A couple of points: (01)
1) Mike Daconta seconded the use of Protege and suggested RDF/S for the KR
language. Protege supports RDF/S (and now UML, XML [DTDs and Schemas]), but is
based on the OKBC knowledge model, roughly equivalent to KIF. This just means
that you might be able to model stuff in Protege that has no expression in
RDF/S, etc., i.e., those formalisms with less expressive power. This is just a
warning, and by no means a show-stopper.
2) the typical application path for ontologies, unless the ontology manager
directly supports efficient inference of the kinds needed, is to transform the
ontologies (by this I mean the class level and the instance level assertions,
the whole shebang) to Horn Clause form for run-time execution by an efficient
Prolog engine. XSB, Amzi!, and I believe binProlog, are free. Most modern
Prologs are WAM based (Warren Abstract Machine), a kind of logical assembler
language, and compile into C, then to an executable. So the target application
model is roughly a deductive database (logic programming + relational database,
so that you get inference and set at a time operations from a RDB). The
OntologyWorks tool (not free, but probably the best overall tool available in
the ontology realm) is like this, transforming (or knowledge compiling) from
KIF/Common Logic to a deductive database. Protege has plugin support also for
FLORA, which is an F-logic implementation that lives on top of XSB Prolog. FLORA
is an OO-like language. Java + Prolog would seem to me to be the best path.
3) an alternate path is to use JESS (Java Expert System Shell based on CLIPS),
which uses production rules rather than inference -- a clear second choice,
since in general production rules "simulate" inference and have potentially
damaging side effects. A side effect-free set of production (condition-action)
rules is probably close to logical implication, however, and may not be
problematic. So this is just a warning. (02)
Leo (03)
Farrukh Najmi wrote: (04)
> MDaconta@xxxxxxx wrote:
>
> > In a message dated 3/3/2003 10:58:11 AM US Mountain Standard Time,
> > farrukh.najmi@xxxxxxx writes:
> >
> >> I would like to propose that the proposed UBL ontologies be managed
> >> using ebXML Registry as an Ontology Server. There are many interesting
> >> features that an ebXML Registry has to offer as an ontology server. A
> >> partial list includes:
> >
> >
> >
> > This is interesting as I have not thought of the ebXML registry as
> > an Ontology server.
>
> Michael,
>
> You are correct in pointing out that ebXML Registry as defined is not an
> Ontology inference engine. More specific Ontology support is being
> planned for V4 of ebXML Registry (V3 will soon be approved by TC).
>
> However, it is a general purpose content management system that can be
> used to manage any type of content. Specific information models (e.g.
> OWL) may be mapped to ebXML Registry using binding.
>
> > For example, I do not believe the RIM supports
> > the formal notion of 'subclassOf" which would be critical. While I
> > believe we could use a custom association with this label, that is
> > weaker than the notion of subclass being built into the RIM.
>
> I agree with above statement. Built-in support for ontologies are being
> planned for V4.
>
> > For example,
> > a formal notion of subclass would allow the child information object to
> > automatically inherit the attributes of the parent. Please correct me
> > if I am misunderstanding the RIM or its implications.
>
> Your assessment is correct.
>
> >
> >
> > Additionally, I would recommend the Ontology classes be associated
> > with a terminology registry for each concept (in essence equating a
> > class with a concept). Following step 3, in the protege Ontology 101
> > document, we need to enumerate important terms in the Ontology.
> > I am proposing a step beyond enumeration to formal definition with
> > concept, terms and referents. Is the ebXML registry suitable for a
> > terminology
> > registry?
>
> This is essentially the use of ebXML Registry that I was envisioning.
> The terminology from an Ontology be mapped to a ClassificationScheme in
> RIM following a specific binding that overcomes limitations of single
> inheritence etc. using custom association types such as (subClassOf).
> Such a ClassificationScheme mapped from an Ontology could be used to
> classify UBL (and any other content) managed within or outside the ebXML
> Registry. The automatic content cataloging feature of the registry could
> be used to classify specific content using the ontology mapped
> ClassificationSchemes. The ontology to RIM binding would also define
> custom ad hoc queries that could be used to do ontology based queries
> such as "Find all objects classified by an ontology class or its sub-class".
>
> The main thing we would be lacking is a truly open-ended ontology
> inference engine. This could be addressed by an external ontology engine
> for now and in future be available as a feature of the ebXML Registry.
>
> > Or do people know of others?
> >
> > - Mike
> > -------------------------------
> > Michael C. Daconta
> > Chief Scientist, Advanced Programs Group
> > McDonald Bradley, Inc.
> > www.daconta.net
>
> --
> Regards,
> Farrukh
>
> _________________________________________________________________
> To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx
> Msg Archives: http://ontolog.cim3.net/forum/ontolog-forum/ (05)
--
_____________________________________________
Dr. Leo Obrst The MITRE Corporation
mailto:lobrst@xxxxxxxxx Intelligent Information Management/Exploitation
Voice: 703-883-6770 7515 Colshire Drive, M/S H305
Fax: 703-883-1379 McLean, VA 22102-7508, USA (06)
_________________________________________________________________
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx
Msg Archives: http://ontolog.cim3.net/forum/ontolog-forum/ (07)
|