Re: blank nodes (once again)

On Wed, Mar 16, 2011 at 10:10 PM, Pat Hayes <phayes@ihmc.us> wrote:
>
> On Mar 16, 2011, at 12:30 PM, Alan Ruttenberg wrote:
>
>> Are graph serializations assumed closed world?
>>
>> In other words, do graph names act the same way other semantic web names do?
>
> By 'graph name' you mean, the name of a (named) graph, right?

Right.

> (As opposed to a name used in a graph, ie some random URIref.)  Well, the answer ought to be as defined by the named graph paper that Jeremy and I co-wrote.

I'm looking at http://www.ra.ethz.ch/CDstore/www2005/docs/p613.pdf  -
that right?

> According to that, the way that a graph name is attached to its graph is like other web identifiers, it is determined by the http protocols.

?
I don't think I've seen spec that defines how http protocols attach
*any* name to anything, at least in the RDF world. (well, other than
this business of how you get a "representation" of the resource via
http. I'd hardly call that attachment)

> In our current RDF WG terminology, the name URI rigidly denotes the graph which is the g-snap of the g-box which the URI identifies.

I'm looking at http://www.w3.org/2011/rdf-wg/wiki/GraphConceptTerminology
now, right?

You are equating "named graph" with g-snap.
I read that page as saying there have been different interpretation of
what "graph" means.
Are you also, then, saying that there are no names for g-boxes? Or
that g-boxes are not "named graphs"?

Seems to me that all the SPARQL/Update stuff must have been talking
about g-boxes since g-snaps can't be "updated".
I thought they were talking about named graphs. Maybe others too...

> (There is, built into this definition, a presumption or requirement that this g-box is static, i.e. always delivers the same g-text when poked.

Yes, it's a set, I see.
I'm not clear that being a set implies that it is static, that word
associated with a particular view of time. For example, can one define
the set of readings of an temperature instrument throughout it's
lifetime, that lifetime not yet over? Is that set "static"?
OK. I see that "the contents of a g-box at any point in time are a
g-snap". Looks like you've chosen the continuant viewpoint ;-)

In any case, my concern isn't time, it's what's known.
In the paper above you say:

"Any assertions in RDF about the graph structure of Named Graphs are
understood to be referred to these graph extensions, just as the
meanings of the RDFS class vocabulary are referred to relationships
between the class extensions."

But class extensions may only be partially known - i.e. we make it
clear that a set of assertions about class membership can't be used to
determine class non-membership. It is this aspect that I find missing
in this talk about graphs and is what my question alluded to. Why is
it assumed that representations of graphs provide complete
information, whereas in our other semantic web standards we assume the
opposite?

>To handle the more general case, one would need to extend the RDF semantics to a temporal logic framework, with resources that are temporal functions and so on.)

As I say, I think that is a different matter.

> BTW, I wouldn't describe this as 'closed-world'. The correct term might be 'rigid': the name-->named relationship is presumed to hold across all interpretations, when describing the model theory.

Well, what I'm trying to address *is* the issue of open/closed world,
which if I understand correctly, is orthogonal to the issue of
rigidity.

I guess another way to discuss this might be to ask whether there
ought not be a fourth concept that should be added to
GraphConceptTerminology - a g-something whose contents are a *subset*
of some set of nodes and arcs.

Going back to the question of whether there are one or two graphs in

G1  :joe :has [a :dog]

G2: :joe :has [a :dog]

to which you answer (as I would) "2".

If "graph" is identified with g-box, then another reason to see these
as two is that their time evolution might be different.
If "graph" is identified with my g-something, then another reason to
see these as two is that the unstated triples that are nonetheless
members of graphs might be different.

--

OK, Pat, please fix me.

-Alan

Received on Thursday, 17 March 2011 03:54:28 UTC