Re: What is RDF Datatyping?

On 2002-04-19 3:15, "ext Pat Hayes" <phayes@ai.uwf.edu> wrote:

>> Please read the following carefully and possibly several times
>> before blasting away at it. Thanks.
>> 
>> --
>> 
>> I assert that the following are true and should be captured by the
>> RDF Datatyping specification:
>> 
>> 1. The purpose of RDF Datatyping is to exchange knowledge containing
>> datatype values.
> 
> Not exclusively, no. That is clearly not true of all of our users, eg
> Dublin Core. If you change the last word to 'information' or some
> such, I would agree.

I disagree. The Dublin Core users are free to comment.

>> 2. The means by which RDF Datatyping captures knowledge about
>> datatype values is by associating lexical forms (literals) with
>> datatypes (URIrefs).
>> 
>> 2. A datatyping idiom is an expression of a datatyped value
> 
> No, that is too limiting. And unnecessary. The point of the
> datatyping idioms is to provide unambiguous syntactic patterns which
> have particular meanings relative to datatypes. Exactly what those
> forms and meanings are is determined by the rules in the MT, and we
> should explain them as clearly as we can, of course. BUt we should
> respect them (or change them.) We shouldnt just ignore them.

I never said we should ignore them. I'm saying the MT should
capture the above. Yes, it does so by defining syntactic patterns
and their meanings, and there are some side effects in addition
to the primary goal, such as assigning the value to a blank node,
but the primary goal is to communicate a datatype value. If a
datatyping idiom does not communicate a datatype value, it is
broken and useless insofar as datatyping is concerned.

>> which
>> associates a lexical form with a datatype, and
>> is comprised of (a) a literal, taken to be a lexical form, (b) a
>> URI, taken to identify a datatype, and optionally (c) a blank node
>> which denotes the datatype value represented by the lexical form
>> according to the lexical to value mapping of the specified datatype.
>> 
>> 3. A literal which occurs as part of a datatyping idiom always
>> denotes itself, the string,
> 
> Right

Please remember that I said that.

>> even though it is also understood to
>> represent a datatype value
> 
> Wrong. It should never be understood that way in RDF. An application
> may choose to do so, but that interpretation would go outside the RDF
> spec and could result in invalid inferences.
> 
>> in conjunction with the specified
>> datatype. 

Again, you seem to be having trouble with qualifying clauses.

>> This is not a contradiction. The literal by itself
>> cannot identify a datatype value. It is only the combination
>> of the literal and datatype which provide sufficient information
>> to identify a datatype value.
> 
> Even in combination with a datatype, the literal still means itself.
> That is what the WG agreed, and what the MT reflects.

Sigh. Isn't that what I just said above?

>> 4. The value denoted by a blank node occurring as part of a lexical
>> form or datatype property idiom is unknown to RDF. It can only be
>> known externally to RDF with the addition of full knowledge of the
>> datatype mapping. Thus, datatype clashes cannot occur within RDF.
> 
> That is a very misleading way of putting it. The clashes occur in
> RDF, in a real sense - they are a consequence of the RDF semantic
> conditions -  but they can only be detected by using information
> about datatypes.

Uhhh... well, since the values cannot be known or compared in RDF,
how can they clash in RDF?

You now seem to be positing constraints on RDF knowledge which
are imposed outside of RDF yet supposedly are part of the RDF
interpretation.

>> I consider sections 2-4 of the datatyping WD to capture these truths.
>> Whether the terminology used is optimal, or whether the concept of
>> datatyped literal pairings is acceptable or not, and whether it
>> remains or is removed, is a separate matter. I am happy to
>> rephrase those sections to the satisfaction of the WG, or allow
>> any other WG member to do so, but I consider the above truths
>> to hold and will expect any rewording of the datatyping specification
>> to reflect them.
> 
> Then we need to discuss what to do, because I think these views do
> not reflect the intent or the actual stated consequences of the MT as
> it exists, nor the clearly expressed description of the datatyping
> model as expressed in my document, nor much of the email and telecon
> discusssions that led us to this point.

Well, I disagree with much of the above, so I guess the WG needs to
say something.

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 Friday, 19 April 2002 04:57:18 UTC