Re: Tim's approach on Involvement

On Tue, Feb 21, 2012 at 11:46, Daniel Garijo
<dgarijo@delicias.dia.fi.upm.es> wrote:

> Why the propery "tracedTo" is not a subproperty of "involved"? It relates an
> Element to an Element too.

Should be involved now.

> Do we need "involved"? What is its use appart from creating a hierarchy?

It shows that a binary relationship can also be expressed using
qualified and Involvement. However the link from individual properties
to that Involvement is not shown - it is by convention only.

> @Stian: I don't see how Start could be an EntityInvolvement, could you
> please explain? It is weird to have "End" under
> activityInvolvement and not "Start" :)

I know, I don't like it.

This is due to prov:Start appearing in both

http://www.w3.org/2011/prov/wiki/ProvRDF#Starting
http://www.w3.org/2011/prov/wiki/ProvRDF#Starting_again

however the second form does not talk about the agent. If I keep Start
under AgentInvolvement, then the activity-to-activity Starting_again
scenario would in OWL interpretation introduce a phantom agent.

So now I left it at the top - but that would allow you to make a Start
without either the agent and the activity. Perhaps this can be used
for qualifying prov:startedAt.


> I agree with Khalid in that "entity" and "activity" properties should be
> renamed. From a usability pov is confusing to use
> the same name for classes and properties (even if they are different URIs).

Yes - we agreed in general earlier to use verb forms in past tense.
The problem is to make the property name make sense for all the
subclasses. "hadQualifiedEntity" obviously did not work well.



> Are we going to delete the subproperties of "qualified"?

Yes, that was an important aspect of Tim's approach.


> Since qualified links an element to an Involvement, I think that we could
> use them wrong:
> :ent a prov:Entity;
>       prov:qualified [ a prov:Usage;
>                             prov:hadRole :role1]
> This is consistent according to our ontology (usage has some entities, but
> in the open world assumption it doesn't have to be asserted). Should we
> allow it?

There are lots of stupid things which are allowed by the OWL, for
instance making multiple inconsistent Generations, having
wasGeneratedAt times that don't overlap the startedAt/endedAt times of
the generating activity, etc. Checking those kind of things should not
be the job of the OWL ontology, but by a set a of rules.


However the ontology should guide the user towards the correct usage.
That's Tim's approach with the "hadSpatialExtent min 0 Location" kind
of subclasses - that gives a hint that it could have a location,
without requiring it.

However super-properties and super-classes make it look like you can
use them directly. It now looks like you can say:

:entity1 prov:qualified [
  a prov:EntityInvolvement;
  prov:entity :entity2;
  prov:hadTemporalExtent :t .
] .

- but this is a half-baked statement where you don't know if we're
talking about derivation, attribution or quotation. All  you can
conclude is :entity1 prov:involved :entity2.  Perhaps that's a useful
statement in a few applications, but for most parts it would be silly.


-- 
Stian Soiland-Reyes, myGrid team
School of Computer Science
The University of Manchester

Received on Tuesday, 21 February 2012 12:18:23 UTC