Re: Expressiveness of OWL

Jeremy,

Thanks for this.

BTW, my comment wasn't meant to be an implied criticism of OWL:  I 
understand that it was never aimed to be a full inference framework for all 
possible things, but I wanted to try an make sure I wasn't writing it off 
prematurely for certain kinds of inference.

Generally, I think it would be very helpful if there were some notes that 
indicated clearly what were the capabilities and limitations of OWL in this 
regard, particularly compared with Horn Clauses. (Maybe there is one that I 
don't know about?)  It's not always clear to me how to best use OWL's 
class-based reasoning.

#g
--

At 20:02 28/04/04 +0100, Jeremy Carroll wrote:
>Graham Klyne wrote:
>
>>I've been doing a bit more reading about Description Logics, and I come 
>>to a conclusion that I haven't noted stated anywhere in the OWL 
>>specifications...
>>(1) No "role composition" in OWL
>>The Description Logic literature makes mention of a "role constructor" 
>>called 'compose'.  I find no equivalent property in OWL;  I don't think 
>>this is surprising, as DL literature indicates that DLs containing 
>>'compose' are undecidable, but I'd like to be sure I'm not overlooking 
>>anything.
>
>Correct - IIRC the DL literature shows that compose over functional 
>properties is decidable (in the DL sense), but this is not support in OWL
>
>Also excluded were unions and intersections of properties.
>
>>(2) Limited expressiveness of OWL-based inferences
>>OWL expressiveness limited to FOL expressions with just monadic and 
>>dyadic predicates and no more than two variables.  Assuming absence of an 
>>equivalent to 'compose' in OWL, this result is noted in a couple of 
>>places, most clearly among those I surveyed in [1].
>>What does this all mean?  Thinking in terms of Horn Clause rules:
>>    A1,A2,...,An => B
>>or
>>    ~A1 \/ ~A2 \/ ... \/ ~An \/ B
>>(The results in [1] are based on conjunctive forms, but I am guessing 
>>that this is a dual result that can be obtained by renormalization.)
>>This suggests that OWL-based inference can handle a maximum of two 
>>variables between the antecedent and concequent of a rule, so something like:
>>    :a :parentOf :b .
>>    :b :parentOf :c .
>>=>
>>    :a :grandParentOf :c .
>>Is beyond the scope of an OWL based reasoner to infer.
>>Is this right, or am I missing something?
>
>that looks correct to me
>>#g
>
>
>Note procedurally what happened was not that these things were considered 
>and rejected but that our charter said start with DAML+OIL, D+O did not 
>have these things and noone suggested adding them.
>
>Jeremy

------------
Graham Klyne
For email:
http://www.ninebynine.org/#Contact

Received on Thursday, 29 April 2004 04:50:10 UTC