- From: Bijan Parsia <bparsia@isr.umd.edu>
- Date: Sun, 8 Jan 2006 15:54:47 -0500
- To: Enrico Franconi <franconi@inf.unibz.it>
- Cc: semantic-web@w3c.org, "Jeff Z.Pan" <jpan@csd.abdn.ac.uk>, Alan Rector <Alan.Rector@manchester.ac.uk>, owl@lists.mindswap.org
On Jan 8, 2006, at 3:29 PM, Enrico Franconi wrote: > On 8 Jan 2006, at 21:05, Jeff Z. Pan wrote: [snip] >> This distinguishes the punning semantics from many other semantics, >> such as the OWL FA semantics [2], Hilog semantics [3] and RDF >> semantics [4]. > > I'm not sure that the mentioned alternative to punning are really > better: [snip] > - Hilog captures more expected inferences than punning but (a) Peter s/Peter/Jeff/ ? > didn't say that it still leave some expected interesting inferences > out (see Boris' paper for an example), and (b) it requires a > re-implementation of the OWL reasoners, as Peter noticed. Ibid? > - The only approach that would capture exactly meta-modelling with > OWL-DL is undecidable (see Boris' paper) - this fact was hidden by > Peter... [snip] Ibid? Or am I missing an email from Peter? From various conversation with people who use OWL Full, and some introspection, I see two primary, if only current, uses of higher order like constructs (be they annotations, punning, or some more full blown species of metamodeling): Metadata about the "symbolic artefact", e.g., who wrote these axioms, when, when last modified, etc. and for ontology alignment (e.g., I modeled Wines as a class and you as an instance). I am not saying that these are the *only* uses of higher order like constructs, but they are *in my experience* what get mentioned. Only the latter has potentially interesting modeling impact, and, in practice, people are just happy to be able to *mark* these alignments and let some other piece of software (usually not a reasoner!) take care of, e.g., conversions of data between ontologies. Since neither of these involves significant cross level entailments, it seems like some version of punning suffices conceived as a liberalization of annotation properites. (This also has the advantage of typically only requiring modification of the input and output of tools, esp. reasoners.) Now, I'm not sure that this is *ideal*. There may be some form of metamodeling with the legs to make it worth adopting even at the cost of revamping tooling. This doesn't seem to be 1.1 material though. It would be nice if whatever we slide into 1.1 was neatly extendable to the Next Great Metamodeling Thing, but eh. We do what we can :) (And actually, most can *probably* by tying more interesting semantics to the presence of certain bridging axioms. This suggest we should *not* overload things like SameIndividual for alignment purposes, but instead introduce new sytnax (e.g., correspondsTo).) Cheers, Bijan.
Received on Sunday, 8 January 2006 20:54:55 UTC