SUO: RE: Problems in SUMO
Pat et al
Note the (superficial) similarity between SUMO and EPISTLE - the latter's root concept is 'thing', with two mutually exclusive subtypes possible_individual and abstract_object. There are differences between the two even at this level (see the recent discussion between Ian N and Matthew W), so that there is no simple mapping SUMO:Physical <-> EPISTLE:possible_individual. However, in both cases developers have found it useful / discovered it necessary to start with a very basic decomposition - in the EPISTLE case, into things that have spatio-temporal extent (possible_individual) and those that don't. The key for EPISTLE (and I assume, from a basis of rather less knowledge, for SUMO) is that the ontology starts from a world view in which it is asserted that everything is a possible_individual or it is not (physical or abstract in the case of SUMO), and that it is anathema for there to be something that is both.
Any such ontology needs to admit of two possibilities:
* that we discover some thing that really is not either an A XOR a B - this would represent a falsification of the theory underlying the ontology that would be better addressed by reconstructing the highest level of the ontology than making ad hoc modifications to it (such as deciding that it is, after all, possible for something to be both a possible individual and an abstract object)
* that the knowledge I have of something may be incomplete - I may at some point discover some thing X but not have determined (or have no way of determining) what kind of thing it is, even at the level of its classification (in EPISTLE terms) as a possible_individual or an abstract_object. The existence of a 'top' or 'thing' class (which you cite as being neither abstract or physical in the SUMO case, but I would interpret as being the union of both) would allow be to acknowledge the existence of X as a thing [1]. Therefore, 'Pat Cassidy' is a member of EPISTLE:possible_individual, 'ontologist' is a member of EPISTLE:abstract_object, and both are members of EPISTLE:thing (which is, in fact, the only class that both can be a member of within this particular ontology).
I would also, by the way, be wary of using inheritance mechanisms as a means to construct an ontology - inheritance should be a consequence of the relationships defined in the ontology.
regards
Julian
[1] The formal definition of the EPISTLE core model doesn't actually allow me to do this, but that's a consequence of itgs schitzophrenic existence as an ontology and as a data model.
-----Original Message-----
From: Patrick Cassidy [mailto:pcassidy@bellatlantic.net]
Sent: 2003-06-27 12:51
Cc: standard-upper-ontology@ieee.org
Subject: SUO: Problems in SUMO
After working on OpenCyc for a while, I will
now inquire about some features of SUMO whose
motivation isn't clear to me.
Just one issue for this note. Is there
a need for SUMO to have only two top classes,
"Physical" and "Abstract"? Other ontologies have
more than those two at the top semantic level.
CYC for example, has "PathSystem" which has
both physical and abstract subclasses. Wordnet
has over a dozen top categories. I have over a dozen
in the ontology I am building, such as "Object" which
can have both abstract and physical subtypes.
The point is that there are concepts that have
both Abstract and Physical subclasses, so cannot be
considered either Physical or "Abstract", where
"Abstract" means inheritably abstract. In common
usage, "abstract" merely means "not physical"
but for classes, that should not necessarily
entail that the "Abstract" quality is inheritable.
If SUMO were willing to admit additional classes
into the top level, it could be made more closely
interoperable with other ontologies.
One additional note:
For those who might be tempted to object that
surely all instances of things in the world can
be classified as "abstract" or "physical", I agree.
But that kind of binary classification to generate
a hypercube space of all possible points in
an abstract concept-space is a different sort of
classification from the inheritance hierarchies that
are far and away more typical for ontologies.
The two methods of classification are incompatible
if one wants to view the points in concept space
as classes that are disjoint in any given dimension.
Consider just the "Top" level. That itself is
nether "Abstract" nor "Physical". So why not have
other concepts that are neither?
Pat
=============================================
Patrick Cassidy
MICRA, Inc. || (908) 561-3416
735 Belvidere Ave. || (908) 668-5252 (if no answer)
Plainfield, NJ 07062-2054 || (908) 668-5904 (fax)
internet: cassidy@micra.com
=============================================