RE: formal semantics strawman



> -----Original Message-----
> From: stian@mygrid.org.uk [mailto:stian@mygrid.org.uk] On Behalf Of Stian
> Soiland-Reyes
> Sent: Monday, September 26, 2011 7:41 AM
> To: James Cheney
> Cc: Myers, Jim; Graham Klyne; W3C provenance WG
> Subject: Re: formal semantics strawman
> 
> On Fri, Sep 23, 2011 at 16:38, James Cheney <jcheney@inf.ed.ac.uk> wrote:
> 
> > Thanks Stian for your detailed comments.  You seem to agree with my
> view on most of the questions where I thought Jim and I might disagree
> (e.g. that entities are statements about "real things")  So I am a little
> surprised that Jim agrees - I am not sure whether your and Jim's comments
> reflect a different view than mine that means we need to change the
> strawman, or simply saying the same things in different ways.
> 
> I am glad we are coming to an agreement. Perhaps it is just a simple matter
> of precision wording.
> 
> 
> > Instead of discussing the rest point by point, can I ask whether the
> following statements are controversial:
> >
> > 1.  Entity assertions (when written down as instances of the data model)
> describe facts about things that are true (or at least the asserter believes to
> be true).
> 
> Right.

Yes, but they actually resolve one of potentially many interpretations of the thing (interpretations that would agree at some instant but diverge over time).

Entity assertions represent a particular characterization of a thing that, through specification of fixed attributes, has a well-defined lifecycle with respect to the activities and events being reported as provenance.
 
> 
> 
> > 2.  Things have attributes that can change over time.
> 
> Yup. Or that *may* change over time.

Yes, but entities do too - everything we don't nail down with attributes. If 'things' are defined to identify how they can and can't change, they can be used as entities (perhaps - one can assert an entity representing the thing as-defined)

Whether attributes of a thing can or cannot change is often left ill-defined. Both things and entities have characteristics than can change. Entity assertions explicitly specify a relevant subset of characteristics (entity attributes) that cannot change.

> 
> 
> > 3.  Entity assertions describe attribute values that are fixed (and may be
> construed as identifying the thing) during the associated time interval.
> 
> Yes, except that we might also want to include other (informational) fixed
> attributes which are not meant to be identifying the thing. If we need to
> distinguish these is the "characterizing attributes"
> question.
> 
Yes, except entity assertions do not identify a thing - they identify/define one characterization of a thing because/when a thing is undefined as a function of time. As for which attributes are identifying - any fixed attribute is useful for discovery, whether or not we would consider that attribute to be part of identity or incidental to it. (While people may distinguish these two types of attributes, I don't know that they function differently in the provenance model...)

Entity assertions describe attribute values that are fixed (and may be construed as identifying a particular characterization of a thing).
  
> 
> It would not be very practical to assert attribute values which are not
> (observed/interpreted as) fixed within the interval observed/described by
> the asserter, so I think we can fairly ignore those. If needed to assert that
> "at some point the entity had this attribute value", they can be asserted on a
> narrower entity which wasComplementOf the original entity.
> 

Yes - an observation could be an entity defined at only one instant that is a complementOf the longer-lived one.

> 
> > 4.  Entity assertions have identities that allow us to refer to / link different
> assertions within the data model, but may or may not be related to globally
> meaningful URIs.
> 
> I agree. And where they do relate to globally meaningful URIs, then the
> attribute values given are not necessarily globally always as stated - they are
> only asserted to be such for this asserter within this provenance account.
> 
Yes, but entities as characterization of things are always globally meaningful - I could give them a URI and in fact there are URIs now that are well enough defined in terms of their temporal behavior that they can be used as entity identifiers.

> (The same asserter might later produce a different provenance account with
> different attribute values for same URIs)

If they do, the URI should be considered as having the union of those attributes fixed. (i.e. open world - you don't have to state all fixed attributes). If entity URIs are considered local in scope and reusable for different purposes, we have no change of integrating provenance accounts.

Entity assertions have identities that are unique and globally meaningful. However, the characterization of a thing that they represent may have limited utility outside of a specific provenance account.

> 
> 
> > If we agree on the above things, then I think the formal semantics
> strawman and data model reflect this common viewpoint as-is, and just
> needs to be updated to reflect the current data model draft and include
> illustrating examples.  If not, please suggest alternative statements that you
> do agree with (or changes to the semantics).
> 
> No, this sounds good to me. Jim?
> 

I've tried to give alternate statements above partly to see if you disagree with them - if you think they mean the same thing as the originals I would be OK with the earlier phrasings... If not, hopefully we'll get a clearer picture of what the differences are...

 Jim

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

Received on Monday, 26 September 2011 15:08:22 UTC