Re: Update semantics LCC

Summary:  The updated semantics LCC still has critical problems.

Comment:  As I am not a member of the RDF Core WG, I don't completely
understand what the semantics of RDF is supposed to be.  Thus I am not a
good person to comment on whether the semantics LCC actually matches the
intent of the RDF Core WG.  Given the number of errors that I have caught,
both internal inconsistencies and mismatches with my partial understanding
of the intent of the RDF Core WG, I feel that the document needs a thorough
review by someone who can check that it matches the intent of the RDF Core
WG.


From: Brian McBride <bwm@hplb.hpl.hp.com>
Subject: Update semantics LCC
Date: Wed, 15 Jan 2003 13:40:37 +0000

> Pat has managed to get an updated LCC semantics doc to me.  It seemed best 
> to load it immediately to the website, which I have done.
> 
>    http://www.w3.org/2001/sw/RDFCore/TR/WD-rdf-mt-20030117/
> 
> There are no accompanying change notes, so I'm not sure what's 
> changed.  However, based on my conversation with him, I'm expecting that:
> 
>    o he has fixed the problem Peter raised about rdfs:Literal
>    o he has made at least one other change in response to a comment from Jan.
> 
> Peter, could you take a look and see if it does indeed resolve the issue 
> you raised?

I've quickly looked over the new version.

The semantics still has strange behaviour with respect to
rdfs:Literal and untyped literals.  
The semantic conditions boil down to

	ICEXT(I(rdfs:Literal)) <= LV
	ICEXT(D) <= ICEXT(I(rdfs:Literal)) for D a datatype

This means that with XML Schema datatyping

	I("1"^^xsd:decimal) in ICEXT(I(rdfs:Literal))
	I("a"^^xsd:string) in ICEXT(I(rdfs:Literal))

but 

	I("a") not necesarily in ICEXT(I(rdfs:Literal))

The Lbase appendix still has errors.  It does not accord rdfs:XMLLiteral
its special status with respect to language tags.  It is not a consequence
of the RDFS model theory that rdfs:XMLLiteral(?x) implies rdfs:Literal(?x).
It is not a consequence of the RDFS model theory that untyped literals
belong to rdfs:Literal.  There are other errors as well.

The semantics has problems with respect to LV.  

LV is mandated to contain all literal strings (which is not defined in the
document), pairs of literal strings and language tags (which is also not
defined), and all well-formed canonical XML documents (there is no pointer
given here for this).  IL is defined to take typed literals with type
rdf:XMLLiteral to XML canonical forms.  It is certainly not obvious that
XML canonical forms have any particular relationship to well-formed
canonical XML documents.

However, even if this is ironed out, there is no particular relationship
between rdf:XMLLiteral and rdfs:Literal in RDFS.  The class extension of
rdfs:Literal in RDFS interpretations need not include any well-formed
canonical XML documents.  This only comes from the (newly modified) fourth
condition in datatyped interpretations.

The special status rdf:XMLLiteral is not respected in datatyped
interpretations, which require that 
    IL(x@"en"^^rdf:XMLLiteral) = IL(x@"fr"^^rdf:XMLLiteral)


The Lbase document is still not a W3C Note.  It should have some formal
status before the Semantics document goes to last call, as changes to it
will affect the Semantics document.

> Other reviewers, could you check any critical issues you raised are dealt with.
> 
> Brian

Peter F. Patel-Schneider
Bell Labs Research
Lucent Technologies

Received on Wednesday, 15 January 2003 14:56:09 UTC