RE: pun in ontolingua KB

fritz@rodin.wustl.edu (Fritz Lehmann)
Date: Wed, 22 Jun 94 03:51:21 CDT
From: fritz@rodin.wustl.edu (Fritz Lehmann)
Message-id: <9406220851.AA26049@rodin.wustl.edu>
To: gruber@HPP.Stanford.EDU
Subject: RE: pun in ontolingua KB
Cc: Doug@SURYA.CYC-WEST.MCC.COM, cg@cs.umn.edu, ontolingua@HPP.Stanford.EDU,
        pdoudna@aol.com, srkb@cs.umbc.edu
Sender: srkb-owner@cs.umbc.edu
Precedence: bulk



     One means of giving real-world semantic content to the
possible meanings of primitive and defined terms in an ontology is
to provide a real-world example for every class (every primitive
class, at least).  It would be a proper name or some otherwise
indexical-for-all-mankind designator.  Example: the Eiffel Tower.
If Tom Gruber has a class "BUILDING" in some Ontoligua ontology,
he may wonder if a class in CYC, say
"FredsFavoriteKindOfJungleJim", can correspond to BUILDING.  The
words don't help.  It could be Gruber's requirement that, to match
BUILDING, a class in some other ontology _must_ include the Eiffel
Tower at least.  This does nothing to guarantee equivalence, but
at least all classes for which the Eiffel Tower is excluded are
eliminated as candidates.  Suppose there were a shared database of
arbitrary real-world examples.  (The Eiffel Tower, the U.S. Bill
of Rights, the rocking-chair in room 206 of the Rijksmuseum, O. J.
Simpson, Bart Simpson, the Risorgimento, the Hope Diamond, the
Petersen Graph, Albania, etc.)  Every primitive class would be
constrained (not determined) by the incidence relation with this
set of examples.  The examples are _not_ descriptions in any
formal language; they are named instances known, or knowable, to
any person on earth.  If the number of examples provided is
increased (either in the whole database, or for a given class),
the reliability will increase monotonically.  This can be
formalized, e.g. with Pawlak's Rough Sets or Wille's Formal
Concept Lattices.  (Tony Cohn and I have a lattice of 42
qualitative degrees of agreement between classes so defined, in
our "EGG/YOLK theory" for knowledge integration.)  Within a
particular field of application, the example database could be
specialized.

     The point of using proper-named indexical examples is to
exploit the shared "real world" we live in for the sharing of
knowledge bases.  This could also flag the undesirable "ontology
puns" noted by Ben Kuipers, although not fully reliably.

     Has anyone tried this?

                          Yours truly,   Fritz Lehmann
GRANDAI Software, 4282 Sandburg Way, Irvine, CA 92715, U.S.A.
Tel:(714)-733-0566  Fax:(714)-733-0506  fritz@rodin.wustl.edu
=============================================================