RE: abstract model

I've reorderd stuff in this response from the original:

> I'm not sure why you feel you need the _mapping_ between 
> statements and 
> reifications.

OK.  Maybe I don't - I'm not sure at the moment.  Its 2.45am
here - I'm trying to open the control panel to adjust my
body clock - but the damn thing is stuck.

I used the notion of a mapping because I felt it gave a clean
explanation of what I had in mind.  I didn't feel it was 
introducing a new concept incompatible with m&s.  For me
the mapping is just a convenient device to explain what
is going on.  Obviously this is not the case for all.  I feel
that this is something we can come back to if these ideas
ever begin to make it beyond this discussion.  An issue to
revisit.
 
> Consider, if you have an RDF model (per M&S section 5 formal 
> model) containing:
> 
>    [A] --type------> [Statement]
>    [A] --subject---> [S]
>    [A] --property--> [P]
>    [A] --object----> [O]
> 
>    [B] --type------> [Statement]
>    [B] --subject---> [S]
>    [B] --property--> [P]
>    [B] --object----> [O]
> 
>    [S] --P---------> [O]
>
> There is no way to know if [A] or [B] is the reification of 
> "[S] --P--> [O]".

This is a really interesting question.  I had written earlier
that there is a unique resource that represents the reification
of a statement.  Back to that decision in a minute.

In a sytem where the reification of a statement in unique, then
[A] and [B] above denote the same resource.  What if we gave them
URI's that are different?  No matter, both URI's denote the same
resource, assuming that a resource can have more than one URI.
If you say that a resource has only one URI, then you would have
an invalid model.

Earlier, I mention identifying properties, i.e. a property which
is defined to have a unique value for each resource.  What happens
if we add an identifying property to [A] and [B] above but with
different values.  Then again, we have an invalid model - i.e. a
 model that is consistent with the base RDF abstract model, but
violates other constraints.  Clearly such models could be created
without using reification.

I do think there is a choice about the uniqueness of a reifed
statement.  And right now, I don't have an argument to support
the choice of it being unique.  What I had in mind at the time
was that statements form a set, something I don't think, Graham,
if recall correctly, you agree with.  So that might be a root
of any differences in our understanding here.

 
> It seems plausible to me.  My only reservation is that it 
> introduces a new 
> type of symbol into the formal model (anon:S doesn't really 
> work as a URI, 
> I think).  But I think that may be handled in other ways 
> while retaining 
> your basic approach to partial reification.

Absolutely.  anon:S is not a URI, and it might well be better
to adopt a different syntax to make that clear.

Yes anon:S is a new type of symbol not in the original m&s
formal model.  I never said the formal model shouldn't change.
Clearly what I have suggested is different.  The issue is what
sort of changes are we allowing.  I was trying for something
that can be loosely described as a refinement - i.e. selecting
one of serveral possible interpretations of the original.
Expressing that interpretation does require introducing
new notation and terminology.

Brian


> -----Original Message-----
> From: Graham Klyne [mailto:GK@Dial.pipex.com]
> Sent: 14 September 2000 23:52
> To: McBride, Brian
> Cc: 'RDF Interest (E-mail)'
> Subject: RE: abstract model
> 
> 
> At 10:10 PM 9/14/00 +0100, McBride, Brian wrote:
> [...]
> > > I
> > > think I am
> > > forced to conclude that it represents an *extension* to the
> > > RDF model as
> > > defined.
> >
> >That may well be :(
> >
> > > Specifically, the "mapping called Reification" is a
> > > new concept
> > > that is not present in the original, and the only way to find
> > > that mapping
> > > using the original model is to find the set of 4 statements
> > > that comprise
> > > the reification.
> >
> >Please forgive my stupidity, but I don't follow the argument.
> >M&S does have the idea that a reified statement is represented
> >by a resource.  I think the mapping called Reification is a
> >component of the abstract model described by m&s.  I think it
> >can be positioned as a clarification of what is already
> >defined.
> 
> I agree that M&S has the concept of a reified statement 
> represented by a 
> resource.
> 
> My view is that the _mapping_ -- the function or relation 
> that associates 
> the resource representing a reified statement with the actual 
> statement it 
> represents -- is not there.
> 
> Consider, if you have an RDF model (per M&S section 5 formal 
> model) containing:
> 
>    [A] --type------> [Statement]
>    [A] --subject---> [S]
>    [A] --property--> [P]
>    [A] --object----> [O]
> 
>    [B] --type------> [Statement]
>    [B] --subject---> [S]
>    [B] --property--> [P]
>    [B] --object----> [O]
> 
>    [S] --P---------> [O]
> 
> There is no way to know if [A] or [B] is the reification of 
> "[S] --P--> [O]".
> 
> Thus, I suggest the formal model does not contain the 
> information needed to 
> construct the mapping you describe.
> 
> (However it could be said that either [A] or [B] reifies the 
> statement "[S] 
> --P--> [O]".)
> 
> >I suspect my problem is that we are thinking in different modes.
> >For me, as a mathematical abstraction it can work as defined.
> >I'm thinking of abstract model here, not of implementation.
> >
> >Maybe you could spell out the problem for me.
> 
> Let me also take a stab from a mathematical abstraction PoV.
> 
> IIRC, a mapping from some domain {d[i] in D} to some range 
> {r[j] in R} is 
> defined a set of pairs
> 
>    { <d[ik],r[jk]> }
> 
> that defines the mapping correspondence between members of D 
> and R.  It is 
> this set of pairs that is absent in the formal RDF model.
> 
> >Let me go through briefly the results of a conversation with a
> >colleague of mine.   Stuart asked me what would the semantics of
> >an incomplete reification be.  I think the answer hangs together
> >reasonably well and may be useful to this discussion.
> 
> Sure, I think it hangs together just fine.
> 
> [...]
> >If we apply the same thinking to reification, and I'm assuming
> >here that the resource modeling the reified statement is
> >anonymous, then given a set of statements like:
> >
> >    S = anon:S
> >    {
> >      (S, [rdf:type], [rdf:Statement]),
> >      (S, [rdfs:subject], [http://aldbaran.hpl.hp.com/bwm]),
> >      (S, [foo:assertedBy, "Graham")
> >    }
> >
> >This is a statement about an incomplete reification of a
> >statement.  What does it 'mean'.  Informally, using the same
> >reasoning as above, its semantics would be "Graham has
> >made a statement about my internal HP homepage", but it
> >doesn't say exactly what statement.
> >
> >That's the intiution.  What do you think?  If it stands up
> >to initial scrutiny, I'll maybe have a go at expressing
> >it in a more formal language.
> 
> It seems plausible to me.  My only reservation is that it 
> introduces a new 
> type of symbol into the formal model (anon:S doesn't really 
> work as a URI, 
> I think).  But I think that may be handled in other ways 
> while retaining 
> your basic approach to partial reification.
> 
> I'm not sure why you feel you need the _mapping_ between 
> statements and 
> reifications.
> 
> #g
> 
> ------------
> Graham Klyne
> (GK@ACM.ORG)
>

Received on Friday, 15 September 2000 06:18:10 UTC