Re: [TF-ENT] OWL Direct Semantics added

On 24 Nov 2009, at 17:05, Birte Glimm wrote:

> Thanks for the comments Ivan.
>
> 2009/11/24 Ivan Herman <ivan@w3.org>:
>> Hi Birte,
>>
>> I share your unhappiness:-)... and I am wondering. I am not sure we
>> discussed how a user would choose among the various entailment  
>> regimes a
>> system provides (maybe different URI-s correspond to different
>> regimes?). Also, I am not sure about a conformance issue: would all
>> SPARQL implementation have to implement simple entailment as a  
>> minimum?
>>
>> However... let us suppose that (a) each system has simple  
>> entailment as
>> a possibility

*Systems*, if by which we mean query engines, don't need to offer this.

>> and (b) the user can choose which entailment is used for a
>> specific query.

The user can always do this by switching systems.

>> Do we then really need this mixed semantics?

I think so.

>> What are
>> the use cases? After all, the user can then choose to run simple
>> entailment for queries on annotations...

It's unnatural, on the one hand, and doesn't handle a bunch of  
reasonable cases.

> For (a) I don't think simple entailment is required, at least I don't
> read that from the spec.

I agree.

> And, if the spec doesn't force me to I would
> most likely not extend my reasoner to work with simple entailment. We
> do not even get to see any triples since we parse with the OWL API and
> work only with the OWL API objects that reflect the ontology
> structures as also found in FSS. Pellet so far also has no support for
> simple entailment I think and it requires quite different algorithms
> and data structures than (hyper)tableau algorithms commonly ued for
> OWL Direct Semantics (DS) reasoning.

I think that simple entailment can be handled entirely at the OWL API  
level, so I'd be surprised if HermiT or Pellet couldn't qua command  
line tools, pretty easily handle simply entailed queries.

But then the combined semantics is both straightforward to implement  
and better for the users.

> The spec now requires for OWL DS
> that the systems have to do some simple entailment, but you would only
> have to work with those triples that represent non-logical things.
> Thus, you can get away with an unoptimised simple implementation since
> only few triples have to be treated with simple semantics. Otherwise,
> it is probably better to use a third-party triple store for simple
> entailment.
[snip]

I wonder if speccing things in terms of triples and simple entailment  
at this level is the most felicitous way for OWL 2 DL semantics.  
Alternatively, we could spec the annotation handling directly in  
terms of OWL 2 FS style structures.

For example, c. could be replaced with:

	c. for each axiom ax in Ax such that ax is non-logical or ax  
contains an annotation, ax is in the axiom closure of O, and

This might not work if we consider treating the annotation  
subproperty etc. vocabulary with semantic force (as I already know  
users want...). So perhaps this is only implementation advice.

So, I'm not sure what's unhappy about it :)

Cheers,
Bijan.

Received on Tuesday, 24 November 2009 17:35:48 UTC