Re: (biassed) summary of the argument so far.

phayes@cs.uiuc.edu
Message-id: <199306072059.AA08379@dante.cs.uiuc.edu>
Reply-To: cg@cs.umn.edu
Date: Mon, 7 Jun 1993 16:03:37 +0000
To: interlingua@ISI.EDU, sowa@turing.pacss.binghamton.edu,
        sowa <sowa@turing.pacss.binghamton.edu>
From: phayes@cs.uiuc.edu
X-Sender: phayes@dante.cs.uiuc.edu
Subject: Re:  (biassed) summary of the argument so far.
Cc: cg@cs.umn.edu, cmenzel@kbssun1.tamu.edu

Hi John

We seem now to agree on most substantial points. A few last comments to 
clarify my position, then no more on this from me to Interlingua (sighs 
of relief subtly change the climate over the USA); then a more pragmatic 
comment about KIF syntax.

>Certainly for many applications, like counting apples and identifying
>sets of people, dogs, and houses, the application is almost trivial.
>But it is very far from trivial when you are working in a new domain,
>especially one where you can't use unaided eyesight to recognize the
>things you choose to call "individuals". And when you trying to get a
>computer to deal with those apples or people, the things that are trivial
>for humans become some of the most difficult to simulate.

Just to reiterate: the application of set theory to domains does not, 
in my view, involve questions of recognition. Of course you are right 
that visual recognition is a difficult problem: but irrelevant.
>
>
>..... there are also very
>fundamental differences between different kinds of "representations"
>inside a computer system, either an AI system or a commercial DB or
>an engineering simulation of some physical system.

I agree, and also that the use of the blanket term 'Krep' ignores 
these distinctions. However, while agreeing that this whole issue 
of relationships between different kinds of representation is 
important and fascinating, I also believe that an adequate semantic 
theory will need to somehow account for how information gets 
converted from one to another, and therefore needs to be able to 
talk about the content of various forms of representation. Without 
this, the question of whether an internal transformation (or 
relationship) preserves content is meaningless. I also, by the way, 
have a longstanding belief that 'new' representational ideas are 
often just old ones in new computational guise, and make it my 
business to burst such bubbles. (I do this not to defend old 
ways, but because I want to get past the old ways, and need real 
changes not just notational variations.) The frequently cited idea of 
'image-like' as opposed to 'language-like' representations is often 
such a bubble, in my view: see a recent issue of 'Computational 
Intelligence' for an extended discussion of this.

>> This is easy to distinguish. What you mean by 'perceived reality' 
>> is something like 'the way the world would (or might) be if my 
>> perceptions were correct', I presume? That is a model of my 
>> (perception-governed) representation of it.
>
>Thank you.  That "model of your (perception-governed) representation"
>is exactly what I meant by "depiction"

But then there is no need for this term. If you are simply talking about a 
model(-theoretic model), use the established vocabulary. "Depiction" 
also has connotations of being image-like, of being computationally 
useful, of having been invented by J.Sowa (another use-mention confusion :-), 
and so forth.

> But what I was
>trying to say is that many of the mathematical operations in model
>theory can be implemented in a digital computer.  In fact, I was
>pointing out that the operations that an SQL processor uses to
>answer a query are exactly equivalent to the operations that Tarski
>defined for evaluating the denotation of a formula in terms of a model.
>

Thats an interesting observation, one of course I don't disagree 
with. (It's always a mistake to disagree with methematics.) Indeed, 
it suggests an alternative way of thinking about a 
database which could be useful. However, it has the problem (compared 
to the view of a database as a vivid representation, ie as encoding 
ground facts) that it seems not to capture the intended content, so 
>From my perspective it seems not to adequately specify what such a 
representation means.

........
>As I said many times, I was not trying to stop anyone
>from saying anything they please in either CGs or KIF. 

Not IN Kif, but you did seem to want to stop people saying some 
things ABOUT it, which is where I came in.
....

>All this metalevel discussion is about my approach to constructing a
>framework that has some useful computational properties.  We have been
>arguing so long about philosophical issues that I have hardly begun
>to list the computational reasons why I prefer this approach.  But
>nothing forces you or anyone else to adopt my approach if you don't
>like it.  

Of course I have no objection to you developing your approach, 
together with whatever theoretical account you find congenial. My 
understanding was that this argument was provoked and fuelled by 
questions about the semantics of a 'standard'.

---------

>Now, to get back to KIF, I would like to endorse your recommendation
>that we avoid the extra levels of parentheses in the variable and type
>declarations; i.e., for the sentence, "Yojo is one a mat", there were
>two suggestions:
>
> 1.  (exists ((?x mat)) (on Yojo ?x))
>
> 2.  (exists (?x mat) (on Yojo ?x))
>
>You preferred the second alternative, since the question mark is
>sufficient to distinguish the variable from the type declaration.
>I agree wholeheartedly.

I should say that Mike Gensereth indicated that there is a more 
substantial reason for the inadequacy of the single-nest syntax. 
He wants to allow one bound variable to  vary over the range of 
an inner variable, so one can write soeming that means something 
like 'for all x, whatever its sort is,...", as in for example 
(using the double-bracket convention):

(forall ((?x set1) (?y set2)) (forall ((?z ?x)) ...))

Here the final occurrence of ?x is bound by the first forall, 
not the second: compare it with 

(forall ((?x set1) (?y set2)) (forall (?z ?x) ...))

where the first ?x is vacuous.

I confess that this seems a fatal objection to the simple syntax, 
and the double-bracket convention might well be the simplest 
alternative.

Pat 

----------------------------------------------------------------------------
Beckman Institute                                    (217)244 1616 office
405 North Mathews Avenue        	   (217)328 3947 or (415)855 9043 home
Urbana, IL. 61801                                    (217)244 8371 fax  	  

hayes@cs.stanford.edu  or Phayes@cs.uiuc.edu