For full text search please use the '?' prefix. e.g. ? Onboarding

Pat Hayes

Re: RDF 2 Wishlist

On Nov 1, 2009, at 11:51 AM, Sandro Hawke wrote:

So, what should W3C standardize next in the area of RDF, if anything? OWL 2 added a bunch of stuff to OWL that users wanted and implementors were willing to tackle. Are there things like that around RDF?

Well, I just used the invited talk to give my speech on this topic,
but here's a more concrete list. I reserve the right to think of more
tomorrow.

  1. Clean up and simplify RDF abstract syntax. Literals can be
    subjects, properties can be blank nodes, etc. In fact, anything can be
    anything in the triple space, in the conceptual graph model. The
    existing semantics can handle this without change. The spec gets to be
    shorter and easier. 1a. All literals are typed, plain literals are understood in
    retrospect to be an abbreviation for typing with rdf:text. 1b. If XMLLiteral is necessary, make it into an ordinary datatype. Or
    get rid of it altogether.

  2. A better syntax than RDF/XML. I also like turtle or N3 style, but
    whatever, striped syntax is just too remote from the actual graph
    model to be useful, and it imposes subtle and harmful influence over
    the whole language (eg it was why we don't have literals as
    subjects.) FWIW, this is likely to be the hardest problem for
    backward compatibility, I suspect. I wonder how hard it would be to
    have a converter from RDF/XML into some other form, as a W3C service?

  3. Add a scope mechanism to the syntax, along the lines of the
    'surfaces' idea in my talk. (Terminology and exact syntax for this can
    be adjusted to suit, but parentheses work well :-) This is an
    extension to the current model, existing RDF documents are understood
    to be on a document-level surface that was always there. As well as
    making bnodes respectable, this provides a clean foundation for a
    whole lot of other things, including: 3a. named surfaces (aka named graphs) 3b. negation (see my ISWC talk) and hence full FOL as a natural and
    well-understood extension (without any retro changes to existing RDF)
    This is close to my heart, but not of immediate priority: Im more
    concerned that nothing is done which would break this as a future
    pathway. 3c. NAF negation. I didn't go into this in the talk, but the big
    problem with any kind of nonmon reasoning on the Web is that nonmon
    (notably NAF) only works inside fixed boundaries, and so far we don't
    have a way to draw the boundaries. Scoped surfaces give us this, and
    so we could have for example 'closed' surfaces which support NAF
    reasoning, and the whole logic can then be monotonic. Best of both
    worlds, and IMO a better solution that the RIF idea of multiple
    logical 'zones' with conflicting semantics. Though it would be
    straightforward to link it to the RIF spec, I am pretty sure (the idea
    would be, for RIF, that it takes its logic-zone cue from the 'color'
    of the surface it finds the rules on. All written in RDF, of course :-) 3d. Named surfaces provide better mechanism than reification for RDF- metadata-of-other-RDF (It is token level rather than abstract graph
    level, which suddenly makes sense.) 3e. Cleaner model for such things as unasserted annotations in OWL,
    use of RDF collections to encode OWL syntax, etc.., where some of the
    RDF isn't really being, like, asserted, exactly.

  4. Deprecate RDF reification, containers.

  5. (related to 4) Choose a recommended way to encode n-ary relations,
    explain it fully, and deprecate all the others. Or, generalize RDF
    triples to n-tuples syntax, with the previous as a backward
    compatibility hack for old RDF.

  6. Build a clean version of SKOS into RDF, to replace RDFS. Current
    RDFS is then OK but deprecated in favor of translation into newer
    vocabulary, eg you should actually say that you want subclass/property
    to be transitive, when you do. In the semantics, this is done by
    changing various 'iff's to 'if's, making reasoning searches easier.

  7. Include a modern, nuanced version of owl:imports into RDF. Should
    be possible to import only a part of a large ontology.

  8. Related to 6 & 7: provide some vocabulary and guidance for how to
    link RDF data without falling into the owl:sameAs black hole. Note, we
    won't completely solve this, but even a beginning would be a huge
    improvement over the current situation, and this should be done at the
    RDF level rather than a forest of conflicting conventions growing into
    use.

  9. Provide some way (an RDF vocabulary?) to support versioning,
    deprecation, etc., if only some best practice recommendations.

Pat

Backlinks