Look into the use of a reasoner

[bartkl]

With the right reasoners we can infer RDFS and OWL statements, which when done right might help us make our jobs a lot easier.

Things like subclassing can imply certain validations that are not structurally explicit, and therefore tend to be ignored.

This nreeds a lot of thinking and refinement though.

[bartkl]

Few things I've learned, about SHACL:

• SHACL inferencing does not mean inferring triples from sh:and and other of its semantics, but refers to inferencing of SHACL rules. This is not what I hoped.
• Many tools seem to tightly couple the implementation of those core semantics with the end goal of validation. This means I can't easily use implementations from others.
  • You might, however, be able to find implementations (in SPARQL or Java, say) of these semantics that you can use, but it's certainly doesn't seem easy. Perhaps looking at the RDF4j source code helps?
    • Search for NodeShape and you'll find an AST for SHACL SAIL in RDF4j's code base.
    • An example is found for logical constraint components.

It's good, also, to remind ourselves of the original view, by contrast:

• Don't use inference, but keep it simple by mapping from SHACL to target schemas purely structurally.
• This means we don't use the semantics of SHACL, only map them onto something that conveys the same meaning.
  • It still feels weird to me though...
• We want to depend only on SHACL shapes, and possibly only optionally extract extra information from the vocabulary.
  • This also means the SHACL graph represents a selection (as does the optionally present vocabulary) of things you decidedly want to see in the target schema
• Using inference (when a vocabulary is present) might cause very wild results, although I'm not sure if that's really the case.
  • Therefore, what I should really consider, deeply, is the subclassing case.

To be continued.