Re: [ratholes, reification, risk] poison-URIref testcases

On Mon, 25 Mar 2002, Graham Klyne wrote:

> At 12:29 PM 3/25/02 +0000, Jan Grant wrote:
> >This is, surprise, going down a rathole fast, so this is my last pulic
> >post on the topic...

don't say I didn't warn you :-/

I don't want to slow down RDF Core process, but do want to get to the
bottom of this. Maybe www-rdf-interest or -logic unless we suddenly all
agree on something?

Any further discussion here should probably be couched in terms of
proposed changes to our Working Draft text, and associated with an open
issue. The closest I can find in the Issue Tracking document is
http://www.w3.org/2000/03/rdf-tracking/#rdfms-assertion

See below for a draft addition to the MT spec.

> >A better bet (I think) is to just do something Herbrandish - that is,
> >keep the current definition of "interpretation"; if you _do_ wind up
> >with one of Dan's problematical graphs (that is, one with a URI-labelled
> >node with no "meaning", ie, which doesn't denote anything*), you don't
> >really hurt yourself by letting it denote some mathematical figment that
> >doesn't collide with anything else.
>
> That sounds very like what I was trying to say.

I think we have some notion of a "Web-sanctioned interpretation"
underlying our thought on this. There are certainly
interpretations (alice-in-wonderland ones, for eg.) where poison-URIs
denote. The concern is that there may be no Web-sanctioned ones, ie.
when we buy into the global meaning of the urirefs, we waive our
right to make up arbitrary mappings. The MT does acknowledge that
urirefs have global meaning imposed by our shared Web understanding
of these node labels.

I would be happy with the Herbrandish workaround, so long as there
could be Web-sanctioned interpretations of a 'poisoned' graph. The
Web-sanctioned interpretations of a graph, as I introduce the phrase,
are those that are respectful of the public meaning of each uriref.
For those, we must defer to specs beyond RDF's (eg. URI2396, TAG work
etc.).

There are perhaps 'hybrid' interpretations in which 'good' URIs denote
things in the world (fixed by Web/TAG conventions for meaning of urirefs), while
'poison' ones denote (some figment of our choice). I'm nor sure that
is a very clear partition, since it is those same Web/TAG/2396 conventions
that determine whether a uriref has public meaning. It might be good enough.
Below is an attempt at characterising such a 'hybrid' interpretation
(without imposing it). I don't think the MT can go any further than that.

We should be wary of trying to make the MT do work that MTs aren't
designed for. But we should also be wary of leaving a gap in our account
of the meaning of RDF/XML documents.

Does the following work for anyone?

[[
Note: urirefs, as referring expressions, have their meaning fixed through
a number of social, legal and technical mechanisms. The RDF MT does not
itself impose any particular interpretation on an RDF graph. An
interpretation for an RDF graph in which the denotation of uriref labels
is fixed by the public meaning of urirefs is termed a <em>Web-sanctioned
interpretation</em>. It is beyond the scope of the MT to provide a
detailed discussion of Web-sanctioned interpretations. At this time, it is
not clear whether the Web's use of uriref identifier syntax allows us to
assume that all urirefs denote. We do not know whether the Web
architecture guarantees that every RDF graph has a <em>Web-sanctioned
satisfying interpretation</em>. An interpretation for an RDF graph in
which the denotation of uriref labels is fixed by the public meaning of
urirefs, augmented by the assumption that each uriref denotes, is termed
an <em>augmented Web-sanctioned interpretation</em>. This notion
allows us to discuss the ascription of propositional content to RDF graphs
that contain nodes and edges labeled with urirefs, without concern for
whether a public ("Web") meaning is assigned for each uriref. Without making this
additional assumption, there may be RDF graphs whose (publically
meaningless) uriref labels ensure that there are no satisfying
<em>Web-sanctioned</em> interpretions for the graph, because nodes are
labeled with non-referring urirefs. This notion is particularly relevant
to Web applications that exchange RDF graphs employing the RDF Core
reification vocabulary (ie. Statement/predicate/subject/object)
since nodes will often be labelled with uriref data from untrusted
sources.
]]

Or we could leave this for version 2... I was just trying to draft a
'health warning' flag to stick outside the rathole... (and something we
could point the TAG or URI CG at).

Dan

Received on Monday, 25 March 2002 09:05:42 UTC