Re: [PROV-O] Proposed OWL change: Dealing with Issue 568 (hadRole)

On Oct 19, 2012, at 11:40 AM, Paul Groth <p.t.groth@vu.nl> wrote:

> Hi Stephan,
> 
> Here's the concrete issue: http://www.w3.org/2011/prov/track/issues/568
> 
> It seems the question is, do want want to make the inference you outlined? 
> 
> I agree that all the relations that are allowed to have a role are influence but dm specifically doesn't list influence as something you can apply a role to. Thus, this is something you probably don't want to explicitly state.

I think right now the best argument for removing Influence from the domain is the confusion that it is causing.

From a modeling perspective I believe it is consistent with the DM, but confusion about the semantics of property domains is causing a great deal of stumbling on this.

We aren't saying that all influence relations can have a role, just that any relation (DM term) that has a role can be inferred to be an influence relation (which I believe is consistent with the DM text through inheritance of the relation 'type').

I think the issue here is trying to get the most reasoning possible while in the RL restriction.  Since Influence is the most specific RL-compatible super-class that covers all the role-able classes, that is the most detailed domain we can set that an RL reasoner will act upon.

I guess at this point I am ok with removing Influence from the domain, but I would argue that the current modeling is consistent with the DM.  We should make the change because the modeling causes more user confusion than the benefit of the inference.

--Stephan

> 
> cheers
> Paul
> 
> 
> 
> 
> On Fri, Oct 19, 2012 at 6:21 PM, Stephan Zednik <zednis@rpi.edu> wrote:
> 
> On Oct 19, 2012, at 10:03 AM, Daniel Garijo <dgarijo@delicias.dia.fi.upm.es> wrote:
> 
>> Thanks Stephan, you are right. However the current problem is that it is not consistent with DM.
> 
> I think it is worthwhile to remember what a property domain in RDFS implies.  
> 
> rdfs:domain is an instance of rdf:Property that is used to state that any resource that has a given property is an instance of one or more classes.
> 
> With the currently modeling, a DL reasoner will infer that the subject is an instance of the union class, and a RL reasoner will infer only that the subject is an instance of Influence.  Since all of the classes in the union class are specializations of Influence, the RL inference is not incorrect or inconsistent with the DM, it is just not as precise as the DL inference.
> 
> An RL reasoner 
> 
> :ex prov:hadRole [ a prov:Role;  prov:label "example role"; ] .
> 
> Will infer the following statement
> 
> :ex rdf:type prov:Influence .
> 
> Which I do not believe is inconsistent with the DM.
> 
> --Stephan
> 
>> 
>> I have been looking further, and there are other properties where we have just
>> a union in the domain (e.g., qualifiedInfluence, wasInfluencedBy, atLocation). In 
>> these cases the properties would have an empty domain in DL. I think that it's better
>> to have it empty rather than allow inconsitencies with the DM.
>> 
>> Thus I still propose to make the change to the documents. Thoughts?
>> Best,
>> Daniel
>> 
>> 2012/10/19 Stephan Zednik <zednis@rpi.edu>
>> Looking at the domain of hadRole again, I believe what we have right now is the result of the RL++ compromise.  The current domain in DL would be the intersection of prov:Influence and the union of prov:Association and prov:InstantaneousEvent, which equates to just the union of prov:Association and prov:InstantaneousEvent.  In RL, the union is ignored so the domain would be recognized as prov:Influence.  There was no way to get the domain aligned with the DM under RL, so adding Influence was a fallback, otherwise the domain would be unspecified.
>> 
>> That is at least my recollection of why it is as it currently is.
>> 
>> --Stephan
>> 
>> On Oct 19, 2012, at 7:49 AM, Daniel Garijo <dgarijo@delicias.dia.fi.upm.es> wrote:
>> 
>>> Prov-o team:
>>> there seems to be a bug in the ontology, which Luc highlighted in the last telecon:
>>> 
>>> prov:Influence is listed as domain of prov:hadRole, and this is not compatible
>>> with PROV-DM. I have checked the latest documents and the only changes to do are:
>>> Remove prov:Inflluence from the domain of prov:hadRole in the ontology.
>>> Remove prov:Influence from the domain of prov:hadRole in the Overview.html document.
>>> Remove prov:hadRole in the "described with properties" box in Overview.html
>>> If nobody disagrees with these changes, I will commit them on Monday.
>>> 
>>> Best,
>>> 
>>> Daniel
>> 
>> 
> 
> 
> 
> 
> -- 
> --
> Dr. Paul Groth (p.t.groth@vu.nl)
> http://www.few.vu.nl/~pgroth/
> Assistant Professor
> - Knowledge Representation & Reasoning Group | 
>   Artificial Intelligence Section | Department of Computer Science
> - The Network Institute
> VU University Amsterdam