to understand language, in the sense of standing back
and trying to see how it works, it needs to be deconstructed. You need to
ascertain the nature of the awarenesses, multiple and conflicting, that
are constructed by the fact, at the very least, that its purposes are intended
to act within the world of human discourse.

And that's why I think that machine understanding of natural language has
failed, and will continue to fail. This isn't to deny that machines can't
be instructed to parse a limited subset of English, defined over a finite
and knowable set of intentions. But this is no longer natural language. This
language makes assumptions that are never made in natural language, one of
which is that "I happen to be talking to a machine which has no real 
understanding." And I don't see the point, except to sell business applications,
of pretending that anything significant has been accomplished by writing
programs that are able to sustain the illusion that understanding has 
taken place for all of three minutes. If an artificer of intelligence is
seriously interested in gaining some basic understanding of what it is
he's supposed to be doing, I think that these expectations and "research
goals" ought to be jettisoned, or, at least, laughed at. They are naive.