From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!caen!garbo.ucc.umass.edu!dime!yee Thu Jan 16 17:20:16 EST 1992
Article 2698 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!caen!garbo.ucc.umass.edu!dime!yee
>From: yee@envy.cs.umass.edu
Newsgroups: comp.ai.philosophy
Subject: Re: Semantics of thoughts
Keywords: formal symbol processing, syntax, semantics, Chinese Room
Message-ID: <41719@dime.cs.umass.edu>
Date: 14 Jan 92 15:58:04 GMT
Sender: news@dime.cs.umass.edu
Reply-To: yee@envy.cs.umass.edu ()
Organization: University of Massachusetts, Amherst
Lines: 71


In article <1992Jan13.023843.12181@cs.yale.edu> mcdermott-drew@CS.YALE.EDU (Drew McDermott) writes:
>                  Both semantophiles and computationalists agree that
>semantics is possible, but only the former require that semantics play
>some kind of active, continuous, functional role in the use of
>symbols.  For a computationalist, a symbol works for purely syntactic
>reasons, and a theory of its semantics is a plausible account of what
>its symbols might be taken to refer to.

and also:

>some Searlite proposed that Searle's most profound point was that
>syntax and semantics were distinct.  My modest reply is that they are
>indeed distinct, but that the point, from a computationalist
>perspective, is not profound.


As both a "semantophile" AND a computationalist I must beg to differ,
both with McDermott and with Searle.

It is virtually impossible to conceive how the formal/syntactic
processing of basic symbols can, by itself, somehow imbue the symbols
with semantics, where by "semantics" I mean Searle's simple kind
wherein the symbol HAN-BAO-BAO, by itself, leaves him cold, while the
symbol HAMBURGER, by itself, produces a visceral response.  Symbols can
be processed formally/syntactically by relying solely on objective
properties such as shape, and they can also (sometimes) be processed
semantically by relying on non-objective information that is associated
with the symbols *by the processor*.  Whether one can tell the
difference between formal and semantic symbol processing solely on the
basis of I/O behavior (i.e., the Turing test) is quite a separate
question.  It is probably impossible to distinguish between the two
types in principle; in practice, it is probably easy.

On the other hand, no computationalist can accept Searle's
(apparent/implicit?) contention that all Turing machines (TM's) are
insufficient for his kind of semantics.  The argument does not hold up.
At best, the logic of the Chinese Room applies only to programmable
computers, i.e., universal Turing machines (UTM's).  Clearly, not all
TM's are UTM's.  UTM's are "universal" only in their I/O behavior (in
fact, only a *portion* of their I/O behavior: programs are also inputs
to UTM's); they are certainly not universal in the why they process
their inputs, which is what semantophiles are concerned about.  No one
has ever shown that all TM's must process their input symbols in the
special manner of UTM's, i.e., purely formally.  Indeed, there are good
(IMHO) arguments to the contrary.

The difference between formal and semantic processing is profound: In
the former case, basic symbols cannot *represent* (literally
re-present) anything... to the processor.  In the latter case, they
can.  Given a symbol, two semantic processors must agree as to its
formal properties, but they may differ, to a greater or lesser extent,
as to its associations or content.  In semantically processing true
re-presentations (as contrasted with formal tokens), each step holds
the possibility of interpreting the basic symbols---using them to form
connections with subjective information.  This can yield inferences not
derivable solely from the intrinsic properties of the manipulated
symbols.  The point is that such interpretations and inferences are
available *within the processor itself*.  A formal symbol processor has
no such leverage with regard to its basic symbols: all additional
interpretation and inferencing, i.e., all additional *semantics*, must
lie elsewhere (e.g., in an external agent's use of "wishful mnemonics"
:-)

				Richard Yee
				(yee@cs.umass.edu)


		Richard Yee
		Computer Science Department
		Lederle Graduate Research Center


