Friday, November 18, 2005
Center of Excellence Proposal
à
Communication from John Sowa and Rick
Regarding the hub approach:
In our first meeting John Sowa suggested that we tried this before and
we better have something new and different. The technical and philosophical
principles of the world wide web answer John's question and the issue at hand
is how ontology and taxonomy co-ordination informs ontology and taxonomy
design, not how design restricts co-ordination.
The more interesting of these principles include tolerance,
decentralization, test of independent invention, principle of least power, free
extension, language mixing, and partial understanding.
See
http://www.w3.org/DesignIssues/Principles.html and
http://www.w3.org/DesignIssues/Evolution.html
Where these principles apply, information flows through channels based
on observable regularities. A channel provides an infomorphism across
classifications at its source and destination based on an interpretation that
meets conditions that can preserve structure or semantics. The expressiveness
of a language restricts the logic used in classifications.
Our team at GSA is currently working through a use case based on this
approach using MDA and Semantic Web technologies.
Regarding FEA-RMO:
TopQuadrant's report calls out a design pattern called an Axiom Bridge
which links the PRM and the BRM. So where TQ has already shown how to
"connect" two FEA-RMO ontologies, what's different about other
connections? Other options also include Marco Schorlemmer's IF-Maps methodology
referenced here
http://www.aktors.org/technologies/ifmap/
So, I think it's in the context of information flow that will give us the new and different we need here ...
That's a very important point:
When two people (or programs) interoperate successfully, the primary requirement is *not* that they have identical worldviews on every detail. The major constraint is that they agree on just that subset of categories that are relevant to the information flow between them.
Mathew West made an important point, which I modified by changing the word "unfortunately" to "inevitably", changing "several" to "untold numbers", and adding the phrase "for the purpose at hand":
MW as modified by John Sowa
[Inevitably], even with one universe
it is theoretically possible to come up with an infinite
number of ontologies - almost certainly none of them truly
correct, but possibly [untold numbers] of them being
accurate enough to be useful [for the purpose at hand].
The professor of industrial engineering George Box made a related point in a pithy and widely quoted observation:
All models are wrong; some models are useful.
A formal ontology is an axiomatization of a model of some domain, which may be as large as the entire universe, but more likely is a much smaller model of some domain of interacting applications. Furthermore, the information that flows among any set of people (or programs) is usually much smaller than the union of what all of them know.
Matthew also added the following point:
MW> Merging ontologies will only be possible where the
same choices have been made for these (and perhaps other)
things. Between ontologies that have made different
choices, the ontologies can be expected to differ in
their account of the same real world phenomena in a
way that cannot be simply merged.
That may be true, but we
should ask the next question: If we want Program A to interoperate with Program
B, why should we merge every aspect of the ontology that was used by the
developers of Program A with every aspect of the ontology used by the
developers of Program B?
One term I like is "task-oriented interoperability": if you try to merge two ontologies, you have to look at the *union* of all the categories in both. But if you want to enable two programs to interoperate, you only need to look at the subsets that are relevant to the task.
Merging two small ontologies is much, much easier. And more importantly, if you are only looking at a specific task, it is very likely that the subsets appropriate to the task will have similar perspectives.
Recommendation: Shift attention from the unsolvable problem of building, merging, and coordinating global world views to the task of developing an open-ended collection of modules that can be selected, assembled, and tailored for particular tasks or collections of tasks.
John Sowa