3

I've been working on an Ontology in Protege.

I started with a high expressivity that was classified in Protege as "DL Expressivity: SROIF(D)".
If I'm correct, the letters indicate -in order- the use of: Transitivity, Role chains, Nominals, InverseOf, Functional, and the D stands for Data Types.Unfortunately, such expressivity should be NExpTime-complete and I need something which is less heavy to compute.

So, I tried to reformulate my ontology with the aim to stay within the OWL2 EL Profile (for which PTime decision procedures should exist). Now I have "DL Expressivity: SRO(D)" as I'm using Transitivity, Role chains (with no recursive definitions), and Nominals, but no Inverse or Functional object properties.

My question is:

Is it really the DL Expressivity SRO(D) within the OWL2 EL Profile?

The correspondence between the expressivity as given in Protege and the OWL2 recommendations is unclear to me. Also, it seems conflicting with the info that I get here: http://www.cs.man.ac.uk/~ezolin/dl/ .

According to the OWL2 EL Profile, role chains should be allowed in SubObjectPropertyOf statements. And I'm using simple role chains of the form a o b -> c (nothing of the form a o b -> b or a o b -> a). But according to general knowledge, role chains should be avoided.

I read extensively on OWL, OWL2, Description Logics, Reasoners, and some benchmarks and evaluations of reasoners etc. but I couldn't find any "final" resource that clearly untangles this aspect.

Second question:

Is there any other "variation" of the OWL2 EL Profile (for example, a slightly different combinations of operators that could be used) that is still PTime? And where can I find this studies?

aky
  • 75
  • 4
  • For your information, Protégé does not provide very precise fragments. As an example of its limitations, the expressivity classifier that Protégé uses only looks at the features that are used in all class and property expressions, regardless of whether thy happen in the left-hand side or right-hand side of a concept inclusion. That excludes the possibility to detect OWL 2 QL or OWL 2 RL ontologies. For OWL 2 EL, the problem is that Protégé considers the simplest fragment to be ALC, which contains features not allowed in EL. – Antoine Zimmermann Aug 20 '18 at 12:04

1 Answers1

2

Does SRO(D) fall in the OWL 2 EL Profile?

No. OWL 2 EL is based on EL++, which excludes the following

  • full negation
  • universal quantification
  • disjunction

SRO(D) extends ALC and since ALC includes all of the above, SRO(D) will fall outside of OWL 2 EL.

PTime variants of OWL 2 EL?

There are various extensions to EL, for which reasoning is PTime. For papers on EL++ see EL++ and EL++ update. For looking at EL variants see for example EL tractable extensions.

Henriette Harmse
  • 4,167
  • 1
  • 13
  • 22