Re: why can't describe the semantic of DAML-S by Description Logic

On Tuesday, September 2, 2003, at 09:07  AM, lsp wrote:
> Hi,
>
> There have papers describing the semantic of DAML-S using situation
> calculus, Petri net, operational semantics. But since DAML is naturally
> described by Description Logic and DAML-S is just ontology for service,
> why can't we describe DAML-S by DL?

Good and natural question

> Any idea?

Several. First, in so far as we're just after the semantics of our 
*ontology*, DL may be up to the job (it's not actually clear given some 
represenational choices which seem to have forced treating some 
properties as individuals, usually a DL no-no; judicious reworking of 
the ontology might well avoid this).

But let's turn to the semantics of the process model alone. It's 
questionable weather *any* description logic can completely specify the 
semantics of the process model construct, in their full interaction 
with preconditions, inputs, outputs, etc. You might look at a prior 
posting of mine on Processes as Properties (which used the even more 
impoverished DL OWL DL). The most likely approach would be to use a 
rather more expressive DL than OWL DL that including various role 
constructors (or use an extension of a subset of OWL DL). Such logics 
correspond to Propositional Dynamic Logic and are capable of expressing 
(and reasoning with) such constructs as if then, repeat, etc. However, 
the Propositional limitation is a nettlesome one.

However, I remain ever so slightly charmed by this general approach, 
and if someone wanted to run further with it, I'd be happy to cheer. A 
good place to start is to identify a natural and useful class of 
problems associated with Web services that would naturally be described 
with PDL.

Finally, notice that it's somewhat tricky, given the standard DL 
reasoning services, to get even such obvious wins as matchmaking right. 
This was brought home to use at the second SWSL F2F by Ian Horrocks (he 
has a paper explaining the problem) on using subsumption for 
matchmaking.

(Of course, this isn't exclusively limited to DLs, in general. KR is 
tricky :))

You might take my Processes as Instances post as a starting point 
(though I've not put in any references). I started trying to do some 
funky and (I hope) clever stuff to get around the more obvious 
limitations. But they rely on various escape clauses  (and perhaps some 
decidedly non-standard reasoning services) in OWL DL.

One of our current moves in the DAML-S coalition is to give a 
(relatively) complete theory of the (or a) process model in full first 
order logic. Given this, it will be somewhat easier to see how one 
might map it, or parts of it, to various DLs (though, really, one might 
argue that there's little sense, practically speaking, in doing so for 
any DL that's not going to be a (future) extension of OWL; of course, 
such a process model might provide motive for such; again, it would be 
good if such a translation were *useful* in some clear way, and thus 
forming a natural subset).

Cheers,
Bijan Parsia.

Received on Tuesday, 2 September 2003 09:43:48 UTC