Re: metadata vs. meta-(meta)-data

> (Following up on an offline question about relation of
> 'tag' to 'ttdb' (The Thing Described By))
> 
> The "tag" scheme you (Sandro) have proposed makes no
> commitment about the resource it designates, at all. It's
> completely unbounded; you're saying that a 'tag' is a
> uniform and unique identifier without ever making any
> commitment that there's a resource. As such, a 'tag' doesn't
> itself identify a resource (in the same way that a http
> URL carries along with it an implicit notion that it locates
> a resource that's served by a HTTP server.)
>
> The only utility of 'tag' URIs would be to use them
> in systems that wanted to supply all of their own
> extrinsic binding of URI to semantics.

You say that like it's a bad thing.  :-)

You're right about tags.  They are strings which can be used as
identifiers rather than being somehow identifiers by their nature.
The degree to which they denote something and the manner in which that
denotation is fixed or communicated is up to the person minting the
tag (and the people it gets sent to).  The sole feature of tags is
that pratically anyone can easily mint them with confidence that they
are unique with respect to all other properly minted tags.  Use in
particular protocols might constitute agreement to additional
restrictions, such as that the denotation will remain fixed.

My expectation is that they will be useful as constant symbols in
declarative languages, enabling people to combine linguistic
expressions from different sources.  But there are other mechanisms
they could use instead, including ttdb.

> So your proposal is fundamentally different than
> 'ttdb';

Absolutely.  They may each be used to solve some of the same problems
(such as what the heck an http URI denotes), but they do it in very
different ways.

>                    I propose explicitly binding the
> URI to "the resource described by the resource identified by".

Indeed.  In my notes from a few months ago, I played around with two
such schemes, "primary-denotation" was the same as "ttdb".  I also had
another scheme, "borrowed-denotation", which used a string and a URI
and claimed to denote whatever the string denoted in the contents
obtained from the URI.  I took a first stab at that (in my first
posting to this mailing list), in [1] and continued to refine it in
private a bit, before I decided this whole thing was better done
in the logic language than in the identifier's internal sub-language.

So instead of saying
  <ttdb:http://www.w3.org>       
in some expression, I could just say
  <tag:sandro@w3.org,1-5:w3c>      (or some other arbitrary string)
as long as I also said
  <tag:sandro@w3.org,1-5:w3c> desc:describedBy <http://www.w3.org> 

N3 people are used to just writing:            
  [ desc:describedBy <http://www.w3.org> ]
which glosses over the generated identifier, but is pretty much the
same thing.

> In fact, I could combine the two things, and make
> 
> ttdb:<date>:<uri>
> 
> as 'the thing described by <uri> on <date>'
> 
>    ttdb:2001:http://www.w3.org
> 
> is a URI for "The World Wide Web Consortium, as described
> by http://www.w3.org at the start of 2001".

That's nice for a URI.   With some RDF statements, we could do
  [ desc:describedByAtTime <http://www.w3.org> ;
    desc:descriptionTime [ ... /* some time structure */ ... ] ]

I think using statements like this makes developing and deploying
systems easier developing and deplaying URI schemes in the current
system.  I'm not sure it has any advantage over URIs in a system where
scheme identifiers are unconstrainted, such as if you think of tags
as open URI schemes.  You could use 
   tag:LMM@acm.org,1-5-23:ttdb:2001:http://www.w3.org
immediately (assuming tag URIs are any more standard than ttdb ones :-)


I'm afriad this discussion will go 'round-and-'round unless we're
talking about particular application scenarios.


"ttdb" works nicely for humans, but for machines (automated reasoning
systems) I don't see how its any better than tags.  Similarly, my
desc:describedBy requires the same kind of human "magic" to get
anywhere.  The one thing computers can do with it easily is know its a
many-to-one relation, so that if I say 
  <tag:sandro@w3.org,1-5:w3c> desc:describedBy <http://www.w3.org>.
  <tag:sandro@w3.org,1-5:w3> desc:describedBy <http://www.w3.org>.
  desc:describedBy a daml:UnambiguousProperty.
than reasoning systems (which know daml) can infer
  <tag:sandro@w3.org,1-5:w3c> daml:equivalentTo 
                                         <tag:sandro@w3.org,1-5:w3>.

Actually, while we're at it, I think
  <tag:sandro@w3.org,1-5:w3c> desc:describedBy <http://www.w3.org> 
should probably be
  <tag:sandro@w3.org,1-5:w3c> desc:describedByURIContents "http://www.w3.org"

Bottom line: I think the semantic web should have "tag:" (or something
similar) for arbitrary identifiers, "data:" (or something similar) for
literals, and ftp/http/https/etc for data inclusion.  Can you give me
a situation where something else would be useful?

[1] http://lists.w3.org/Archives/Public/www-rdf-logic/2001Jan/0080.html

Received on Thursday, 24 May 2001 11:16:02 UTC