Re: httpRange-14

On Saturday, Aug 2, 2003, at 07:33 US/Eastern, Bill de hÓra wrote:

> Tim Berners-Lee wrote:
>
> Tim,
>
>> Sorry, but the semantic web architecture absolutely needs the idea
>> of information resources. The RDF identifier foo#bar is
>> used by dereferencing foo and parsing it a get information.
>
> Is it? I'm sure there are fragids out there that are being used as 
> just proper names.

I am sure that is the case, but that doesn't stop it being a good thing 
to be able to dereference them.

> And when it comes to machine reasoning, that is /all/ they are. The 
> RDF MT explicitly does not care abut URI structure. You can't carry 
> over that implication without going beyond the RDF MT. An RDF machine 
> that did this would be embracing and extending RDF semantics in much 
> the same way cwm does.

An "RDF machine" is presumably one which draws inferences in ways which 
have been authorized in the RDF spec.  A "URI machine" is one which 
makes inferences based on what it find in the URI spec.
A very interesting machine is composed of both.

You can't use RDF without committing to the RDF spec, and the RDF spec 
can't use URIs without commiting to the URI spec.

So yes, a "pure" RDF machine can b conceived of which only does RDF-MT 
allowed operations.
The semantic web, though uses URIs for a reason.
That's why cwm mixed an RDF machine and a URI machine (which includes 
an HTTP machine).

> To state, in a machine reasonable way that a URI denotes an 
> information resource rather a non-information resource requires an 
> ontology of resources - URI indexicals are not sufficient.

Yes.   (What are the domain and range of owl:imports, for example?)

> If you are going to mandate a privileged status for URIS with fragids 
> in the semantic web, then before the URI is forwarded to an RDF MT 
> compliant reasoner that URI needs to be annonated with new triples 
> that claims it is an information type resource. So, you /still/ need 
> the resource ontology.

No, it does not *have* to be.  Just as is an imaginary isbn: URI 
scheme, you wouldn't *have* to add code to decide that anything with an 
ISBN is a book.  And you may not want to clutter your kb with these 
these conclusions of the URI machine.  Decisions about when to use say 
your rule machine and when to use your web machine (when to look for 
inference from the local kb and when to go look on the web for more 
data will be the source of much research  and many products for a long 
time to come.

> This automagic must be stated explicitly in the architecture, as being 
> automagic.

"Magic is anything you don't understand" (I have been known to tell 
kids).
Inference can be made based on the various specs.
Inference which is made using URI and HTTP specs (as well as RDF) is 
"magic" to RDF.
But then, inference which is made by OWL is "magic" to a URI machine.

> That's not to say it isn't useful, people hack at URI structures every 
> day to great effect, but as things stand it's beyond RDF and any 
> semantic web MT I've seen.

An MT which includes URIs and HTTP is an interesting idea.

> And I suspect this opens a Pandora's box of hacks, optimizations, 
> localizations, my-need-is-special, non-portable extensions and 
> presumptions to the semantics that involve frigging with URIs.

We are *not* talking hacks, here, except in as much as the whole WWW is 
a great big hack.
There is a well defined set of protocols which define what you can do 
with and conclude about URIs without hashes.  There is the rule that 
the language of a document defines the syntax and semantics of URIs 
which are the document's URI followed by a hash and some local 
identifier.  That is the architectural hook which actually allows RDF 
identifiers with hashes


> You name it. This is since I don't see why we'd would stop carving 
> things up at information resources, if that turned out to be useful.
>
> If so, where this heuristic annotating is done needs to be clearly 
> outlined. Is there a layer for it?

In the block diagram you will see "URI" and "HTTP" mentioned.
What we have not done is formalize it in MT or in OWL. yet.

You seem to feel that this is the small end of a wedge.
It doesn't have to be.
Cwm has a simple formalization of the whole web using just 
log:semantics, a property of an information resource which gives an RDF 
graph (or N3 formula) obtained by dereferencing the document's URI on 
the web. It hides all the intermediate steps.

OWL has, I suppose, bitten off this by getting into owl:imports, which 
has some semantics along the lines of the log:semantics of the current 
document is a Truth, then the log:semantics of the imported document is 
a Truth.


> The more I think about it, the more this seems like a way of avoiding 
> defining a new URI scheme for the semantic web by finessing the 
> opacity axiom.

Don't define a new URI scheme for the semantic web!  The whole point is 
to build on the existing web, the information space, as a foundation.  
We know how to distribute information.  We can't make the semantic web 
work without doing it. The semantic web is defined as being built on 
the web.

The opacity axiom is sometimes misunderstood as "URIs have no meaning". 
It should be "URIs have no meaning apart from that given them by the 
web specifications".

Tim

PS: See  <http://www.w3.org/DesignIssues/Stack> for an article about 
how all the specs hang together.


>
> Bill de hÓra
>

Received on Wednesday, 6 August 2003 11:20:09 UTC