[Top] [All Lists]

Re: [ontolog-forum] Foundation ontology, CYC, and Mapping

To: "[ontolog-forum]" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: "John F. Sowa" <sowa@xxxxxxxxxxx>
Date: Fri, 05 Feb 2010 09:02:27 -0500
Message-id: <4B6C24F3.7080401@xxxxxxxxxxx>
Dear Matthew, Gian Piero, Pat C, Pat H, and Sjir,    (01)

For a long time, I have been impressed with the results that Matthew
obtained at Shell.  I also support the analysis of documents written
in natural language, which Gian Piero has been doing.  I believe
that these two approaches, which are very different in nature, are
complementary.  But a lot of explanation is necessary to show how
they are related.    (02)

PH>> You [PC] are claiming that if your hypothetical 'foundation
 > ontology' were available, that the US economy would gain 100 billion
 > dollars in a matter of weeks, simply by more efficient information
 > transfer? First, I simply do not believe that anyone, with any
 > methodology, could possibly sustain this estimate based on any
 > actual data.    (03)

MW> Well I have looked at this, and I don't know what the number
 > is, but it is very large. Just in Shell I found that it was so
 > unbelievable that I scaled the answer I got back by a factor
 > of 10 before talking to anyone.    (04)

I don't see any conflict between what Pat and Matthew have said.
Pat H. was criticizing Pat C's claims that an FO, by itself, would
magically provide very large savings in a very short time, and
Matthew said that the approach he and his group had developed
saved a great deal of money for Shell.    (05)

In the next response, Matthew supported Pat H's point:    (06)

MW> This is more where the flaw lies. The myth that if you have
 > a better mousetrap, people will beat a path to your door. I think
 > it would take at least 20 years for a widely supported ontology
 > to have this sort of effect, possible more. Certainly I think
 > that managing the politics of take up is a much tougher problem
 > than developing the ontology itself.    (07)

That 20 year estimate is realistic for the kind of savings that
Pat C. claimed.  I believe that we should aim for that goal, but
I don't believe in the brute force method:  "If you build it,
they will come."  That quotation is from a movie about a man who
built a baseball park in a cornfield.  He conjured up some ghosts
from the past (literally), but he didn't make a profit.    (08)

MW> As another example, about 5 years ago when I was still working
 > for Shell I was charged with developing a conceptual data model
 > for Shell's Downstream business (oil tanker to petrol pump). We
 > started with ISO 15926 as a foundation ontology, and developed
 > a 1700 entity type data model in 12 months with a team of ten people.    (09)

That is the kind of work that brings results.  But note that Matthew's
approach had almost no similarity to what Pat C has proposed.  He
and his group developed an ISO 15926 over a period of years by an
incremental approach of solving one specific problem at a time.
For more about ISO 15926, see the Wikipedia article:    (010)

    http://en.wikipedia.org/wiki/ISO_15926    (011)

That article has some references, including a pointer to a critique
by Barry Smith, who said "Conclusion: ISO 15926 Is Not An Ontology":    (012)

    http://ontology.buffalo.edu/bfo/west.pdf    (013)

I don't want to get into the details of that debate, but I'd point out
that ISO 15926 is probably much closer to an ontology than 90% of the
things that are implemented in RDF or OWL.  In any case, Matthew and
his group at Shell have demonstrated that something along those lines
can be very effective.  But -- and this is a very big **BUT** --
ISO 15926 and its applications were developed in an incremental
approach for a company that had crucial problems to be solved,
and the developers had to solve them under the constraints of
budgets and deadlines.    (014)

SN> Could you indicate how many relationships were used? And were
 > they all binary, or did you include n-aries?    (015)

That sounds like a question designed to elicit a factoid that
could be used to support some kind of debate.  I don't see any
reason for a debate.  There are many different kinds of logics
and databases that use many different kinds of relations, and
we need to support mappings among them all.    (016)

GPZ> To be concretely useful, a system of primitives must be
 > a) simple; b) general. This rule out both Pat's thousands of
 > primitives and Wierzbicka primitives, which are too "surface level"
 > (see terms like > "I", "You", "People", "For some time" etc.).    (017)

I agree that I wouldn't choose such terms as the foundation for an
ontology, but I would be happy to include them in the microtheories
of a more general framework.    (018)

GPZ> In NKRL (Narrative Knowledge Representation Language) we have only
 > 14 primitives, seven "semantic predicates" (Behave, Exist, Experience,
 > Move, Own, Produce, Receive) and seven "functional roles" (subject,
 > object, source, beneficiary, modality, topic, context). Terms
 > pertaining to generic ontologies - we have our own ontology in NKRL,
 > including presently more then 7,000 terms, but we have successfully
 > used other ontologies in a standard Protégé style - are associated
 > with the predicates through the functional roles to build up
 > conceptual structures like, in simplified notation,
 >   (MOVE (SUBJ John) (OBJ BOOK_1) (BENF Mary)).
 > ...
 > A powerful inference system, including causal reasoning, is used
 > to take advantage from the advanced characteristics of the language.
 > http://www.springer.com/computer/artificial/book/978-1-84800-077-3.    (019)

This is very different from what Matthew is doing, but many people
have been working on processing documents in order to extract
ontologies and to relate them to reasoning systems that are also
tied to detailed business applications.    (020)

GPZ> My last NKRL application has concerned dealing in an 'intelligent"
 > way with the "storyboards" built up in a gas/oil industry context...    (021)

That would be very interesting, especially since "gas/oil industry"
would include Shell.  Could you elaborate your ideas about how an
ontology could help relate Matthew's applications to yours at Shell
(or any other large business)?    (022)

At our VivoMind company, we have also been using ontologies for
relating unstructured data in NL documents to highly structured
data in databases and computer software.  The following slides
summarize the ideas.  (The last slide has further references.)    (023)

    Pursuing the goal of language understanding    (024)

Slides 24 to 27 describe an application for relating the software
of a large company to the documents (manuals, reports, email, etc.)
Slides 30 to 38 describe an application that analyzes a query by
a geologist and derives an answer by relating it to the contents
of documents that had previously been analyzed.    (025)

There is much more to be said, but I believe that ontologies can
be useful for relating the highly structured information used in
computer systems to the much more flexible formats of NL texts.    (026)

John Sowa    (027)

Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx    (028)

<Prev in Thread] Current Thread [Next in Thread>