Re: Heads up: RFC2996bis, possible problem for RDF

Jeremy,

It appears we are in agreement about RDF (phew! ;-)...

At 12:42 18/02/04 +0000, Jeremy Carroll wrote:
>Graham:
>
>>Again, I don't think we are silent.  In RDF, they are different URIs by 
>>my understanding.  And what about:
>>   http://foobar/ and http://foobar:80/
>>compared with:
>>   ftp://foobar/ and ftp://foobar:80/
>>RDF knows nothing about schemes, so they must be different, surely?
>
>
>My understanding is that RDF Semantics does not say that these are the 
>same, however, it does not prohibit adding
>
>http://foobar/ owl:sameAs http://foobar:80/ .
>
>(with an understanding of owl:sameAs).

Exactly.  My concern is that the revised RFC2396bis suggests that it is 
*always* safe to replace one with the other, by the language:

[[
Even though it is possible to determine that two URIs are equivalent, it is 
never possible to be sure that two URIs identify different resources. For 
example, an owner of two different domain names could decide to serve the 
same resource from both, resulting in two different URIs. Therefore, 
comparison methods are designed to minimize false negatives while strictly 
avoiding false positives.
]]
-- http://gbiv.com/protocols/uri/rev-2002/rfc2396bis.html#equivalence

Specifically: "while strictly avoiding false positives"

>A SemWeb reading of the section of RFC2396bis that we are reviewing could 
>be that a web service could automatically generate such a triple for every 
>http URI it sees. I would see such a web service as compliant with both 
>RFC 2396 bis, and with RDF Semantics. (It is adding new information, but 
>information which one of the authorities involved (IETF) underwrites).
>
>
>>I figure I must be misunderstanding what you are saying.  Maybe my 
>>responses will flush out the disconnect.
>
>I have tried to turn into it a test triple:
>
>http://foobar/ owl:sameAs http://foobar:80/ .
>
>is not implied by RDF Semantics combined with OWL Semantics
>but is implied by RDF Semantics + OWL Semantics + comparison ladder from 
>RFC 2396bis.

Yes.  But I have two concerns:

(a) a future version of RDF would presumably cite RFC2396bis, in which case 
it acquires an additional axiom, which I think is a potential source of 
surprise.

(b) the RDF rules for formation of URIrefs from namespaces and local names 
mean that URIrefs may be changed (by the RFC2396bis normalization) in ways 
that cannot be described by additional assertions of this kind.

>Conversely
>
>NOT
>http://foobar/ owl:sameAs http://foobar:80/ .
>
>is not implied by any recommendation.
>
>Hence, most RDF systems should not endorse the test triple, but must not 
>contradict it.

Yes.

>RDF systems may explicitly implement the comparison ladder and then would 
>endorse the test triple.

Then some given RDF may have different meanings when processed by different 
systems.  If the triple is to be endorsed, I think it should be explicit 
(somewhere) in the RDF data.

>In the open world, the test triple is true;

That seems like an unjustified leap to me.  It may be true in the "open 
world" *when used for the purposes of retrieval*, but there are other 
purposes (e.g. URIref formation from namespace URI) for which it is not true.

In practice, we would hope that people do not assign different meanings to 
URIs that differ in these ways, but we seem to agree that this is not 
prohibited by RDF.

>... but it is not part of the open world that we expect RDF systems to be 
>aware of.

#g


------------
Graham Klyne
For email:
http://www.ninebynine.org/#Contact

Received on Wednesday, 18 February 2004 10:41:18 UTC