Re: Do we loose anything? (discussion on ISSUE-131)

As I understand it, there's two aspects to the proposal:
	1) Make OWL-R syntactically distinguishable from OWL Full (that is,  
rule out some graphs)
	2) Bring OWL-R Full and OWL-R DL together (or as close as is feasible.

The first is criticial, the second is helpful (i.e., 1 fragment is  
easier than 2). Everything I see in your list is related to 2).  
Personally, if we have an OWL-R Full that is a bit more than OWL-R DL  
(with respect to builtin vocab, punning, etc.) that would be ok by me  
(assuming there are no other problems).

On 3 Jul 2008, at 10:19, Ivan Herman wrote:

> Boris, Zhe,
> thanks for the discussion on the OWL-R issue yesterday. I asked the  
> question at the meeting on whether we loose anything if we go along  
> Boris' proposal[1] and both of you said 'no'. However, I still need  
> some extra information to put my mind at ease... Here are two areas  
> where I feel some problems may arise (or where I hit the limits of  
> my understanding:-). There may be more...
> 1. Punning
> From an OWL-R-Full point of view, punning is of course not an  
> issue. However, the current state of OWL2 is that object/data  
> propery punning in DL is _not_ allowed. Doesn't that mean that, if  
> we go along your proposal, it would be disallowed in OWL-R (if one  
> wants to bind to the official profile) to use the same symbol for  
> data property _and_ object property? This may be considered as a  
> major restriction for OWL-R-Full users.

One point I'd like to raise here. I see you saying that such punning  
is a major feature (since not having it is a major restriction). I  
recall right after we removed object/data punning Michael gave a  
little example of owl-r that relied heavily on it to showcase a nice  

...given all this, shouldn't we consider adding back to OWL 2DL? I  
mean, we technically know how to do it...there's just some extra  
vocabulary (roughly).

> 2. Reserved vocabulary
> One thing we 'formally' loose, of course, is to use the RDF/RDFS  
> vocabulary in OWL constructs in OWL-R, too. For 99% of the cases I  
> do not really believe anybody in the community would really mind. I  
> do have two issues/questions, though, that we might want to look at.
> 2a. There was a long discussion[2] on what to do with lists,  
> sequences, etc. The way I read the discussion is that these  
> construct remain disallowed in DL. However, this means that they  
> will stay disallowed in OWL-R although the rule set would work  
> perfectly with those being used, too. This may be a _major_ loss of  
> functionality for OWL-R-Full users, which may be a deal breaker...
> 2b. The current rule set essentially says that, for example,  
> rdfs:subClassOf behaves as a transitive property. However, I am not  
> allowed to _declare_ (eg, in a set of axiomatic triples, which also  
> relates to issue-116[3], b.t.w.) that rdfs:subClassOf is of type  
> owl:TransitiveProperty. Eg, if one does a SPARQL query on the  
> dataset after all rules are executed, the query
> SELECT ?x WHERE { ?x a owl:TransitiveProperty }
> will _not_ return rdfs:subClassOf, although, well, rdfs:subClassOf  
> walks like a duck and quacks like a duck...

Just to pick a bit on this example...would you really want this in  
your answer? I mean, rdfs:subClassOf is a *builtin*  
"transitiveproperty". So including it in your answer set just adds  
noise. Indeed, SPARQL engines could build in that knowledge themselves.

> It is not entirely clear in my mind what an RDF user would expect  
> in this case, and we may very well decide that this is not a major  
> issue. But we should be clear in our mind that, well, this question  
> may come up!

It's worth considering. There's similar user issues about, e.g.,  
including owl:Thing or owl:Nothing in "ancestor/descendant" queries.  
In general, trivial theorems are a funny thing to deal with.  
Sometimes, some people want them excluded as noise; other times,  
other people want them for sanity checking and avoiding special  
casing. I don't think there is an ideal answer. I tend toward a bit  
of minimalism.

By and large, an OWL-R Full with property punning and syntax  
reflection would be, afaict, ok with me as long as the way its done  
actually has an appropriate effect on entailment. The thing about the  
current OWL-R Full that's unpalatable to me is that there's this  
silently ignoring of parts of the graph. When the relationship is  
simple (simple entailment to rdf entailment), it's not such a big  
deal. When the relationship is complex (i.e., you aren't just  
ignoring triples, but triples *in certain patterns*) I think it's a  



Received on Thursday, 3 July 2008 11:55:20 UTC