Re: Cross-ontologies reasoning

   [me]
   >* and so forth
   >Translating back and forth can be done by straightforward deductions.

   [Evan Wallace]
   and this is overly optimistic.  Add to that list:
   * One ontology represents a concept as a class, the other as an instance

   * A qualifying attribute in one ontology translates to an additional class 
   with a different attribute in another ontology

   and many more.

Well, I don't see "and so forth" as necessarily more optimistic than
"and many more."

   Different factorings like this will be common.  We see them all the time
   in models of overlapping domains.  For example, right now there are 
   discussions on a manufacturing integration standards list which are aimed
   at aligning planning and execution views of manufacturing processes.  Both
   of these models cover many of the same entities, but different things are
   important to each.  

I agree completely.  When I say "deduction," I don't mean the stuff
you can state in OWL.

   Different factoring of concepts has made it difficult
   even for people to see the mapping.

True.  That's why I think most of the work on finding ontology
"mappings" automatically is of theoretical interest only.  For many
overlapping ontologies, a committee of several humans will be
necessary to find how expressions in one are to be approximated by
expressions in the other. 

   Assuming that we can create mappings between ontologies, how can we test
   their quality.  One concern I have is preserving consistency.  One ontology 
   may well have constraints (disjointness for example) that another doesn't.  
   How can we ensure mappings are restricted between these two ontologies such 
   that importing data consistent with one through a mapping to another won't 
   violate local constraints (KB pollution?)?

Waldinger and Stickel have used SNARK to find inconsistencies in
ontologies.  They found bugs in early versions of DAML+OIL.  (Sorry I
can't seem to locate a reference to some of their work.)  I think
using a theorem prover for this task makes perfect sense.  Granted it
will take a while, but it doesn't have to be done that often.

Bureaucratic note:

This interchange, while interesting, probably belongs on www-rdf-logic
rather than public-sws-ig.

-- 
                                   -- Drew McDermott
                                      Yale Computer Science Department

Received on Tuesday, 16 December 2003 20:36:42 UTC