RE: A possible way of going forward with OWL-R unification (ISSUE-131)

Zhe Wu wrote:

>For the record, I am happy with the way OWL-R is written. Like I said in
>the conference call, I haven't heard any
>complaints about it myself. Frankly, the rule set in OWL R Full is what
>I care the most.
>I will be happy if Boris's proposal is accepted, as long as we clearly
>point out that the rules can be applied to any
>RDF graphs. The OWL-R unification idea can reduce a bit confusion. I
>don't consider it a must-have though.
>
>In short, I don't have strong opinions either way.
>
>Thanks,
>
>Zhe

Here is my opinion.

For me, a crucial point is whether OWL-R, when applied in the "relaxed mode", 
i.e. as a triple rule language on arbitrary RDF graphs, will produce *every* 
RDFS entailment.

If OWL-R/relaxed turns out to not be a proper RDFS extension (an "enhanced 
RDFS", "RDFS plus a bit", "RDFS 3.0", or whatever you like to call it), then I 
would regard OWL-R to be pretty pointless from an RDF community point of view. 
It would probably be hard to explain to RDF people, why particular 
RDFS-features are missing, while others are supported. It would certainly give 
to RDF people a warmer feeling to learn that an OWL-R reasoner will produce 
all the results of an RDFS reasoner, if applied to RDF graphs, which don't 
contain any of the additional OWL vocabulary.

Also, having OWL R as an RDFS-extension would probably ease the production of 
prototypical implementations for OWL R. For example, in order to create an OWL 
R Full reasoner, as defined at the moment, it is as simple as taking Jena's 
RDFS reasoner, and adding the additional rules as given by the OWL R Full 
spec. That is a matter of a few hours, I suppose.

However, I do *not* believe that OWL R is rejected by the RDF community, if 
OWL R does not extend RDFS. I would rather expect that in such a case RDF 
people will just start to create such RDFS-extending implementations of OWL 
R -- simply because it's easier and more straightforward for them to do so. 
But then, we would be in the undesirable situation that the "typical" OWL R 
implementation would not conform to the W3C spec. In fact, such 
implementations would be OWL-R *non-sound*, since they would produce 
additional inferences, which are not entailments in the official OWL R/relaxed 
language.

So in order to be allowed to put the "W3C OWL-R" brand on their product, 
implementers would need to create an additional "strict OWL/R mode" of their 
reasoner. This would demand additional effort for something, which is unlikely 
to be applied by many people, simply because the RDFS-extending OWL-R 
implementation is more capable than the official one. On the other hand, it is 
even more unlikely that implementers will /exclusively/ produce "strict mode" 
versions, because then they would be in a disadvantage compared to their 
competitors.

In such a situation, I wouldn't be much surprised if RDF people (both 
implementers and users) soon start to claim that the W3C OWL-R spec is broken. 
:(

Best regards,
Michael