RE: does RDF require understanding all 82 URI schemes?

: Graham Klyne:
: Should there be exactly one resource corresponding to a given literal, or
: may there be many?

It does not seem sound to unify literals by conveting them to URIs. Any such
attempt is to say the least, lossy. Converting a literal to a data URI doesn't
give any more information about the literal that would allow it to be matched.
Consider a trivial example:

"Paris" -> data:,Paris
"Paris" -> data:,Paris
"Paris" -> data:,Paris

The first "Paris" was the name of a perfume, the second was the name of a god,
the third was the name of a city. By converting them all to URIs, I've lost
information. If I convert them to different URIs I add no information. One way
to gather information is via the local context in which these were found, so a
machine might be able to abduct with a reasonable probability that as it

"Paris" -> data:,ParisTheCity
"Paris" -> data:,ParisTheGod
"Paris" -> data:,ParisThePerfumeTM

Of course the machine will have to make a best effort attempt to do this for
each and every occurence of "Paris", there's no chance of a general search and
replace. The machine might get the mappings wrong. But so do people.

Yes it is annoying that literals can't be matched in a general way, but that's
not a problem unusual to RDF, and no one here should be expecting the wg or the
ig to solve it by URI hacking. We could disallow the literals altogether, but
that's not really practical. One suggestion would be that the next M&S
explicitly acknowledege this limitation in using literals and suggest
non-normatively and in the nicest possible way, that people who want to
interoperate and allow their data to be inferred over by others ought to ground
it and stop being so lazy :)

Bill de hOra

Received on Wednesday, 14 February 2001 17:50:47 UTC