owl:intendedProfile (proposal for ISSUE-111)

Here's a possible solution for ISSUE-111 [1], with some variations and
discussion.  I was only considering the RDF/XML serialization here.  For
the XML serialization, a MIME type could a good solution.

* Basic Approach:

Use an ontology property.  So the users adds a triple like this:

  <> owl:intendedProfile owl:DL.

This would have processing-model semantics, much like owl:imports.  The
processing model is something like this:
   1. You fetch the content from some URI U.
   2. If its Content-Type is "application/rdf+xml", proceed.
      (otherwise, this procedure doesn't apply; it's not OWL
      in RDF/XML.)
   3. Parse it into an RDF graph.
   4. Look for any triples matching { U owl:intendedProfile ?x }
   5. If there are zero or >1 distict values for ?x, then
      this algorithm does not apply.  Fall back to whatever
      you did before OWL 2.
   6. If you implement the profile named by ?x, use it.
      Otherwise, you MAY proceed using some other profile that you do
      implement, but you SHOULD warn the user about how the results may
      be different.  (You may be able to characterize, for the user, how
      the results will be different and ask them whether to proceed.)

That seems pretty simple, and I don't see any problem with it.  The key
point is that the failure mode in step 5, which also occurs if you
eliminate the triple by some kind of RDF entailment, leaves us no worse
off than we are in the OWL 1 world.

Some variations....

   1.  Allow one to specify multiple profiles.  In this sense, they
       become more like "Features" or "Semantic Extensions" than

            <> owl:usesSemanticExtension owl:DL.
            <> owl:usesSemanticExtension owl:DL-Safe-Rules.

   2.  Perhaps #1 has to be done with an RDF List, since if it were
       done with just more triples, you wouldn't be able to detect the
       case where the triple naming some extension was removed.  You'd
       think you knew the exact intended semantics but would be wrong.
       I'm not sure if this is a problem -- we don't seem to mind it
       with owl:imports.   Anyway, if it's a problem, we can do
       something like this:
           <> owl:usesSemanticExtensions ( owl:DL owl:DL-Safe-Rules )

   3.  Call it "imports", or something like "imports".  As a rules user,
       it seems perfectly natural to me to say that RDF graphs have only
       RDF Simple Entailment until/unless they "import" various
       semantics.  I picture import as loading the rulebases which
       compute entailments under those semantics, but I would expect
       that in some cases (eg owl:DL) there would be vastly superior
       implementations that simply recognized the owl:DL URI and turned
       on their local DL reasoner.

           <> owl:imports owl:DL  # this ontology uses DL
           <> owl:imports owl:DL-Safe-Rules    # and also these
           <> owl:imports <some_std_facts>.
           <> owl:imports <some_other_ontology>.

Ideally, one could have an ontology which uses owl:DL and imports
another ontology which use owl:OWL-R.  My sense is that allowing that
would complicate matters a lot -- you couldn't just compute the imports
closure and then hand it off to the reasoner! -- so we probably want to
make that an Extra Credit thing for now, saying system MAY do that, but
are also free to simply report an error, that ontologies are being
combined which use incompatible profiles.

Oh...   Another idea: users might want to say "at least" OWL DL-Lite.  

      Ontology A:
           <A> owl:exactSemantics owl:DL.
      Ontology B:
           <B> owl:exactSemantics owl:DL-Lite.
      Ontology C:
           <C> owl:minimalSemantics owl:DL-Lite.

Now, A+B could only be run by some very clever (impossibly clever?)
system which somehow combined A and B keeping their separate intended
semantics.   Is that even well defined?  I dunno.   Avoid this case.

But A+C could be handled, no problem, by an owl:DL reasoner.


Somewhere in here, it would be nice to tell people they SHOULD NOT (even
MUST NOT?) rely on the places where these various languages are not
neatly stacked semantic extensions.  So that, while falling back from
owl:DL to RDFS semantics obviously makes the reasoning incomplete (with
respect to the intended DL semantics), the fact that in some corner
cases it also makes the reasoning unsound..., well, we'd like that not
to cause real-world problems.

      -- Sandro

[1] http://www.w3.org/2007/OWL/tracker/issues/111

Received on Tuesday, 22 April 2008 20:00:56 UTC