Re: Named graphs etc

Ok, after more thinking.

Clearly some sort of vocabulary solution would be the least painful 
route to take, along the lines Jeremy suggests.  We have to face up 
to the complications it introduces, though.  For example, what if an 
agent wants to change their mind about G, but there's an assertion 
out there that says that they are asserting G? What's to stop a 
malicious publication of assertion that A asserts P by some other 
agent B? What if A asserts G  but then B says that G is deprecated? 
I think that we have to assume that there is a clear notion of the 
agent-thing doing the asserting, and that any asserted graph has an 
identifiable 'owner' who is the agent asserting it. I'm not sure how 
we can guarantee that this is always possible, so I'll leave that on 
the table as an issue.

How to do it, is the question. It can't be done simply by tweaking 
truth-conditions (my previous gripe), so OK, we do it some other way. 
It will be a rather unusual semantic extension, but still a semantic 
extension.

Lets hypothesize a class of 'RDF agents'. We don't need to say what 
they are exactly (RFC 2396 doesn't say what a 'resource' is exactly) 
but we suppose that they are resources which can perform 'web acts' 
with respect to RDF graphs. Web acts are 'speech acts on the Web': 
they include minimally

asserting
quoting

and likely others not yet specified. Every assertion is made by an 
agent. Given any record of an assertion, we have to be able to figure 
out the agent of the assertion.

Now, how does this relate to triples in graphs?? One view (1) (the 
one I've been using, implicitly) is that a web act is somehow 
connected with the publication of a graph: that the act happens when 
you hit the 'upload' button and something gets written on your web 
server.  On this view, a triple in a graph would at best be a kind of 
noting or recording or a statement ABOUT the act, and the asserting 
agent is the owner of the web resource where the (original) graph is 
located.  Another view (2), which I think Patrick is using, is that 
the 'act' is performed by the graph itself: that to assert *is* to 
have a triple of a certain kind in a graph.  He wants to be sure that 
graphs can only assert themselves in this authoritative way, which 
makes sense on this view. Also, on this view, as long as asserting 
triples can be firmly anchored in named graphs, the 
locating-the-asserter problem is solved.

However, I don't see how we can firmly locate triples in graphs in 
this way, because how can we possibly legislate against people 
copying triples from one graph to another? And this mixes up several 
things, such as the propositional nature of RDF and the distinction 
between a graph and a document containing the graph, that seem to be 
fundamental; and it muddies up RDF with things that look like 
locatives or indexicals ('this' or 'here') .

So let me revert to the idea of a graph being a description of 
something, and ask what it would be describing in this case. We could 
say that a triple

ex:agent rdfx:asserts ex:graph1

is true when the 'state of the Web' is such that ex:agent is indeed 
asserting, or has asserted and has not yet un-asserted, that graph. 
So this can come to be true and false depending on the state of the 
Web, which gives a model theorist nightmares but is something we have 
to face up to.

OK so far, but this still only has the graph describing - reporting 
on - some event of asserting, rather than actually making anything 
happen. So we need a kind of web act rule (compare Grice's rules of 
conversational implicature :-) which says that if an agent publishes 
a graph which describes a web act by that very agent, then the act is 
deemed to have been done by that publication. That is, you cannot 
tell RDF lies about your own web acts. (Cannot as in 'its impossible' 
not as in 'its naughty')  Just like in English (actually, more like 
in civilized English-speaking society) you cannot say "I promise..." 
without actually promising. I take it that this is what Jeremy meant 
about English doing the speech acts. However, if you say "He 
promises..." then unless you are empowered by him to to speak for 
him, eg by a power of attorney, then nobody has promised anything: 
which is why we need the link between the agent responsible for the 
triple, and the agent referred to by the triple. The weakness of this 
whole story is then how to establish that link between a triple and 
the agent responsible for it, so people can tell genuine assertings 
from mere descriptions of assertings. I don't have any good answer to 
that, but maybe we don't need to give a good answer to it, if we 
state the issue clearly. (Is there a good answer to the question "Who 
owns a web resource?")

This story could be applied to either (1) or (2) above, but if with 
(1) there is no particular need to have the assertion be an assertion 
of the graph it is in. What matters is that it is an assertion BY the 
agent it says it is by.  Its the domain rather than the range of 
rdfx:asserts that needs to be anchored in some kind of checkable 
reality.

One problem seems to me to be, who is responsible for stating a 
general Gricean 'web act rule' like this? (Grice was the first person 
to describe these rules for English, but he didnt impose them, they 
seem to have just kind of happened.) It goes beyond the RDF or OWL 
semantics, which don't say anything about web actions or agency. I 
don't see the TAG group doing anything this formal-sounding in the 
forseeable future, and I don't see that we have the authority to do 
it. Maybe all we can do is to suggest that this might be a good way 
to behave and hope that the world in general listens (?)

Anyway, I'll send this off and see what happens :-)

Pat



-- 
---------------------------------------------------------------------
IHMC	(850)434 8903 or (650)494 3973   home
40 South Alcaniz St.	(850)202 4416   office
Pensacola			(850)202 4440   fax
FL 32501			(850)291 0667    cell
phayes@ihmc.us       http://www.ihmc.us/users/phayes

Received on Wednesday, 10 March 2004 18:31:32 UTC