Re: PROV-ISSUE-347 (review-provo-luc): feedback from Luc on prov-o WD [prov-n]

Hi Luc,

I am trying to help Tim to address some of your feedback in this issue. 
A lot of your other feedback will be implemented during the editorial 
process separately.

 > Section 4 is arid, and not systematically handled. Suggestions below.
 > * Are the comments within the OWL file adequate to familiarize with 
the structure? If not, what kinds of comments would help?

This is a todo for us now. We will add more annotations to the OWL file 
rather than linking to the DM.

 > 11. Section 3.3:
 > choice of name: you have prov:qualifiedUsage, etc
 > why not simply prov:usage?

Because these properties are to be used for expressing "qualified 
relationships". I think with a qualitied- "prefix", it's easier for 
people to realize that this is a qualified property. Without it, 
prov:usage becomes vague and less straightforward, and people have to go 
and read documentation to understand what this is about.

What do you think?

 > 23. Role is defined in the context of usage, generation, association,
 > start and end, but hadRole has all involvments in its domain, 
including derivation and collection-derivations.

I think this has is also related to thread [1].

Such similar feedback have come several times. And I would like us to 
consider two additional things:

1) The prov-o is an OWL-RL ontology that implements DM. Given the set of 
semantics that can be expressed using OWL-RL, we couldn't have a 
one-to-one implementation of DM, just like a prov-xml schema cannot 
either. It's a compromised made for implementation.

2) Even if we do have a perfect prov-o implementation of DM, there is 
still nothing we can do to prevent people from using it wrongly or 
saying stupid things. The abuse of ontologies/vocabularies on the 
Semantic Web is nearly universal.

3) A possibility of saying something is different from being able to 
automatically infer that something. Our use of OWL-RL does provide a 
possibility for users to associate a derivation with a role, but the RL 
semantics would *not* infer that if something is a derivation, then it 
must have some sort of role. This is the best I can do for explanation 
and I hope I got it right:). A sensible user of prov-o, if read the spec 
properly, then would/should not have abused "hadRole".

See also Tim's original example in [2].

(prov:hadActivity rdfs:domain prov:Involvement .
:s prov:hadActivity :a
:s a prov:Involvement

this is different than saying:

:q a prov:Quotation
:q prov:hadActivity :activity)

The current implementation of prov-o is a compromise that we made so 
that we can use OWL-RL. A justification of OWL-RL is a separate issue. 
The prov-o team were strongly advised to use OWL-RL and that's what we 
did. I don't think you are *not sympathetic* of their situation and the 
problems that they had to deal with or had dealt with. :)

I suggest what we can do with this problem is in section 4, when we 
define the properties like "hadRole" or "hadActivity", we make it clear 
what/how we exactly expect people to use these properties.

Would that sound like a good compromise to you? If you do, then we will 
act upon that.



-- Jun

Received on Friday, 13 April 2012 16:16:56 UTC