Re: CG: Re: A simplistic definition of "ontology"

John F. Sowa (
Wed, 15 Oct 1997 19:40:41 -0400

Further comments on Mike Uschold's comments on my comments:

>I'm not sure that many or any of the various groups WANT to expand their
>interests, precisely because focus and progress in their area is what counts
>for the most. This is my whole point. Unless there are clear advantages for
>more than one group to expand their interests and cooperate to address the
>deeper common problems -- there is little motivation to do so.

I would agree that many people see no need to do anything other than
continuing the kinds of things they have been doing and feel comfortable
with. But there are many others who are frustrated by their inability to do
new things that cannot be done without much better developed ontologies.

In particular, I have been actively working in two such areas: NL processing,
which requires much better ontologies that are well supported with formal
definitions and axioms that enable NL systems to go beyond syntax to semantics;
and the work on the ANSI and ISO CSMF (Conceptual Schema Modeling Facilities)
project, whose aim is to develop richer KR systems to support databases,
software design, requirements analysis and specification, and other similar
projects whose economic impact can be enormous.

>We had an `upper level' enterprise ontology, but when it came to specific
>applications which required the ontology to be extended, there was a big gap,
>which was bridged in a very minor way....

Yes, that tends to happen. I don't believe that you can develop ontologies
in isolation from major applications. But at the same time, I don't think
that you can develop good ontologies just by grubbing away at the low
levels of the hierarchy. There has to be constant interaction between
the high-level theory and the low-level empirical data mining. It is
an iterative process that has to be approached top-down, bottom-up, and
middle-out with continual interaction and adjustments at all levels.

The Cyc project is the largest example of a major effort that has done
good, if not perfect work at all levels. I have a great deal of admiration
for what they have accomplished, although I still am not completely satisfied
with all of the design decisions that they have made so far. I must admit,
however, that I don't fully understand why many of those design decisions
were made, and I am very frustrated by the lack of published papers that
explain it. Their only major document, the 1990 book by Lenat & Guha, is
already 8 years out of date, and every level of their ontology has undergone
major revisions. I and many other people would love to know why they made
the revisions that they did, what was wrong with their previous choices,
and what kinds of directions they would like to take in the future.

I have spoken with Lenat and his colleagues and complained about that
lack of documentation. Their answer is that they have a small group
that is paid by their customers to implement working applications and
their customers have no desire to pay for research publications. Furthermore,
the customers consider their applications proprietary and would rather not
have the details published.

I think that we have won a minor victory in getting the top several thousand
of the Cyc categories made available on the Internet ( But
that is without a detailed rationale for the design choices and without
the detailed axioms that accompany those categories. Scientific progress
cannot be made without much more open analysis, debate, and testing of
extensions and alternatives.

John Sowa