CG: Re: A simplistic definition of "ontology"

John F. Sowa (
Wed, 15 Oct 1997 04:37:11 -0400

Some comments on the note from Mike Uschold:

>2. there is a deeper distinction than logic vs language:
> --the extent to which meaning is specified for terms--

Yes, I agree wholeheartedly. In fact, this point is so central that
I think the issue of "logic vs language" is irrelevant. If you specify
meaning properly, it should be equally useful for both KR and NL systems.

>To me, the overall goals of the two 'camps' are not extremely similar.

That is because both "camps" are looking at narrow issues that are
obscuring the deeper issue of how meaning is specified.

>The logicists specify meaning via axioms, and generally are unconcerned with
>word meaning of a grammatical and morphological nature.

Grammar and morphology are part of NL syntax. The NL semantic issues
are very directly involved with axioms and definitions in the same sense
as any KR system. And by the way, I prefer to avoid the term "logicist"
because it tends to divide the KR camp into those who use logic and those
who don't. When you get to the deeper issues, I believe that the issues
of meaning are identical for the NL people, the logic-based KR people,
and the "scruffies" who are using logic, but think that they aren't.

This point may seem shocking to old timers in AI, but it is the working
hypothesis of the people who have been going to the conceptual graph
workshops and ICCS conferences for the past 12 years. CGs are a system
of logic that has a direct translation to and from KIF and predicate calculus,
but the people at these conferences have been using CGs for NL semantics,
expert systems, learning, theorem proving, planning, problem solving,
product specifications, etc. Some people have a very scruffy Schankian
style of using CGs, and others have a very neat logic-oriented style.
But they're all using the same notation with the same kinds of axioms and
definitions -- any or all of which could be translated to KIF.

I am not implying that people have to use CGs in order to get the benefit
of knowledge sharing between the different groups, because CGs can be
translated to KIF and other notations. But the CG notation has enabled
people who grew up with a strong logic-oriented education to cooperate
closely with people who like the notation, even though they never thought
they liked logic. Perhaps there are other notations that might be able
to promote the same or similar kind of cooperation, but CGs are one example
of a notation that has succeeded in getting these differing groups to talk
with one another -- instead of talking past one another.

>There are wide differences in the nature and extent to which meaning of terms
>is specified for these different purposes. Although there are some
>similarities in the goals of these efforts, I am not so sure that they are
>similar enough that we can clearly define what we mean by a successful
>cooperation interms of specific benefits that all camps could enjoy.

I agree that there are wide differences in the amount of semantics that
various groups have been using for different purposes. But when any one
of these groups expands their range of interests, they get into the deeper
issues that are identical for all of them. There is much more that could
be said about this topic, and perhaps we can get into further discussions
at the T2 meeting in November.

John Sowa