W3C home > Mailing lists > Public > www-tag@w3.org > September 2004

Re: Information resources?

From: Sandro Hawke <sandro@w3.org>
Date: Thu, 09 Sep 2004 14:18:40 -0400
Message-Id: <200409091818.i89IIeh2028673@roke.hawke.org>
To: Norman Walsh <Norman.Walsh@Sun.COM>
cc: www-tag@w3.org


> / Sandro Hawke <sandro@w3.org> was heard to say:
> | I don't think dogs and cars are obscure.  It's hard to work with RDF
> | for a day without having URIs for people, organizations, books,
> | conferences, etc, ... and being stuck with the very real problem of
> | distinguishing between those things and web pages about those things.
> 
> In what context have you had this problem?

[read on]

> I have used the URI 'http://xmlns.com/wordnet/1.6/Hoary_Marmot' to
> identify the Hoary Marmot (really, I'm serious[1]). If I GET a
> representation of that URI, I get some RDF that tells me things about
> the Marmot. I consider that data to be a representation of the
> physical creature.

To clarify, by "the Marmot" and "the physical creature", you
mean the class of Hoary Marmots, right?  There is a particular
instance of that class which you seem to identify as
"http://norman.walsh.name/knows/what#hoary-marmot".

> Assertions that http://xmlns.com/wordnet/1.6/Hoary_Marmot is a web
> page or has a particular creator or last modified date or
> what-have-you are inconsistent.

How could web client software not have such assertions interally,
whether or not expressed as RDF?  My RDF web client software [2],
infers exactly that when I use that URI.  It says to itself, "that's a
nice looking URI... maybe I can dereference it and learn
something... oh, cool, I can... so a Hoary_Marmot is a kind of
Marmot...  meanwhile, we got back some HTTP headers; let's remember
the expiration time so we can refresh it when it expires...."

And it ends up thinking (and telling users) that a wn:Hoary_Marmot is
an rdfs:Class (a subclass of wn:Marmot) *and* a retr:RDFSource.

As you say, that's inconsistent.  I've defined RDFSource as a subclass
of InformationResource (called retr:Source in the code), and I've
proposed a definition of InformationResource which includes the
statement that it is disjoint from wordnet:Physical_object -- a
superclass of wordnet:Marmot.  (Okay, I haven't actually published
that ontology yet, but that's what's emerging here, and it would be
sensible to publish it.)

My system wont actually break under this inconsistency since the two
sets of facts are kept in different graphs with different provenance,
but it's still a bad state of affairs when WordNet and the system's
internal knowledge base are in obvious (or not-so-obvious)
contradiction.  Given the right trust rules, one would find
inconsistencies in data believed to be (and which apparently is)
correct.

Do you see a way around this problem, other than by having wordnet
switch to using some kind of indirection as I've suggested?  Wordnet
(as currently published at http://xmlns.com/wordnet/1.6/) is one of
the few vocabularies with this problem; I expect to motivate DanBri
to fix it soon.  

     -- sandro

> [1] http://norman.walsh.name/2003/07/25/images/marmot.html
[2] http://dev.w3.org/cvsweb/2004/semwalker/
Received on Thursday, 9 September 2004 18:16:22 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Thursday, 26 April 2012 12:47:28 GMT