Re: use/mention and reification

On 2002-01-23 23:01, "ext Frank Manola" <fmanola@mitre.org> wrote:


> If we really determined that they truly denote the same "thing", we
> certainly would like to try!

It depends on the application, and of course, specific applications
are free to interpret, modify, and merge RDF graphs as they need.

There are, however, very real contexts where even if we know with
100% certainty that two resources with different URIs are the very
same thing, we can never merge those nodes, unless we wish to
discard all traces of "perspective".

E.g. consider a digital resource with DOI, ISBN, and proprietary
identities, each of which is a URI, and different users of that
resource make statements about it using different identities,
yet it is known that these are all synonymous identifiers.

You may decide to merge based on e.g. DOIs, but then some
application presents a query based on the ISBN, so you
still have to maintain the variant identities and their
equality.

If you merge all synonymous URIs into a single URI:

1. How do you choose which one to use?
2. How do you accomodate queries from applications that only
   "know" the resource by one of the other URIs?
3. How do you maintain historical perspective for statements
   made using one URI vs another?
4. What if equality of identifiers is not universally agreed?
5. What if equality of identifiers is later found to be false?

Thus, it seems to me that one can never escape having to
accept all synonyms in a knowledge base and thus any benefits
of actual merging is either limited to a very narrow
application space (or closed systems) or is an illusion.

> There are many highly practical examples
> where you'd want to get rid of one of the identifiers and agree to use
> the other.  

Some examples would be very appreciated (though feel free to
send them offlist, if you think they would be noise for the
majority)

> ...  Also, there are practical
> difficulties with "legacy data" using both URIs that is hard to revise.

Right. As I outlined above (apparently unnecessarily ;-)

> If by "different resources in RDF space" you mean RDF has to treat them
> differently, I agree (and in any case, Superman and Clark Kent are
> slightly different, as I've suggested above)).  But they may still be
> the same resource, right?

Different RDF resources, having different URIs, but perhaps
the same "thing". Just like (xsd:integer, "5") and (xsd:integer, "05")
are the same "thing".

It is correct, though, that even if two RDF resources correlate
to the same "thing" in some universe, that doesn't mean that
those resources are actually identitical, since as is the case
with Superman versus Clark Kent, resources can denote a subset
of qualities of a given "thing" that are significant to a given
perspective or context.

 
>>> I don't see how we can reasonably expect URIs to be used only in
>>> situations where everyone on the Web has a full understanding and
>>> agreement on all the attributes of the thing the URI names before they
>>> start using it.
>>> 
>> 
>> There need not be agreement about all attributes of a thing, only
>> about what thing a given RDF resource denotes.
> 
> 
> The *intent* is certainly that use of the same URI indicates that
> agreement.

Hmmm... I can't say I agree. Use of the same URI is agreement
about the referent, not agreement about everything that anyone
might say about that referent.

We may both agree that <person:John> denotes a specific person,
but not agree that that person is smart. You may assert he is
smart. I may assert he is dumb. But all the common use of the
URI implies is agreement that we are both talking about the
same person, nothing more.

Of course, without mechanisms for authority, scope, source, etc.
it would be very difficult to benefit from such an interpretation,
and I agree that insofar as the present RDF is concerned, it's
pretty hard not to have to accept the view that use of the same
URI implies agreement about everything said about the resource
denoted by that URI.

Still, I hold fast to my view that use of a URI only implies agreement
about the identity of the referent, and am working on the
other bits needed to make it workable...

>  But I repeat that I only know what someone else denotes by a
> URI (at least, general URIs denoting non-digital things) by their
> description of it in terms of other attributes.

Agreement can be (and to date usually is) based on knowledge
not expressed in RDF but expressed in natural language and
exchanged in written prose or word of mouth.

Perhaps in the future we will have a sufficiently rich and
mature set of upper level ontologies which will allow us to
define our understanding of the identity of resources based
on semantic primities and expressed in RDF, but not yet.

>   And I may, on the basis
> of the description you present, think we have an agreement when we
> really don't, because we disagree on other attributes that you didn't
> include in your description.

Insofar as they are attributes which establish only identity,
I agree. Other attributes need not be the basis of agreement
for common use of a URI.
 
>> Of course one cannot prevent misunderstanding, or even disagreement,
>> on the SW, but agreeing on the same URI to denote the same thing
>> seems a reasonable basis to conduct discussions, and we can hope
>> that knowledge expressed in terms of RDF and URIs will be less
>> ambiguous than knowledge expressed in natural language ;-)
> 
> 
> I agree.  My only point is there may be some additional conversation
> necessary to resolve such misunderstandings, even after we *think* we
> have agreement (and the Web provides a medium for conducting those
> conversions).

Absolutely.
> Strictly speaking though, the problem isn't that we're allowing a given
> URI to denote more than one thing.  The problem is how well (accurately)
> we can transmit that denotation from the minting authority so that it is
> (again, accurately) shared by others who are going to use it.

Quite so.
 
Patrick

--
               
Patrick Stickler              Phone: +358 50 483 9453
Senior Research Scientist     Fax:   +358 7180 35409
Nokia Research Center         Email: patrick.stickler@nokia.com

Received on Thursday, 24 January 2002 10:40:06 UTC