Sunday, February 26, 2006
Generative Methodology Glass Bead Games
On the limits of the OWL standard à [184]
Reading material [1]
Reading material [2]
Reading material [3]
Summary of the discussion up to this point à [186]
The issue of n-articulated ontology
“multi-articulated” ontology
Related discussion on ontologyMapping thread à [17]
On Erick’s comment on the same subject à [204]
Communication from Irene
Foot notes by Paul Prueitt
Paul,
There is some incompatibility between RDFS and OWL that created difficulties in the layering of OWL and other higher order languages. Ian Horrocks has written on this topic in several papers with a lot of clarity .
The problem had to do with the fact that the model theory of RDFS is outside of the First Order Logic model theory. It can be transformed into FOL. However, the transformed language falls outside of decidable FOL. (By decidable I mean something that can be proven to be decidable. In practice, the fact that something is not proven to be decidable, does not necessarily make it undecidable.) à comment by Paul Prueitt (footnote) [4]
DL people working on OWL were very interested in the decidability. (In addition to DL, other languages that are subset of FOL are those that use Horn rules. F-logic from Ontoprise is one example. Datalog engine you have been working with from Intellidimensions (RDF Gateway) is another example.)
The incompatibility problem was addressed in the OWL standard by creating OWL-DL subset of OWL. This was done before OWL became W3C specification.
This quote is from one of Ian's papers mentioned below:
"It is interesting,
however, to consider the ``FOL subset'' of RDFS. This language would correspond
to RDFS with a reinterpretation of the meta-classes and other structural
vocabulary of RDFS. Classes such as rdfs:Class and rdf:Property would be
removed. Vocabulary such as rdf:type and rdfs:domain would simply be a way of
stating the appropriate FOL statements and would no longer denote elements of
the domain of discourse. The resulting language would be equivalent to a very
simple description logic, and would also be expressible in Horn languages. This
language would form an ideal base layer within the FOL framework and would, via
progressive extensions of the syntax, allow for a full semantic layering of
languages such as DAML+OIL and OWL."
This is essentially what was done [5]. Note that in OWL, we do not use rdfs:Class, but use owl:Class instead.
OWL DL is based on a subset of RDFS. In other words, within OWL DL not every rdfs:Class is owl:Class
OWL Full includes all of RDFS, as well as all of OWL DL
You can read more on the topic in http://www2003.org/cdrom/papers/refereed/p050/p50-horrocks.html and http://www.cs.vu.nl/~frankh/postscript/OntoHandbook03OWL.pdf
At this point this is pretty much history. But there are lots of other things going on right now. These are resulting from learning from the uses of the standard in the industry and understanding additional requirements. They include:
Btw, if you are trying to learn OWL by using Protege, it may (and has) created confusion.
Protege is a capable tool, but it has its own long history that precedes RDF and OWL. Protege's foundation is quite different from RDFS and OWL. It is easy to make wrong conclusions regarding these languages based on Protege's user interface. Common misconceptions I came across include:
I would recommend you to try SWOOP and to play with its built-in inferencers. It also gives you the ability to see the RDF/XML (and other serializations) as you define a class or a property.
I do not necessarily suggest that you switch out of Protege completely. It has some nice features not present in SWOOP and the user community is large and supportive (although the degree of knowledge varies and you can get some wrong or misleading answers). I just think that SWOOP is a better way to learn.
[1] http://dip.semanticweb.org/documents/ECIS2005-A-Methodology-for-Deriving-OWL-Ontologies-from-Products-and-Services-Categorization.pdf
[4] Comment by Paul
Prueitt: This are technical academic
arguments that concern a certain class of academic scholars, NOT ONTOLOGY. Ontology is nature. Ontological
modeling using first order logics, or any of the logics that is in favor within
this class of academic scholars, may not (it might turn out) fulfill the needs
that are “there” in reality. So the
discussion that you are making at this point is an academic discussion which
MUST be balanced by arguments by natural scientists that web ontologies are
“mis-named”. I see nothing clear about
any of the papers that Horrocks has authored.
There is no question in my mind that the work is technically correct, it
is the issue of relevance that concerns us (the Second School). Same is true for fuzzy logic or rough
sets.
[5] Comment by Paul Prueitt. When? When was this done? Is it part of the Protégé software tool?