Re: Contexts and quantifiers in KIF

sowa <sowa@turing.pacss.binghamton.edu>
Date: Fri, 9 Apr 93 23:21:11 EDT
From: sowa <sowa@turing.pacss.binghamton.edu>
Message-id: <9304100321.AA05070@turing.pacss.binghamton.edu>
To: cg@cs.umn.edu, interlingua@ISI.EDU, sowa@turing.pacss.binghamton.edu,
        srkb@ISI.EDU
Subject: Re:  Contexts and quantifiers in KIF
Cc: jfulton@atc.boeing.com
Jim,

There are a number of issues in your note that indicate a major
difference in how we would relate formalisms to the world, although
I don't believe it would change the way that we would use to formalisms
computationally.  Following are some comments on your comments:

> The SITUATION construct you discussed adding to your TARGET language, not your 
> METALANGUAGE, is a MODAL construct.  It is formally intensional in that it (or 
> constructs like it) resist quantification across the construct and they resist 
> replacement of identicals or equivalences.  Moreover, if the word 'situation' 
> is taken literally, it appears so similar in meaning to Kripkean "possible 
> worlds" that the differences appear to be quibbles.  

I agree with your statement up to the word "moreover".  I am using
the term "situation" in a way that is consistent with situation semantics
and the AI use in situation calculus (although the term "consistent"
may be a bit too strong, since there are many variations of both those
approaches that are not consistent among themselves or with each other).
The basic point is that a situation is limited to a specified duration
and extent of space-time, unlike the possible worlds of Montague and
Kripke, which have no bounds.  The book by Barwise & Perry (Situations
and Attitudes, MIT Press, 1983) is a bit out of date as a description
of situation semantics, but its introductory chapters are a good
statement of why they prefer finite situations to the open-ended
possible worlds as a basis for semantics.

> Secondly, the fact that Tarski restricted the scope of his original paper to 
> formalized languages is not particularly telling in this context.  The purpose 
> of his paper was to demonstrate an approach to semantics, and he succeeded 
> for a broad class of formal languages.  What he did not do in that paper was 
> demonstrate that the same approach could be applied to natural languages.  

Again, I completely agree with this statement.  But I begin to disagree
with the following:

> But he also did not say that the approach could not be applied to natural 
> languages.  Indeed applied logicians and philosophers have applied the 
> approach successfully to significant subsets of natural language.  

My only disagreement is with the word "successfully".  The primary
example of the application of Tarski-style semantics to natural
language is the work of Richard Montague and his followers.  I have
a great deal of respect for that work, but I believe that the approach
is fundamentally flawed, for many of the same kinds of reasons that
Barwise & Perry give for situation semantics.

> The work
> we have done on the SUMM shows that the approach works for that subset of 
> natural language that is captured in information systems of the kind that 
> can be captured by EXPRESS, NIAM, ER, etc.  So the issue is not whether the 
> information is initially expressed in a formalized language, but the 
> communicants can agree that a formalization captures the essence of what 
> they are communicating with sufficient accuracy.

I agree that the Tarski-style of semantics works quite well for a
broad range of formal languages, among which I would include NIAM,
ER, EXPRESS, KIF, and even the version of CGs that we are mapping
to KIF.  My reservations about its adequacy as a foundation for
natural language semantics, especially in Montague's formulation,
should not be considered as a rejection of its use for defining the
semantics of formal languages.  That's what Tarski designed it for,
and it is still the best approach available for such systems.

John