"On the Web" vs "On the Semantic Web" (was Re: resources and URIs)

Jonathan Borden wrote:

> > > If we are going to concern the "SW" at least as it is incarnated in
current
> > activities and specific software products (i.e. RDF and OWL related
> > software), then you may certainly be a part of the Web without a URI.
Hence
> > the statement is not correct in the SW context -- I'm not entirely
convinced
> > that it is correct in the 'current Web' context e.g.

To which Tim Bray responded:

> This might be a nice clear clean differentiating principle, because I'm
> pretty convinced that at the moment, something that doesn't have a URI
> isn't part of the Web.

I think Tim's distinction is valid. A resource not denoted by a URI is
not "on the web".

The Web and the Semantic Web are (IMO) two distinct things which
intersect via a common set of URIs with (presumed) consistent
denotations.

Here's how I've been viewing the relationship between the Web
and Semantic Web:

The Web is a network of linked representations of resources, where
those representations are accessible via URIs. Thus it is quite correct,
I think, to say that a resource that is not denoted by a URI is not "on
the web". In fact, I think an even tighter claim can be made -- that a
resource that is not denoted by a URI that is meaningful to the HTTP
protocol and does not intentionally** resolve to at least one representation
is not "on the web".

** barring practical technical problems such as system being offline,
     routing updates/problems, etc. etc. i.e. the authority for the URI
     intends that the URI will reliably resolve to a representation.

Thus, a URI that consistently and expectedly always produces a 404 error
does not denote a resource that is "on the web", even if the URI is used
in some non-web fashion to actually denote some resource.

And here's an important point, that addresses concerns expressed by
Pat and Jonathan:  a resource that is "on the web" is not (necessarily)
itself part of that physically realized network of interlinked
representations
managed by web servers. Nor is every representation necessarily
"on the web" (if it is not unambiguously denoted by its own URI).

Being "on the web" simply means that the resource in question is
denoted by a URI which can be resolved via HTTP to a representation.
And that representation may contain references to other resources
which may resolve to other representations, etc.

The only resources which are both "on the web" and part of that physical
network of representations are representations which are denoted by URIs
which resolve to a bit-equal copy of themselves.

One can say that there are two realizations or facets or dimensions
of the Web:

   (1) the physical network of interlinked representations managed by
         HTTP servers

   (2) the abstract network of interrelated resources having interlinked
        representations managed by HTTP servers.

Thus, any resource whatsoever may be "on the web", such as a star
cluster, by being denoted by a URI which is meaningful to HTTP and
resolves to one or more representations. That doesn't mean that that
star cluster is part of the physically linked network of representations.
But it is still "on the web" and thus one may interact with representations
of that resource via the web machinery.

This distinction between being "on the web" and being part of the
physical network of representations is one that I think needs to be
clarified in the TAG's web architecture document.

The Semantic Web also has two realizations or facets or dimensions:

    (1)  a virtual (ever changing) global graph of statements (both explicit
          and inferrable) constituting the sum of knowledge available to
agents
          at any given time, even though any given agent or physical
knowledge
          base may (usually always will) possess only a fraction of that
virtual
          all inclusive graph

    (2) an implicit network of interrelated resources related by statements
in
          that graph.

--

"On the Web" vs "On the Semantic Web":

A given resource may be "on the web" yet not "on the semantic web"
because even though that resource is denoted by a URI which resolves
via HTTP to a representation, there may exist no statement referring to that
resource. The moment such a statement is asserted, that resource is
then "on the semantic web". Since a given SW agent or KB will nearly always
possess only a subgraph of that virtual global graph that is the SW, it
will in practice never be possible to conclude absolutely that a given
resource is not on the semantic web.

A given resource may be "on the semantic web" yet not "on the web" because
even though that resource is referred to in one or more statements it is
is not denoted by a URI that is meaningful to HTTP and intentionally
resolves to one or more representations. The moment that resource is
denoted by a URI that resolves via HTTP to a representation, it is then
"on the web". Since a given web application may not be aware of all URIs
synonymously denoting the resource in question, it is unlikely that a web
application can ever conclude absolutely that a given resource is not on the
web.

Thus, the intersection of the Web and Semantic Web constitutes those
resources which are both (a) denoted by URIs that are meaningful to HTTP
and intentionally resolve to at least one representation and (b) are
referred
to in at least one statement in that virtual, global graph that is the SW.

Yet for both the Web and the Semantic Web:

1. We are talking about the same infinite set of resources, namely things
     in the universe, with the Web and Semantic Web embodying finite
     subsets of that infinite set of resources.

2. URIs serve the same function for both the Web and the Semantic Web,
    namely they denote resources. The Web provides machinery for
    interacting with representations of the denoted resources. The Semantic
    Web provides machinery for interacting with formal descriptions of
    the denoted resources. But the denotation (is presumed to) remain
    consistent for a given URI irrespective of Web or Semantic Web
    operations.

--

This brings us to a single critical missing piece to this puzzle, namely,
how SW agents consistently and reliably access authoritative
descriptions about resources in a standardized way which is
analogous to how web clients access authoritative representations
of resources.

This is where I see solutions like URIQA [1] playing a crucial role, by
providing that missing key piece of Semantic Web architecture.

Regards,

Patrick

--

Patrick Stickler
Nokia, Finland
patrick.stickler@nokia.com

[1] http://sw.nokia.com/URIQA.html

Received on Thursday, 17 July 2003 05:44:23 UTC