Re: Changes to make S&AS consistent with RDF Semantics document

In this message I summarize, for convenience, all
the options discussed so far for making the spec consistent with 
respect to XMLLiteral.

Jeremy's message describes an implementation sketch 
for working with inconsistent specs [1], which therefore seems to 
belong to a different category.  I give further reaction to Jeremy's 
message separately.

On further reflection, solution 1 needed to be changed
from possible to impossible, for the same reason that solution 3
was already noted to be impossible, namely that it violates OWL 
being a semantic extension of RDFS.

Conclusion: there are two possible solutions.
There is one possible solution, #4, which could be performed by Webont 
on its own.
Another 'possible solution', #2, is less likely, as it would 
require going backward processwise since RDF Core would need to 
change its design.

==
Details:

>>*Impossible solution 1*: no changes to RDF Semantics, and two
>>"local" changes to S&AS:
>>-1a: add OWL DL semantics without XMLLiteral by means of a
>>detour via the abstract syntax (details below)
>>-1b: exclude the possibility of having OWL Full without
>>XMLLiteral semantics (after all, OWL Full is *OWL Full*)
>>
>>In view of 1b, this would not solve the problem completely, but
>>perhaps this would be acceptable?
>>In addition, 1b would require a small change in OWL Reference.
>Actually, 1b does not describe a change in S&AS, since S&AS as it
>is does already include XMLLiteral semantics in OWL Full.
>However, 1b does lead to changes to make Test consistent.
>For clarity, I list the other change required when this
>solution is taken:
>-1c: remove all tests that allow OWL Full without XMLLiteral
>semantics
On further inspection this 'solution' violates proper semantic
layering.  What does it mean that OWL is a semantic extension
of RDFS?  It means, in particular, two things:
1) each RDFS-inconsistent RDF graph is OWL-inconsistent
2) each RDFS entailment is an OWL entailment
Both statements can be contradicted under this 'solution'
with the graph G mentioned earlier:
>>      v p l
>>      p rdfs:range rdfs:Literal
>> where l is an ill-typed XML literal
G is RDFS-inconsistent but would become OWL-DL consistent (when
XMLLiteral is not in the datatype map)
A graph H can be found such that G RDFS-entails H but such that
G would not OWL-DL entail H (take the vocabulary of H to be large
enough).

>*Possible solution 2*: no changes to S&AS, and two changes
>to RDF Semantics:
>-2a: move the XMLLiteral conditions from Section 3 (on RDF) to
>Section 5 (on datatypes)
>-2b: allow the possibility to leave XMLLiteral out of a
>datatype map.
Pat Hayes has said twice that is too late for this change to RDF
[2] [3].

>*Impossible solution 3*: no changes to RDF Semantics, one
>change to S&AS:
>-3a: delete the (RDF-)conditions on XMLLiteral from the 
>definition of OWL interpretation in S&AS.
>
>This solution seems to be excluded, since it would violate the
>central assumption on the semantic layering of OWL on top of RDF,
>namely that OWL is a semantic extension of RDFS.

>*Possible Solution 4*:
>-4a: add the requirement in Section 3 of S&AS that datatype maps
>include XMLLiteral
>(This change is suggested in part for clarity: it does not change
>any entailments for OWL DL and OWL Full.
>This change does remove the currently present third OWL semantics
>in S&AS - abstract syntax-based semantics without XMLLiteral -
>which should not be there because the WG decision on semantic
>layering prescribes only two semantics, OWL DL and OWL Full.)
>-4b: remove all tests that allow OWL Full or OWL DL without XMLLiteral
>semantics

Herman

[1] http://lists.w3.org/Archives/Public/www-webont-wg/2003Dec/0057.html
[2] 
http://lists.w3.org/Archives/Public/www-rdf-comments/2003OctDec/0234.html
[3] 
http://lists.w3.org/Archives/Public/www-rdf-comments/2003OctDec/0236.html