% notes from JC telecon 2/10/04 % by Benjamin Grosof participants: Mike Dean Benjamin Grosof Sandro Hawke Ian Horrocks Peter Patel-Schneider Deborah McGuinness Drew McDermott Gerd Wagner Harold Boley Said Tabet Pat Hayes (joined late) o news: OWL and RDF have become official W3C recommendations Sandro: the doc's are pretty much frozen, but in addition there is a process to collect errata and filter for which corrections are "normative" -- look near top of the top-level pages o Drew McDermott: introducing DRS (see also his recent posting to sws-ig list) similar to the OWL Rules language, slightly different motivations and design decisions, thus hope they can evolve to converge. conjunctions are a class, and so on a regular formula is an object "disguises" atomic formulas or triples latest version can have SWRL near bottom-level in this regard DRS gives explicit quantifiers, unlike SWRL which has implicit outer universal quantifier invented it to represent planning, cf. PDDL students have extended it to use it for ontology translation Said Q: what would you like to see different/new in SWRL? Drew A: liked the SWRL having an explicit class for variables, so changed to use that, can declare types, would like less tie to Prolog syntax wrt the above issue of implicit outer universal quanitifier, e.g., to have quantifiers or lambda binders for variables. E.g., in OWL-S sometimes preconditions have existentials. Benj: the plan all along for future RuleML directions was to have explicit quantifiers, converged hopefully with SCL/FOL syntax, including to handle Lloyd-Topor kind of expressiveness in LP's, e.g., existential subexpressions in body or universal subexpressions in the head. Drew: fine, agree with doing that, view the version omitting explicit quantifiers as simply shorthand for the version with explicit quantifiers. Benj: so we could summarize by saying the time has come soon to put in explicit quantifiers. Ian: yes, often find myself putting in explicit quantifiers to understand what SWRL rules say. (some discussion of planning and decision procedures and complexity) Ian: we don't have a good fast procedure even for OWL-DL, much less SWRL of course one can use a generic FOL theorem-prover Benj: so applications may want to avoid using too much OWL expressiveness in combination with rules Sandro: what about putting OWL into a rule system? Ian: see work by Raphael Volz where he did some experimental implementation including using DLP cf. WWW-2003 paper plus skolemizing and axiomatizing equality. He found quite a lot of what's in the DAML ontology library is in this expressive class, but much of this is legacy stuff from much simpler languages. Benj: as suggested in the DLP paper, those experiments show that use of rule system techniques can be advantageous in particular when there are large numbers of instance facts Ian: if the non-instance ontology axioms don't change too often Harold Q: are binary predicates -- which can map more easily/directly to RDF -- distinguished in DRS (syntactically) from n-ary predicates? Drew A: not really; could use type declarations to distinguish them (some discussion of dark triples; Drew: just reify atomic formulas not everything) Sandro Q: handling of variable declarations is a bit tricky in SWRL too, wrt interactions with general semantics of RDF Drew: pretty much assume that you have an entire document and you can't take any subset of it -- don't like the RDF design spirit that wants to be able to take any subset of it and make sense of that