Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: SUO: RE: RE: RE: Comment#1 'Foundations'




> > >
> > > 4. Ontology Parts
> > >     - A series of parts, each probably quite small
> > >       (smaller than the Ian Niles document) that
> > >       can be used in groups together for particular
> > >       purposes.
> >
> > Actually, the "Merged Ontology" is comprised of nine relatively
> > self-contained sections, viz. "Structural Ontology", "General
> > Classes and
> > Axioms", "Agent Hierarchy", "Number Hierarchy", "Units of Measure",
> > "Organism Hierarchy", "Temporal Definitions/Axioms", "Mereotopological
> > Definitions/Axioms", and "Positional Relations".  Are you
> > saying that we
> > need a more fine-grained division that this?
>
>MW: This is a problem (I think) with KIF first of all.
>
>What I am saying is that it is already clear that a monolithic ontology will
>become unmanageable, if it hasn't already become so. I think we need what I
>think John has described as a Lattice of Theorems, or at least my
>interpretation of what he meant. That is that each theorem should be managed
>as a separate piece of KIF - this give manageability. But we also need to
>know as part of the theorem, which other theorems it is dependent on (I
>think we can allow dependence to be inherited).

Im not quite clear what you mean by 'dependent on' here. If you mean 
'could resolve with and produce a conclusion', then almost everything 
is going to depend on almost everything else.

>This implies that we need
>constructs in KIF to support this. So we need to be able to define a
>collection of KIF statements as a theory, and be able to give in a theory
>the list of other theories it is dependent on.
>
>You do not have any of this structure even informally, so no user would know
>which bits of the SUO he needs if say he is just interested in holes.
>
>If this can already be done with KIF as it is, I would appreciate an
>explaination.

KIF does not of itself provide this structure, but software could be 
written to extract it automatically. Cycorp have extensive tools to 
allow the user to 'browse' through a large axiomatic Kbase, including 
such facilities as asking for a concept related to an existing 
concept, view and navigate through the (highly complex) class 
heirarchy, and see all axioms which mention a concept or group of 
concepts. It takes some getting used to (its considerably more 
complex than hyperlinks), but one can do most of what one wants to do 
in this kind of framework. I honestly don't think we can expect to do 
much better, and to develop usable tools like this is not a job to be 
undertaken lightly: it is many man-years of effort. No large ontology 
is going to fall apart neatly into separate theories which can be 
taken or left in isolation. The SUO isnt going to have 'bits', its 
going to be more like a kind of large network, with some areas more 
tangled than others.

Pat Hayes

---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes