RE: Comments on the new RDF Test Cases draft

>On Fri, 31 May 2002, Jan Grant wrote:
>
>[I've lost the attributions, sorry]
>
>>  > >Therefore, such "word smoothering", plus a precise definition of
>>  > >isomorphism, suffice. But, note that if we go along the
>>  > >"smoothering way", the same problem of a precise definition of
>>  > >isomorphism can be nicely dropped as well, as the wording can well
>>  > >say that the "expected output" is the given N-triple one, and just
>>  > >be silent on the isomorphism issues at all (as, it's rather clear
>>  > >that N-triple output is defined modulo renaming of blank nodes, and
>>  > >in any case, crucially, no *formal* definition is then needed as
>>  > >the Test Cases contain clarification guidelines, and not formal
>>  > >normative definition of "test passing for parsers").
>
>Ack! I didn't notice this before!
>
>I should say that jeremy Carroll is producing a document describing what
>we mean by "an RDF graph" including an expression of the appropriate
>notion of isomorphism.
>
>But I see you're talking about "renaming blank nodes". Blank nodes
>_don't_ have names. They have identity wrt the graph they're a part of;
>the "names" are simply an artifact of the serialisation syntax and have
>no non-local meaning.

Right. However Ive begun to see that Massimo (author of above, as I 
recall) does have a point. Even though blank nodes don't have names, 
they do have an identity. Suppose we start with an RDF/XML document, 
parse it, and generate a(n instance of a) RDF graph A. That contains 
urirefs and literals and bnodes. OK so far. Now suppose that we parse 
it *again* and generate *another* RDF graph B. The urirefs and 
literals in B are the *same* as in A - literally the same- but the 
bnodes are different, right? That is what we meant by 'local to the 
graph'. But in some sense A and B are the same. They express the same 
content, are derived from the same document, they would look exactly 
the same if you were to draw them as pictures, etc. . They are 
virtually indistinguishable, and ought to be interchangeable with 
each other. Maybe it would be useful to have a notion of 'isomorphic' 
which allows one set of blank nodes to be swapped with another 
without 'changing' the graph.

Or, we could say that two graphs are equivalent if they are instances 
of each other. That is a stronger relationship, since equivalent 
graphs need not be graph-isomorphic (example in 
http://www.w3.org/TR/rdf-mt/#graphdefs) but they might as well be as 
far as their meaning is concerned. It was tricky to get 'instance' 
right, by the way.

>An N-Triples document is just a description of an RDF graph, which may
>contain some blank nodes; they really are (honest!) blank.
>
>jan
>
>PS. I'm still somewhat surprised by complaints about abusing
>graph-theoretic terms. Like I said before, the graph theorists I've met
>all seem to play fast and loose with terminology (where context is
>obvious) and have few qualms about dealing with graphs, multigraphs,
>partially-labelled multi-digraphs, etc. and will still tend to refer to
>them all as a "graph" when talking.

But I bet they are much more anal about things that get printed.

Pat
-- 
---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes

Received on Friday, 31 May 2002 16:59:01 UTC