Redundant axioms?

[dlutz2]

Found 8 instance where there was both a simple subclass axiom as well as more Aristotelian equivalent classes axiom for the same class i.e.
A is a B
A is a B that C's.

• AgentOntology
  • CivilOrganization
  • CommercialOrganization
• ArtifactOntology
  • ImagingInstrument
  • OpticalInstrument
  • PortionOfCoolant
  • WireReceiver
  • DishReceiver
  • PatchReceiver

Not sure about CCO policy/convention re this type of stuff. let me know if this is too much.

[swartik]

At least for some of those, I think the “A is a B that C’s” is necessary. These are equivalence axioms, not subclass axioms. Consider: CivilOrganization ≡ cco:Organization and (cco:has_role some cco:CivilianRole) Without the cco:Organization qualifier, a reasoner would infer that person with a civilian role is a civil organization. I didn’t check all the property domains, so you may be right about other classes.

[dlutz2]

Sorry, should have been clearer. I intended to say that if you have both the subclass and the equivalence you don't need the subclass:
A ≡ B that C's implies A is a B.

By policy/convention you might always require a purely taxonomic backbone. Certainly for readability/comprehension have the explicit subclass axioms makes the overall structure clearer.

re: domain/ranges there were also about 30 domain/range axioms which can be inferred from other domain/range + inverse properties and sometimes a subproperty e.g.
ObjectPropertyRange( has_recipient Agent ) + InverseObjectProperties( has_recipient receives ) => ObjectPropertyDomain( receives Agent )

Removing these would drastically reduce the readability of the ontology even if they are redundant from an entailment perspective.

[mark-jensen]

Interesting discussion. Thanks.

We don't currently have a policy in place re redundant axioms. My guess is that we land more on the side of readability unless there is a performative concern. I have been told that certain applications cannot work with equivalency axioms, eg- treating the ontology as a graph, thus having the redundant subclass assertions may be helpful to some users.

[dlutz2]

Shouldn't be any performance issues, readability is a good thing. I'll close this as resolved.