On 2/2/2012 1:37 AM, Cory Casanave wrote:
While
we may be dancing around it I don’t think you can separate
executability from ontologies, the entire point of ontologies is
to be able to “enact” what the ontology implies.
Hi Cory,
I think this is a case of trying to oversell what ontologies are/can
do. In my view, being able to 'enact' implications is in not the
'entire point of ontologies'. A reasoner is an implemented
algorithm, an ontology is not. To me, the entire point of ontologies
is actually the formal clarity of definition that other technologies
lack. Of course, implications are interesting but for something like
systems engineering they cover a very small scope compared with
what's required to do the job.
In a large scale application, ontologies are IT artifacts, just like
RDBs, reasoners, analysis algorithms, visualization algorithms,
access control mechanisms, etc. and ontologies and those other IT
artifacts must interact. Analysis algorithms can 'execute over'
ontologies that contain measurement data that's incomprehensible to
a reasoner, just like it could execute over an HDF5 dataset of the
same measurements - so we're going too far in suggesting that the
ontology is itself executable or that it's not useful otherwise.
A perhaps related comment : I had a debate a few months ago with
someone who claimed that 'you cannot test a reasoner'. In the end, a
reasoner is just an implemented algorithm (similar to a compiler)
and so normal software testing approaches can be applied ... this
stuff ain't magic i.e. it's complicated, but not complex. To me a
more interesting concern is 'How do you test a large-scale
ontology?' and ... 'Can you only really test the algorithms that
execute over it?'
Cheers,
David
--
Managing Director and Consultant
TopQuadrant Limited. Registered in England No. 05614307
UK +44 7788 561308
US +1 336-283-0606
|