AW: AW: Conformance Layers in WebOnt

Hi -

besides CARIN, I'ld also suggest to have a look at AL-Log [0], which
was proposed in the DWQ project [1]. In a nutshell, AL-Log is a simple
DL plus datalog. Of course, as [Levy and Rousset, 1996] pointed out in their
paper on CARIN, one has to restrict the expressive power of the DL to remain
decidable.

Unlike CARIN, AL-Log has an A-box.

Refs:
-----
[0] Donini, Lenzerini, Nardi, Schaerf:
AL-Log: Integrating datalog and DLs
Journal of Intelligent Information Systems (JIIS), 27(1), 1998

[1] Calvanese, De Giacomo, Lenzerini:
DLs for information integration
In: Computational Logic: From Logic
Programming into the Future,
LNCS 2001

[LR96] Alon Y. Levy, Marie-Christine Rousset:
Carin: A representation language combining horn rules and description
logics.
In: Proceedings of the 12th ECAI, pp. 323-327, 1996
-----Ursprungliche Nachricht-----
Von: Ian Horrocks [mailto:horrocks@cs.man.ac.uk]
Gesendet: Sonntag, 28. April 2002 23:01
An: Frank van Harmelen
Cc: Steffen Staab; volz@aifb.uni-karlsruhe.de; Daniel Oberle
Betreff: Re: AW: Conformance Layers in WebOnt


On April 10, Frank van Harmelen writes:
>
> > Frank van Harmelen wrote:
>
> >>One proposal is to remove defined classes out of level 1
> >>(ie only <=, not =). This means no classification of instances in level
1.
> >>
>
>
> Steffen Staab wrote:
>
> > do you then syntactically disallow
> >
> > A <= B and B <= A
>
>
> Sorry, unclarity on my part.
>
> The proposal (by Mike Dean) is to have definitions for "primitive classes"
in
> layer 1 and definitions of "defined classes" in layer 2.
> Only defined classes are characterised by necessary & sufficient
conditions,
> primitive classes are characterised only by necessary conditions.
> Thus, when you want to classify instances to named classes, you can only
do
> this for the defined classes (and thus this type of reasoning is not
possible
> (and thus not required to implement) for level 1).

As I said in my email to the WebOnt WG, I suggest we don't talk about
definitions, primitive or defined, as they don't have any clear
meaning in our setting (e.g., membership of a class can be inferred as
a result of range/domain constraints or, as Steffen pointed out, as a
result of cyclical inclusion axioms).

I think that the idea of using OWL level 1 plus rules is an
interesting one. I don't think that it is necessarily important that
OWL level 1 is the largest language that is still decidable when
combined with rules (besides which, there may not be any single "most
expressive" language meeting this condition). As Raphael has
suggested, there is a well know literature on this subject, in
particular the Levy papers on Carin (I guess that we are all familiar
with these) where the DL is Classic, but where there are some
restrictions on how classes and rules can be combined.

Regards, Ian

>
> Frank.
>     ----
>
>
>

Received on Sunday, 28 April 2002 18:12:13 UTC