Datalog rules and triples in RDFox

[dyamon]

Setting

The problem came up while investigating why RDFox didn't seem to provide any automatic derivation of some obvious axioms (e.g., every class in a subclass of owl:Thing, every role is a subrole of owl:topObjectProperty, etc.).

Relevant documentation

• RDFox docs

Problem

Let's say we have an axiom stating that A is a subclass of B.

Importing this axiom in RDFox as a triple ([:A rdfs:subClassOf :B]) will cause the solver to derive also some additional information, such as

[owl:Nothing, rdfs:subClassOf, :A], [owl:Nothing, rdfs:subClassOf, :B], 
[:A, rdfs:subClassOf, :A], [:B, rdfs:subClassOf, :B],
[:B, rdfs:subClassOf, owl:Thing], [:A, rdfs:subClassOf, owl:Thing]

The problem is that we are converting axioms into Datalog rules and hence we have something like this

[?X a :B] :- [?X a :A]

Doing this we lose the esplicit information about class subsumption and we don't get the same additional derivations as before.

[dyamon]

Now we need to decide if this is really bothering us, and if it does, how to solve it. Note that this is somewhat connected to #1

Solutions

• One naive solution is to introduce additional rules (similar to the ones implicitly added by RDFox) to fill the gaps in the derivation. This has a negative impact in the materialization process.

[dyamon]

Looking at this more closely, this is not something we should care about. In the "implementation" provided by the paper, our same rules are given to DLV, which, as a Datalog reasoner, does not provide any additional information/derivation either.

The only thing we need is the axiomatization of \top which is provided by these additional derivations. We will achieve this in a more straightforward way (see #1)