From W3C Wiki
Jump to: navigation, search

What's up with RDF, RDFS, OWL, OWL Lite, OWL DL, OWL Full...? Is each in some sense a superset of the former?+

Full OWL is something like the spiritual successor to RDFS, in that it continues the 'anyone can say anything' tradition of RDF, boosted by the OWL extensions to allow some very powerful generalisations about RDF vocabularies and data. OWL DL and OWL Lite are careful restrictions of this powerful language, created to ensure that inference/logic tools can guarantee to draw all allowed conclusions from the data at hand. OWL Lite and OWL DL are designed with similar tools in mind, but take different choices regarding expressivity versus implementation cost.

I'm not sure of the etiquette here - I hope interjecting comments like this is ok. There seems to be some doubt that real DL tools can, in practice, be complete. JJC is good person to ask about this. Secondly, I think OWL DL is a superset of Owl Lite - it's true that the choices are different, but I think the superset relationship would be worth bringing out.

RDFS (RDF Schema) in recent times has partially rebranded as a practice rather than a technology, the practice of RDF "vocabulary description", of which OWL is our 2nd phase. OWL vocabulary descriptions are called 'ontologies'. OWL basically amounts to a language for making generalisations, in much the same way that the RDF core specs provide a means for making simple assertions. In plain RDF, you can say things like 'john is an unmarried man'. In OWL, you can make certain kinds of sweeping generalisation, such as saying 'Something is a Bachelor if it is a Man that is unmarried', or that people are Male or Female (but not both). As a 'semantic extension' to RDF, this is less expressive than a full Semantic Web logic language (such as those suggested by N3, Lbase, RuleML etc)., but still far more powerful than the inbuilt facilities of core RDF, which offers (through the RDFS vocabulary) only the most basic forms of generalisation (eg. Persons are Living Things).

With this increased power comes responsibility: the ability to say such things creates opportunities for complexity, confusion and contradiction. OWL generalisations can interact in unexpectedly rich ways; reasoning about these interactions is difficult. For such reasons W3C defines two limited profiles of Full OWL, 'OWL Lite' and 'OWL DL', which provide very careful restrictions of OWL that draw upon recent academic and industrial research.

OWL Lite is optimised for ease of implementation while still allowing tools to draw many conclusions from RDF data that are not guaranteed to be answered by plain RDF Core tools. In simple terms, this means that OWL Lite-aware systems will often be better at answering questions about RDF data. OWL DL systems will be better still, but at a greater implementation cost, since OWL DL allows more of the full expressiveness of OWL, ie. more ways of stating generalisations about the world. This means that OWL DL systems are both 'smarter' (better at answering questions) but harder to implement. OWL Full is simply the full combination of RDF and OWL vocabulary descriptions without the safeguards imposed by the two profiles, ie. Lite and DL. In OWL Full you don't have to be so careful about what you write, but you pay the price: this greater expressiveness means you give up guarantees from tools about the kinds of questions they'll be able to answer...

All this still leaves us with a rich (too rich?) choice of terminology. This refects the multiple communities that have come together to collaborate on the development Semantic Web. Earlier RDF discussions used 'Schema' as a noun, eg. when talking about 'an RDF schema which...'. More recently we're used 'vocabulary' in this role, with 'Ontology' being a common noun for a kind of vocabulary we use in the Semantic Web. We still talk of the RDF 'schema language' though, but not so often of RDF schemas. This is because RDFS vocabulary and OWL can be freely mixed, and there are no things that might be called "RDF Schemas" which couldn't equally well be called Ontologies (and vice-versa).

Notes on OWL and RDF, originally by DanBri in reply to questions from Ivan Herman.

See also OwlTalks, OwlPuzzle

see also: diagrams by hayes at March tech plenary in Cambridge (LinkMe)