Re: Good and Bad IS-A hierarchies (Peter Clark)
Message-id: <>
From: (Peter Clark)
Date: Mon, 24 Jul 1995 12:39:56 -0500
X-Mailer: Mail User's Shell (7.2.5 10/14/92)
Subject: Re: Good and Bad IS-A hierarchies
Precedence: bulk

> [Pat Hayes wrote]
> The point is that all this fussing over is-a heirarchies and getting
> vocabularies straight really misses the core issue, which is locating the
> key conceptual clusters and trying to get enough of the content axiomatized
> in a useful way. Once this is done, it is easy to be flexible about
> particular vocabularies.

Exactly! But...

> There is no way to organize the concepts, or even the axioms, into neat 
> little packets so that the various alternatives can be assembled by 
> choosing some and ignoring others. There just are genuine alternatives, 
> and one has to make committments in selecting a temporal theory to work with.

Well, the above paragraph suggests that we simply have N different 
representations and we pick one. It ignores the idea of being able to 
compose representations from components (I'm not sure if that was 
intensional) given a particular ontological commitment.
While composition isn't a trivial notion of pick any 
vocabulary you want, it's not impossible either as demonstrated by
AI research in compositional modelling, and software engineering
research in software reuse. What would such components look like? 
Maybe a collection of pick-and-choose axioms which share the same
underlying ontology. Some formulations of use of "contexts" and "views"
seem to embody this kind of compositionality.

For example, we might want to reason about objects moving in/out of 
containers. Different tasks might require us to model different aspects
of the container: eg. whether it has a portal, its volume, the portal's
size, whether or not the portal is locked. We could write down axioms
about containers:
   1. To move in/out of a container, the container must have a portal.
   2. Moving in/out of a container involves moving through a portal.
   3. To move an object out of a container,
           - it must be in the container to start with
           - after the move, it will no longer be in the container
   4. If you move an object into a container, it will be in the container
           as a result.
   5. To open/close a portal, the portal must be unlocked.
   6. To move an object through a portal, the object must be smaller than
           the portal.
   7. For an agent to lock/unlock a portal, the agent must have the 
      portal's key.
   8. ...
Now we do have more of a pick-and-choose representation -- I can select
some axioms and ignore others, depending on which aspects of the
world I care about and which I want to consider irrelevant. This
is a more constrained notion of composition: all the axioms must share
share the same vocabulary or ontology, ie. they are all in the same
"family" (context?), and may be inconsistent with other families. 
But it does still allow some notion of pick-and-choose.

So while you do have to make ontological commitments, you could still have
pick-and-choose within a commitment. So Ed Hovy's (and the rest of AI's) 
desire for pick-and-choose component libraries can still be realized.
[I'm not sure if you meant otherwise: I'm reacting to the suggestion
 there are 4 (say) alternative temporal theories -- whereas 
 I'd prefer to say there are 4 families of temporal theories (based on 4
 different ontological commitments) but each family contains a large space
 of different theories which can be constructed from components.]

	Best wishes,


Peter Clark (   Department of Computer Science
tel: (512) 471-9565                  University of Texas at Austin
fax: (512) 471-8885                  Austin, Texas, 78712, USA.
Project homepage: