RE: complementOf -> viewOf: proposed text



> -----Original Message-----
> From: stian@mygrid.org.uk [mailto:stian@mygrid.org.uk] On Behalf Of Stian
> Soiland-Reyes
> Sent: Thursday, January 19, 2012 9:08 AM
> To: Myers, Jim
> Cc: Luc Moreau; Paolo Missier; Paolo Missier; public-prov-wg@w3.org
> Subject: Re: complementOf -> viewOf: proposed text
> 
> Thanks for a great summary!
> 
> On Thu, Jan 19, 2012 at 13:29, Myers, Jim <MYERSJ4@rpi.edu> wrote:
> 
> 
> > Alternate is so hard to define because we're are having to define
> > entities in terms of the things that people have named, which are not
> > prov:things - they are prov:things that have some characterization in
> > terms of some theory of the world.  (When fully characterized with
> > respect to what the asserter wants to report, we have entities.)
> 
> Agree - although I would argue that they are entities no matter how 'fully'
> they are characterised. As long as the asserter has some kind of implicit or
> explicit idea about what the entity is and is willing to use them in the
> provenance account, then they are characterised. It could be as little implicit
> as simply entity(paolo).

I agree - there's an open-world sense that the characterization is all implicit - the only constraint is that saying e1 = entity(paolo) means that one characterization has been chosen and all statements about e1 should be consistent with that characterization.

> 
> 
> > So two choices (could we vote?):
> 
> At least informally..?
> 
> 
> > Alternative applies only between two entities that characterize the same
> prov:thing over the same interval and is transitive: to use it in the customer-
> in-chair case you have to create specializations of both poalo and customer in
> chair.
> 
> So in my silly usecase, there would be a paoloInChair (specialization of paolo)
> and customerInChairFrom5pmTo6pm (specialisation of customerInChair),
> and these are alternateOf each other (but don't have a common
> specialisation).
> 
Right - the asserter has chosen to define these specialized entities in terms of two different colloquial things (person, customerinchair) and thus hiding the fact that paolo-in-chair and customer-in-chair-5-6PM do indeed refer to the same prov:thing - alternateOf asserts that otherwise hidden relationship.
 
> Does it mean their generation and destruction time have to be the same for
> alternates?

For this case, I think the answer is yes (we might be able to formulate it in terms of events rather than times, but yes). Paolo-in-café is an example of the issue - presumably there was a time paolo got his favorite drink where he was not in the chair and so there's an interval when transitivity won't work (Stian was in the chair while paolo got his drink).

> 
> If not, then what does 'same interval' mean in alternativeOf(A,B)?
> Lifetime of A is fully contained by lifetime of B? Or just that they
> characterized the same thing during the complete time they overlapped?

This is really the other choice I described below - pseudotransitivity works. I think what you're pointing out is that one way to specify the interval of the alternateOf relationship is to make it be the full time of overlap of the two entity's lifetimes. If this is of interest (we think there are enough cases and not having to explicitly define the interval of the relationship is easier) I think it could be added to option 2 - alternativeOf applies for an interval and, if not specified, it defaults to the intersection of the lifetimes).

> 
> 
> > Alternative applies between entities that at some point in their lifecycles
> describe the same prov:thing. Transitivity cannot be applied. Having a
> time/event interval specified on the relationship would allow inference that
> there are specializations of both entities that are aliases (symmetric
> specializations of each other?) Pseudotransitivity in this way could be a
> separate vote...
> 
> -1 - this is the old ivpOf with its confusing interval overlap. This easily covers
> the customerInRedChair example, but as the overlap (where the
> characterisation is true) can be anything from tiny to fully contained it is not
> that useful concept.


?? Not sure what the -1 applies to - I would agree that this option without an interval (I'd like to have an explicit option, but having a default of the intersection of lifetimes works too) is not as useful for the reason you state. I think this option is better than the first choice iff we have a defined interval. I would still make the question of whether transitivity within that interval makes sense as a separate vote simply because I can imagine the possibility that with a series of different-enough colloquial things we could construct an example where we'd get a non-intuitive result (due to a mismatch in implicit parts of entity characterizations, e.g. things like paolo having a different mass than paolo and his coffee cup which is really what we meant by customer in chair may not be important to start, but if there's a long inference chain, unexpected consequences may arise - we might be better off not allowing pseudotransitivity by definition and leaving it as a domain judgment call - pseudotransitivity- transitivity in the interval - works if there are no mismatches in how the entities involved are characterized and only the user can decide. Said another way, pseudotransitivity MAY (I don’t know) depend on entities being more fully characterized than what was required for the direct assertions and therefore we chose not to make it automatic...)

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

Received on Thursday, 19 January 2012 16:03:04 UTC