RE: defaults

From: "Tim Finin" <>
Subject: RE: defaults
Date: Tue, 22 Jan 2002 12:44:39 -0500

> > From: 
> > [] On Behalf Of Peter F.
> Patel-Schneider
> > ...
> > My memory may be failing, but I don't remember any support  for
> defaults in
> > any version of Classic.  It would be possible to achieve some of the
> effect
> > of defaults by using procedural attachment in Classic, but that would
> > violate the procedural attachment contract, unless all that 
> > was done was a kind of input completion.
> I must have been thinking about the use of procedural attachment in
> Classic.  
> What was the procedural attachment contract in Classic?  I'm guessing
> that
> Any conclusions had to be monotonic.

Actually a bit more than this.

Very roughly:

	When the conditions for the procedural rule first become true, the
	rule must determine the information to be asserted and cannot
	change its answer in any way.

Two kinds of things have to be disallowed:

1/ Computing a value non-monotonically from known information, such as
   computing the number of children of all people via a rule like:

   for all people, count up the number of known children and make this be
   the value for number-of-children

2/ Computing information monotonically but depending on information not in
   the antecedant of the rule, and which can be augmented later and which
   would, further, augment the consequent of the rule, such as:

   for all people, take their children and also assert that they are their

The first violates monotonicity, the second is monotonic but to be correct
would have to be run whenever any information about an object that is a
person changes, not just when the object is first known to be a person.


Received on Tuesday, 22 January 2002 12:55:54 UTC