Re: OWL and RDF lists

>
> It would also be really nice to understand if there is a fundamental
> reason for this limitation in OWL.
>

I believe this is to help distinguish the ABox from the TBox.

I realize you know this, but I'll set it out for clarity:
OWL is expressed in RDF using collections (owl:intersectionOf, owl:unionOf,
owl:oneOf, etc). If the rdf: namespace can be referenced in OWL, then it
becomes difficult for a reasoner to know the difference between lists that
are part of the ABox, which should be processed, and the TBox, which
shouldn't be. It may be possible for a reasoner to use some analysis to
identify triples that appear in the TBox and ignore those, but it still
gets difficult to do queries across a large ABox and filter these out.

In practice, there can be workarounds for this, such as keeping the TBox in
a separate graph, but that sort of thing is not considered in the standards.

If OWL collections had been done with predicates in the owl: namespace,
then this would have provided a solution, but that would not have worked
with any of the serialization formats. And OWL needs a list structure in
order to keep collections closed.

Paula

On Fri, Aug 12, 2022 at 8:31 AM Eric Prud'hommeaux <eric@w3.org> wrote:

> Molstly protégé, and other OWL-capable tools like StarDog. I'd like to
> figure out for FHIR whether to use RDF lists, but I think it's
> important for the SemWeb as a whole to solve this. As it stands, it
> looks like one should never use RDF lists.
>
> Given that way more RDF data processing happens with SPARQL than with
> OWL, and SPARQL *can* access lists (albeit awkwardly), it sounds crazy
> to forgo using lists just because someone someday might want to do OWL
> inference over the entities in the list. Voilà the dilema.
>
> It would also be really nice to understand if there is a fundamental
> reason for this limitation in OWL. I beleieve that Jim's research
> shows that at least OWL-API can be updated to allow lists. If this is
> the case, and this can be replicated in other OWL implementations,
> then perhaps the answer is a period of civil disobedience where folks
> violate the spec, use tools that support that violation, and
> eventually update the OWL spec. This interim solution could be made
> official with an OWL 1.1. errata stating that rdf:first/rest *can* be
> used in axioms.
>
>
> On Thu, Aug 11, 2022 at 08:47:27PM +0000, Mark Wallace wrote:
> > Eric, are you looking for a solution that runs Within protege? Within a
> triple store?
> >
> > ________________________________
> > From: Eric Prud'hommeaux <eric@w3.org>
> > Sent: Thursday, August 11, 2022 8:26 AM
> > To: semantic-web@w3.org <semantic-web@w3.org>
> > Cc: Jim Balhoff <balhoff@renci.org>; dbooth@dbooth.org <
> dbooth@dbooth.org>
> > Subject: OWL and RDF lists
> >
> > RDF lists (technically "collections" ¹) have terse abbreviations in
> > Turtle/SPARQL and a "ladder" representation as triples.
> >
> > <https://en.wikipedia.org/wiki/Flag_of_France> :orderedColors _:1 .
> > _:1 rdf:first "blue" .
> > _:1 rdf:rest _:2 .
> > _:2 rdf:first "white" .
> > _:2 rdf:rest _:3 .
> > _:3 rdf:first "red" .
> > _:3 rdf:rest rdf:nil .
> >
> > The SPARQL 1.2 WG is wrestling with lists ², and JSON-LD 1.1 has added
> > support for them ³. OWL however, specifically disables them by
> > prohibiting inferences across predicates in the rdf: namespace à la
> > Jim Balhoff's example ⁴;
> > [[
> > :contains rdf:type owl:ObjectProperty ;
> >   owl:propertyChainAxiom ( rdf:rest :contains ) .
> > ]]
> >
> > FHIR is a set of models for clinical record. It has representations in
> > XML, JSON and RDF. There's a playground ⁵ to explore alternatives
> > which illustrates alternatives, including whether to use
> > rdf:Collections (see button at top-right). With collections turned
> > off, we have to roll our own order (fhir:index 0, 1, 2...), which
> > kinda goes against RDF standards.
> >
> > I put together a gist which illustrates three observations we might
> > encounter in a patient's record. The codes for the first two appear in
> > a SNOMED hieararchy you might query for evidence of bone density loss
> > (clinical example, balancing corticosteroids against osteoporosis).
> >
> >
> https://fhircat.github.io/fhir-rdf-playground/?axes=rdvCh&manifestURL=https://gist.githubusercontent.com/ericprud/8e53eef196ccdc2c43f40238fdd06691/raw/224261f5055a3980acd79570fe5caeaf4a4b2d84/osteo-manifest.json
> >
> >
> > Solbrig et al demonstrate how the SNOMED hierarchy can be used for
> > valuable clinical insights ⁶ *iff* we can work write OWL axioms which
> > simultaneously access the SNOEMD hierarchy and the codes in the
> > paitent data. But as Jim demonstrated, that requires OWL axioms that
> > reference the forbidden rdf: namespace.
> >
> > Thoughts? Advice?
> >
> >
> > ¹ https://www.w3.org/TR/rdf-primer/#collections
> > ² https://github.com/w3c/sparql-12/issues/46
> > ³
> https://www.w3.org/TR/json-ld11/#example-82-specifying-that-a-collection-is-ordered-in-the-context
> > ⁴ https://gist.github.com/balhoff/62fb8f2c1e29bc0d4d27c3df0d005154
> > ⁵ https://fhircat.github.io/fhir-rdf-playground/
> > ⁶ https://github.com/BD2KOnFHIR/BLENDINGFHIRandRDF
> >
>
>