This is a page to discuss various aspects of the design of fragments of OWL. Currently, all the proposed fragments are subsets of the OWL 1.1 submission for which some interesting inference problem can be computed in polynomial time with respect to either the number of facts in the ontology or the size of the ontology as a whole.
Other fragments of OWL have been defined by implementors:
With the advent of the WG some members believe named fragments of OWL Full are important. At the time of chartering, it had been suggested that the OWL 1.1 WG should NOT be the place for this to happen, but it was overruled by W3C, so this is all and only the place where this can be done within W3C OWL specs, so we must consider it. Those of us advocating believe an important user group is the Web Application Developer, who is willing to use a small amount of OWL if it helps, does not often have a classfication/realization problem (yet) so are not using DL reasoners, care about scaling from a computational (rather than formal) view - i.e. being able to do map/reduce parallelism on the feature set would be a valuable plus.
One name that has been suggested for this is "RDFS 3.0" as it would be the third update of RDFS and it would resonate with being a language of use to the "Web 3.0" developer community. (additional info -November, 19 2007 - I was afraid W3C might have trouble with this name, since we are an OWL, not RDF, group, but Ivan Herman said he thought this would be okay if the WG decided it liked this name)
- - is an extension of RDFS but a subset of OWL Lite vocabulary
- - can have a clean operational semantics defined via a axiomatic specification
- - if it is closely tied to a particular formalism, then the rules and restrictions can be explained in an easy and intuitive way. Better is that the language is not tied to a particular formalism (a reason why a axiomatic, as opposed to model theoretic, formalism may be prefered)
- - is syntactically as well as semantically easy to learn/use
- - does not require the understanding of DL restrictions to use (i.e. is a subset of OWL Full) - although if there is a way that many of these can naturally be covered by DL-based OWL reasoners that is a plus
- - allows efficient and scalable implementation. Scalability is certainly one key requirement to push this fragment to an enterprise (or web) level.
The following subset of language terms, similar to those included in Oracle's OWL Prime, is the suggested starting place. Specifically we could extend RDFS with:
- '"Annotation" Properties:'
- differentFrom (and allDifferent)
- Property Characteristics:
Some people have suggested we include class relationships - intersection, union and complement - but only for named classes.