Re: PROV-ISSUE-153 (complementarity): Complementarity description differs from model definition [Primer]

Hi Paolo, comment below...

On 23/11/2011 09:56, Paolo Missier wrote:
>> Out of interest, do you have a use-case for which complementarity depends on
>> there being an actual overlap of attributes, as opposed to both being
>> contextualizations of some common thing?

> the old "royal society" example is meant to exemplify a pattern where each
> observer has a partial view on some system state (state = complete set of
> attribute value pairs), but there is no "common thing" that is given to them:
> nobody has the /complete/ state. Indeed, the idea is that the common thing
> emerges by taking the union of the two sets of attributes, on the basis that the
> overlapping portions are mutually consistent (i.e. a mapping can be established).
> We may be saying the same thing: a "common thing" that subsumes both exists, but
> in this example it only becomes manifested /as a consequence of/ the observers
> agreeing that they are each looking at two projections of it.
>
> Indeed your last comment seems to agree with this view: in an open world, there
> is some common entity that subsumes our views, but it may not have been explicated.
>
> The inspiration for this is the notion of "record linkage", or the process by
> which you "discover" such common entity, and you benefit from the discovery by
> taking all that you know from each of the individual pieces. I just would like
> to have this setting expressed as part of PROV because it's the only place where
> you can make an attempt at "joining up" or reconcile different assertions made
> independently about what is in reality the "common thing".
>
> I think what you are referring is complementary to this, namely you do have a a
> priori "common thing", you derive views from it, and you call them the
> complement of each other.

Yes, complementary (sic) indeed.

Maybe I'm missing something here, but absent knowledge of the existence of some 
a priori common thing (whether or not one knows what that thing is), it seems to 
me that there's very little one can reliably infer from some common attribute. 
Consider:

   Entity( e1, [membership=50] )

   Entity( e2, [membership=50] )

If it happens that e1 is a contextualization of "Royal Society", and e2 is a 
contextualization of the "Historical Society", I think the common attribute here 
tells us nothing of import regarding provenance.

(I should re-check PROV-DM on this, but I'm out of time right now.)

#g
--

> I see no conflicts here, I believe that both should be expressible.
>
> Regarding terminology, either "restriction" or "projection" work relative to the
> common thing, but they don't work in relation to each other. In any case, both
> are meant in their algebraic sense:
>
> restriction: "Any function can be restricted to asubset
> <http://en.wikipedia.org/wiki/Subset>of its domain. The restriction of/g/ : /A/
> ? /B/to/S/, where/S/?/A/, is written/g/ |_/S/ : /S/? /B/."
> (http://en.wikipedia.org/wiki/Function_restriction)
>
> projection: well, this we all know :-)
>
> --Paolo
>
>
>>
>> (Arguably, in an open-world environment such as the web, the fact that two
>> entities contextualize some common other entity suggests very strongly that
>> there does exist some attribute that is common, even if it has not been
>> explicated.)
>>
>> #g
>> --
>
>

Received on Wednesday, 23 November 2011 16:35:47 UTC