Re: Good and Bad IS-A hierarchies

Peter Clark <pclark@cs.utexas.edu>
From: Peter Clark <pclark@cs.utexas.edu>
Message-id: <199507242000.PAA12906@firewheel.cs.utexas.edu>
Subject: Re: Good and Bad IS-A hierarchies
To: lenat@cyc.com (Doug Lenat)
Date: Mon, 24 Jul 1995 15:00:30 -0500 (CDT)
Cc: pclark@cs.utexas.edu, phayes@cs.uiuc.edu, srkb@cs.umbc.edu, cg@cs.umn.edu
In-reply-to: <19950724183032.6.DOUG@SHANDRA.MCC.COM> from "Doug Lenat" at Jul 24, 95 01:37:09 pm
X-Mailer: ELM [version 2.4 PL24alpha3]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Content-Length: 2023      
Sender: owner-srkb@cs.umbc.edu
Precedence: bulk


> I think some generalization of what you're saying will turn out to be
> the right (i.e., adequate!) answer.  Namely, one will pick and choose
> the ontological commitments as well, and then pick and choose from among
> the axioms which are "allowed", and each "pick" will then further
> eliminate some axioms from consideration.  
>
> An example of the former sort of ontological choice is "size" -- as
> in your axiom number 6.  An example of the latter is the adoption of
> your axiom 1 which then precludes osmosis etc.

Yes, that sound's right; I like that incremental notion more than 
the idea of fixed partitions. 

True too: axiom 1 (containers need portals) makes an ontological commitment 
about containers which precludes osmosis. We've been looking a bit at this:
we can "localize" this commitment by introducing a specialized container, 
("portaled-container", say), and then asserting this axiom just for that 
specialization, rather than all containers. Now we can introduce osmosis into
our representation somewhere else without conflict. In fact -- an "isa" 
hierarchy falls out naturally by looking at what things there are in the 
world, identifying which axioms are potentially applicable, and then looking 
for subset/superset relationships between those axiom sets. The (general) 
axioms are the building blocks, and the taxonomy reflects the things which 
you can build with them, composed out of those building blocks but using
specialized names (eg. "container" -> "portaled-container") to localize the
ontological commitment. (If anyone's interested we've a rather half-baked 
working note at http://www.cs.utexas.edu/users/mfkb/notes/004.ps with a bit 
more on composing action descriptions from components).

		Best wishes,

			Pete

----
Peter Clark (pclark@cs.utexas.edu)   Department of Computer Science
tel: (512) 471-9565                  University of Texas at Austin
fax: (512) 471-8885                  Austin, Texas, 78712, USA.
Project homepage: http://www.cs.utexas.edu/users/mfkb