The question of whether to treat roles as classes has been
a vexing one in ontology development, and needs to be
decided at the first point in a real project. (01)
Dean Black wrote:
> I don't see Buyer as a subclass (or interface) of Party (or another
> class), because Buyer is a totally transient notion, a role as you
> pointed out, and a class hierarchy in an ontology (I believe) should
> be trying to model more permanent notions. For example, a Zebra is
> always permanently a Mammal, and it doesn't transition or morph from
> being one species to another, therefore it would be a legitimate
> subclass of Mammal in an ontology. (Hope I'm making a clear and
> understandable distinction between "permanent" and
> "transient" states of being.)
>
Some roles persist indefinitely after some time point, such
as creator of an artifact, or biological parent of a person.
Otherwise for all real-world applications and particularly
business, roles will only be asserted within some event
occurring in a bounded period of time. If assertions about
real-world events are always accompanied by specification of
the time interval in which they are true (I think they should be)
then there shouldn't be any logical problems having roles as
subclasses of the objects that play those roles, with the
differentiating characteristic of the class being the (transient)
relation that gives rise to the role. Role classes can also
be differentiated from "permanent" classes in various ways,
such as being instances of a RoleType metaclass. However, the
only *instances* of such role classes that would occur in a base
ontology are those that are persistent, or those relating to
time-bounded events that occurred in the past. (02)
e.g.
(holdsInInterval (isInstanceOf BillClinton USPresident)
Jan1992-Jan2000)
and (isSubclassOf USPresident Person)
from which we can conclude that BillClinton is a Person
(Person being a stable or "rigid" class type). (03)
I prefer this style of representing roles because it
allows a less clumsy association of linguistic phrases with
logical representations -- "The chairman called the meeting
to order" etc. It's also a simple way of specifying
restrictions on what sorts of things can play roles.
I could live with a "cleaner" ontological approach if
someone can suggest a straightforward alternate method for
handling the various ways roles interact with time intervals.
I know the "4-dimensional object" style of representation
is one way of dealing with roles, but I prefer to avoid it
for other reasons.
I think that there may be more than one logically compatible
way of handling such such issues, and perhaps several can
be implemented, with a translation utility to convert
one style to another. Not having any practical experience
in ontology applications, I can only surmise that different
styles may be more computationally convenient for different
applications, and if it is possible to find precise translations
between styles of representation, it would serve the
purpose of developing a widely acceptable ontology to allow
all logically compatible styles. (04)
Pat (05)
=============================================
Patrick Cassidy (06)
MICRA, Inc. || (908) 561-3416
735 Belvidere Ave. || (908) 668-5252 (if no answer)
Plainfield, NJ 07062-2054 || (908) 668-5904 (fax) (07)
internet: cassidy@xxxxxxxxx
============================================= (08)
_________________________________________________________________
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx
Msg Archives: http://ontolog.cim3.net/forum/ontolog-forum/ (09)
|