Re: datatype maps and annotations

[ I'm not sure what the correct answer is or should be, but I can think of
a quick workaround...
We could define a SuppressWarnings annotation, applicable to datatype
declarations, which could be used  to indicate  which uses (as a range, or
in literals) should be ignored by profile checkers.
Definitions are a different situation, since reserved namespace  types
should not be  redefined, but subtyping (using unvalidated facets) could be
ok.
]


One aspect of the motivating instance for this issue that makes the case
harder is that it involves a lexically valid literal of datatype of
xsd:date [I think].

Intuitively this seems like it ought not to be a violation arising from the
datatype map, but it is less clear that the mention of a reserved
vocabulary term would still be in DL.

If we modify the scenario different issues are exposed.

1. If the datatype were in a different namespace, it is difficult to
justify rejecting the literal from FULL.

2. If the lexical form were invalid then various RDF 1.1 issues come into
play; if the ontology enters RDF space, the literal is ill-typed, and the
entire ontology document would be inconsistent.  SPARQL 1.1 could be read
as requiring this if OWL DL is the reasoning profile, since xsd:date is
recognized.
  If the datatype is user defined in the ontology closure, then any use in
a literal is necessarily ill-typed, and recognized by an OWL 2 processor,
thus making it potentially illegal to write a document to an RDF syntax.

3. Literals in DL are MUST'ed to have lexical forms that belong to the
literal space of the datatype. This would seem to imply that full  can
ignore this restriction, which would make any restriction on datatypes for
full pointless.
It also implies that all literals in DL must be able to be validated by any
OWL processor, and thus must be restricted to the default map. This may be
a bit harsh.


4. If the value of the annotation is an anonymous individual (and this
individual has assertions) , then it seems that this would be a violation
of EL, since that individual could be visible to EL even if it is only used
as the value of annotations.
If annotation values are invisible to profiles, it seems strange to allow
for anonymous individuals to be used as annotation values in this case, but
to prohibit those individuals from having object / data property assertions
made about them.
It's possible to edge around this issue if the blank nodes are converted to
IRIs (e.g. applying the transform from RDF 1.1), but then any o/d property
assertions on the IRI would have to be made in a different ontology
document, and that document not be imported, if the individual is to remain
invisible to the reasoner.

Annotations and annotation properties are a little bit messy. Especially
when people use annotation properties to model things that  are fundamental
parts of the domain of discourse... [/me removes angry KOS SME hat].

Simon

On Mar 30, 2017 6:43 AM, "Ignazio Palmisano" <ipalmisano.mailings@gmail.com>
wrote:



On 30 Mar 2017 09:53, "Ian Horrocks" <ian.horrocks@cs.ox.ac.uk> wrote:

I agree that the clear intention was that annotations play no role in
determining the profile, even if the spec is not so clear on this point.

This POV is supported by the stated motivation for the profiles, which is
to "trade some expressive power for the efficiency of reasoning”.
Annotations do not contribute to expressive power and play no role in
reasoning, so it would be very strange if they were to play a role in
determining the profile.

Ian



OTOH an undeclared annotation property, while not affecting reasoning - and
so not affecting the profile - becomes an annoyance for parsers, among
other things.

So maybe the profile checker could produce a list of profile violations and
one of warnings? The second to be dealt with for the purpose of better
parsers and in general improving the world?

Cheers,
Ignazio




> On 29 Mar 2017, at 21:15, Alan Ruttenberg <alanruttenberg@gmail.com>
wrote:
>
> I'm in a discussion about whether annotation axioms and ontology
annotations are subject to the datatype restrictions that are checked for
conformance to a profile. I don't recall discussing this, but it seems like
it would not have been the intent since the datatype map is only of
relevance to the reasoner. Some parts of the spec read like they might
apply, others part ambiguous (at least to me).
>
> *I* don't think that the values of annotations should play a role in
determining the profile, but I wonder if others who were in the WG can
speak to this.
>
> Alan
>
>