gist does not meet OWL DL spec; import SKOS?

[dylan-sa]

Recently we had some internal discussion at SA about which OWL profiles gist is consistent with. Robot's validate-profile tool reveals that gist would meet the OWL DL spec but for the fact that it uses annotation properties like skos:definition and skos:prefLabel that are currently "undeclared" in the ontology. (gist also does not meet RL, but that is because the "distinctionary" pattern that we employ for many formal definitions is not expressible in RL.)

To meet the DL spec, one option is to import the SKOS ontology; then properties like skos:definition would be declared in gist. There has been some discussion of importing SKOS in the past--e.g., in #373. What do folks think of this option?

[rjyounes]

The issue you cite is not terribly relevant here; I recall another issue related to importing SKOS that that we decided not to act on, but I can't find it nor remember the counter-argument. I don't think there's a choice here: if we want gist to be DL-compliant we need to import SKOS.

[uscholdm]

I often import skos into my enterprise ontologies to stop PRotege from adding them.

Is there a real practical advantage of staying in DL, if the only departure is two undefined annotations? I can't imagine it matters.

[uscholdm]

gist also does not meet RL, but that is because the "distinctionary" pattern that we employ for many formal definitions is not expressible in RL

Is the workaround here to just use subclass instead of equivalent class?

[rjyounes]

Let's not bring RL into the discussion now. There are several ways in which gist goes beyond the RL profile. We are working on this for a client, and we have an open issue for gist to make the subclass assertions in a separate file. So let's limit this issue to DL.

[dylan-sa]

Is there a real practical advantage of staying in DL, if the only departure is two undefined annotations? I can't imagine it matters.

This is a good question. Arguably this particular deviation from DL doesn't matter much in itself b/c it's not like there are inferences we're expecting but not getting. At the moment, there are just a handful of properties we use in the SKOS and SHACL namespaces that cause the issue. However, if we would generally like to maintain gist in DL (for the purpose of being able to confidently rely on DL reasoners), there is value in weeding out these undeclared property reference errors so that we can more easily catch more impactful deviations from DL in the future (as unlikely as they might be).

[rjyounes]

Oh, it's SHACL as well. While I wouldn't be much opposed to import SKOS, I hate the idea of importing all of SHACL into gist just to use sh:declare and sh:PrefixDeclaration. (Too bad you can't, as in programming, import only portions).

It also seems that other reasoners we typically use are not balking at the lack of import, since we don't see these errors when running reasoning in Protege. @dylan-sa Does robot validate-profile by any chance have an option that would silence these errors?

I change my vote to not importing.

[rjyounes]

It turns out importing SKOS, as opposed to adding the declarations as Protege does, will not solve the problem, because SKOS itself uses dcterms predicates without importing dcterms. Interestingly, it doesn't help if gist imports dcterms - it has to be imported by SKOS. And we don't want to get into that downward spiral.