Date: Tue, 26 Mar 1996 18:48 CST
From: Jim Nell <nell@cme.nist.gov> (by way of Jim Fulton <jfulton@redwood.rt.cs.boeing.com>)
Part 1: Call for Participation in Joint Workshop on
Standards for the Use of Models that Define the Data and
Processes of Information Systems
[...]
The root of the trouble (with the incompatible standards mentioned in
the Call) is not-expressive-enough languages and notations. That's why
STEP couldn't use IDEF, etc. The Basic Semantic Repository (BSR) is now
limited to modification-chains. Each time somebody makes a minor
incremental addition to a language, but fails to go for universal
expressiveness (at least that of natural language, logic and mathematics
combined), it is a step forward in progress. But then there are so many
steps left on the road ahead. To skip over these incremental
improvements (and the pattern of incompatibilities they perpetuate) you
have go right to a universal language combining the expressiveness of
natural language with the precision of logic and mathematics.
Is it easy to state ordinary things exactly right in such a language?
No, it usually requires careful analysis. It's often simpler to say
things in a specialized language tailored to the job at hand, but then
you get incompatibility.
The CSMF efforts (involving KIF, SUMM and Conceptual Graphs) come
closest to full expressiveness, as does the language CycL used in Cyc,
and (I assume) the language now used at Ontek -- although standard KIF
might have had its expressiveness trimmed too much to be useful. Full
higher-order predicate logic should be included, with some general modal
capability. The only noteworthy technical problem now is integrating
declarative and procedural languages, such as the logic-ish versus
Pascal-ish parts of EXPRESS. There is a lot of relevant theory
available for that, in the form of dynamic logics, denotational
semantics, and numerous other approaches, but most practitioners will
have to ignore these since the theories are too hard to learn. This
particular problem can benefit from benign neglect, since nobody
actually needs a quick solution.
Once you decide on a fully expressive language, you can get to the key
task, which is to agree on your underlying common "ontologies".
The thing to avoid is another "lowest common denominator" (= lowest
expressiveness) modelling language. That will just re-start the
historic cycle of incremental improvements and incompatibilities.
Yours truly, Fritz Lehmann
Cyc, Cycorp, 3500 W. Balcones Center Dr., Austin, TX 78759 fritz@cyc.com