Re: "do not occur"

On Tue, Jun 2, 2009 at 5:30 PM, Alan Ruttenberg
<alanruttenberg@gmail.com> wrote:
> On Tue, Jun 2, 2009 at 4:59 PM, Axel Polleres <axel.polleres@deri.org> wrote:
>>> Incidentally, the fact that you can filter using the DATATYPE function
>>> in sparql is another hint that something is amiss. By my earlier
>>> analysis, the DATATYPE function should never return rdf:PlainLiteral,
>>> according to our spec.
>>
>> Indeed, *according to our spec*. This is why I prefer Option 2 which makes
>> this point clear.
>
> I would modify it to not try to make it invalid, but instead to make
> it clear we say nothing about the lexical to value mapping of such
> literals within RDF and SPARQL.

In RDF, before you attempt a D-interpretation or to prove a
D-entailment, you have to decide what D you want to use. You are
forced to make a choice. So saying nothing isn't an option if
rdf:PlainLiteral is in the domain of D. To implement what you suggest
(saying nothing), the datatype map would have to be prohibited from
having an assignment for the URI rdf:PlainLiteral, as an individual
datatype also does not have the ability to "say nothing". If I
understand RDF semantics correctly you would then get no entailments
involving such typed literal nodes (loosely speaking), making them as
ill-behaved as a literal node with a randomly selected URI after ^^
(and I'm not sure exactly how ill-behaved that is). You could
separately constrain interpretations by saying that the URI
rdf:PlainLiteral must be interpreted to be the datatype described in
section 3 of our spec; this would allow use of the URI in other
contexts. It just wouldn't be in the datatype map.

This seems really weird to me, and I haven't chased through the
consequences to SPARQL, OWL, RIF, or anything else, so I can't
recommend this approach.

Jonathan

Received on Wednesday, 3 June 2009 14:33:54 UTC