embedding logics in RDFS

hi All,

until recently I really had difficulties to understand the benefits of the
semantic web "layer cake"-model (which says that ontology languages like
DAML+OIL should build on a generic RDF(S)-layer, which, in turn is written
down in XML.)
Well, now I think I may have grasped the idea, but I am not sure if I got
it right. Therefore I post my interpretation to this list, in hope of
responses.

The trouble I had with the layer cake model was that I did not see any
benefits of the use of RDF/RDFS: The main point was, that even if
formalisms like DAML+OIL or other DL's build on top of RDF(S), an agent
still needs to have *built-in* capabilities to interpret and reason on
DAML+OIL ontologies.

So, even if ontologies in DAML+OIL and other languages are written down in
one common way (RDF), they still represent completely different worlds!
Many papers tried to relativate this, stating that RDFS-based agents which
are not capable of e.g. DAML+OIL could understand DAML+OIL ontologies at
least partially, because some properties of DAML+OIL constructs are defined
in RDF schema. (However this never convinced me. I do not trust systems
which act upon partial (=uncertain) information, should i?)

So, my main problem was, that most of the semantics of languages like
DAML+OIL where declared *outside* their respective RDFS definition (e.g.
DAML+OIL's semantic is defined using KIF axioms in [1])

But why not encode the *complete* semantic interpretation of a language
like DAML into RDF(S)? This would enable agents to "load" and use different
problem solving mechanisms "on the fly"  - - and that would really justify
the role of RDF as a generic KR layer.

I am aware of two papers which approach this problem: In [2] W. Conen and
R. Klapsing propose an extension to RDFS by introducing a property
"isDefinedAs" which "hosts" arbitrare logical constructs, e.g. prolog
clauses, which might be applied as follows:

<Property rdf:ID="union">
  <rdfs:isDefinedAs rdf:parseType="Literal">
    in(X,S,P,O) :- P = union, instanceOfSet(X,S).
    in(X,S,P,O) :- P = union, instanceOfSet(X,O).
  </rdfs:isDefinedAs>
</Property>

Now, if for example the whole DAML+OIL specification would be encoded this
way (much work already done in [3], which expresses some DAML+OIL-semantics
as sequences of prolog clauses), any RDFS capable agent in combination with
a prolog interpreter could make use of the semantics of DAML's
epistomological constructs.

One problem with this approach is of course, that the use of a specific
logical notation (e.g. horn clauses) may not scale in Semantic Web context
(again we deal with "different worlds")

Another approach is presented by D. McDermott and D. Dou in [4] which
proposes the encoding of logics as reified RDF statements. The logical
formulas are "quoted" in RDF in a clean, RDF conform way. Thus, logical
formulas written down in reified RDF statemens are parseable by any RDF
parser.

however, one problem still remains: is there a logical calculus (e.g. HOL)
or notation (e.g. KIF) available which is able to express all other
(relevant!) logical calculi? [which calculi may be considered relevant,
which not?]

If this question could be solved, then the SemWeb layer cake would finally
make sense to me: a nearly selfdescribing multilayered semantical
framework, building on top of a generic syntactical framework would be
possible. perfect infrastructure for a semantic web ! :-)

what is your opinion on this?

thanks in advance,
Joachim


References
--------------------

[1] Fikes, R; McGuinness, D.L.: "An Axiomatic Semantics for RDF, RDF-S, and
DAML+OIL",
2001http://www.ksl.stanford.edu/people/dlm/daml-semantics/daml+oil-axioms-october2001.htm

[2] Conen, W; Klapsing, R: "Utilizing Host Formalisms to Extend RDF
Semantics", 2001
http://wonkituck.wi-inf.uni-essen.de/rdfs.html

[3] Zou, Y. : "DAML XSB interpretation", lab notes,
2001http://www.cs.umbc.edu/~yzou1/daml/damlxsb.P.txt

[4] McDermott, D; Dou, D. :"Embedding logic in DAML/RDF", 2001
ftp://ftp.cs.yale.edu/pub/mcdermott/papers/wwwconf.pdf


Joachim Peer
Research Assistant
Institute for Media and Communications Management
Blumenbergplatz 9
CH-9000 St. Gallen
Tel.: ++41 (0)71 224 3441

Received on Monday, 14 January 2002 09:19:23 UTC