Re: Expressions in OWL-S

On Oct 12, 2006, at 7:56 PM, Drew McDermott wrote:

>> [j;AX Qn]
>>   I'm doing a research about semantic web service matching and
>>   composition, and I have noted that in the article "OWL-S: Semantic
>>   Markup for Web Services ", you have mentioned:
>>   "our approach is to treat expressions as literals, either string
>>   literals or XML literals. The latter case is used for languages
>>   whose standard encoding is in XML, such as SWRL [8] or RDF
>>   [11]. The former case is for other languages such as KIF [10] and
>>   PDDL [6]. "
>>   My question is: Is there other ways to describe the P.E
>>   "expressions" except SWRL, RDF, KIF and PDDL, e.g., using the
>>   Description Logic formula?
>
> No, not in general.  DL is a subset of first-order logic (FOL), so  
> if you
> need all of FOL you can't use it.

Neither SWRL nor RDF encompass all of FOL (of course you can use RDF  
to encode swrl sytnax), so that's not necessarily a problem. But  
since OWL has (a couple of) XML syntaxes, including RDF/XML, one  
would use the XMLiteral form. If you were using Turtle or Abstract  
Syntax...one would use string literals.

Of course, SWRL is a superset of OWL DL.

>>   I want to try this way because the DL reasoner is already made, so
>>   please tell me is this method feasible?
>
> There are reasoners for FOL, too, and there are other interesting
> subsets besides DL.  SWRL uses essentially Horn clauses, which means
> that a logic-programming system can process it with reasonable
> efficiency in many cases.

Well, sort of.

SWRL uses essentially OWL-DL as well, which means that an OWL-DL  
system can process it with reasonable efficiency in many cases.

> Even though DL reasoners come with some
> guarantees that other reasoning systems don't,

Well, if you compare to decidable formalisms of similar  
complexity...well, no :) I mean, Compare a SHIQ reasoner to a  
disjunctive datalog system...both decidable, both EXPTIME (I think).  
Both have optimizable algorithms (with different strengths).

(I would be remiss if I didn't mention KAON2 here:
	<http://kaon2.semanticweb.org/>)

> that doesn't always
> mean they're more efficient in particular problem areas.

That's, of course, true. In general, I'd be surprised if FOL  
reasoners would beat any of these tuned reasoners for decidable  
logics on the *same* problem, at least, as we scaled it up. It's  
possible of course, but unlikely I think. Interesting thing to  
investigate. For some work in this direction, see:
	<http://www.cs.man.ac.uk/~horrocks/Publications/download/2004/ 
TRBH04a.pdf>

You can go a long way by some preprocessing.

The real question here is exactly how one intends to use the DL to  
create preconditions and effects formula. I think I'd generally want  
conjunctive query like things, but I can imagine that there are cases  
where a class expression would do.

Cheers,
Bijan.

Received on Thursday, 12 October 2006 20:39:24 UTC