% notes from JC telecon 7/29/2003 % by Benjamin Grosof Participants: Mike Dean Benjamin Grosof Ian Horrocks Sandro Hawke Pat Hayes Harold Boley Said Tabet Deborah McGuinness Peter Patel-Schneider (? perhaps one or two others) agenda topic: requirements and basic approach to layering rules on top of OWL seeded by msg sent out by Mike Dean earlier today initial comments by Mike Dean reviewing that Mike: propose restricting rules to what stacks nicely on top of OWL, and thus keep it to binary unordered relations and all names being URI's; this will help serve goals of DAML and OWL users Pat: yes, it's non-trivial to do n-ary on top of OWL Benj: rules are more mature in deployment than OWL or RDF, it's easy to do n-ary in Horn/Datalog, so let's not keep it restricted to only using OWL and RDF; rather, we can have rules help pull users to adopt OWL and RDF Deb and Benj: hopefully we can do both relatively soon, in V1 of Rules Ian: i.e., have Lite (binary, on top of OWL) and Full (with n-ary, not necessarily on top of OWL) versions Pat: we're dealing with intersection of rule and ontology languages and DAML program objectives -- let's try to keep things smaller to make it more doable Harold: n-ary unordered, with slots/roles, is very close to RDF binary; also can use collections Pat: one potential requirement is to transmit rulesets as RDF graphs Deb and Pat: can view all this as baby step that's in part political and is outreach to the DAML community Benj: an important concern is not to make later versions more difficult, i.e., anticipate moving later to n-ary, nonmon, procedural attachments, etc. Benj: there will also be issues for layering on OWL that are more hard to work out, that will have to be deferred as well Mike: yes, I do anticipate some issues at semantic level e.g., unique names assumption, explicit equality and how it's treated in unification Ian and Mike: rules with property chaining on top of OWL-DL immediately loses decidability all: need more than DLP, e.g., want rules to do property chaining discussion around LP (least Herbrand) vs. FOL semantics for Horn, to add on top of OWL-DL - LP is a weakening of the conclusions (rather than the premises) - concern: do we need to weaken the conclusions of the OWL-DL subset of a ontology+rules language Benj: it's not well studied what are the expressiveness/semantics of either: - Horn FOL + full DL (tho' this is known to be a subset of FOL) - Horn LP + full DL - (tho' we know Horn LP + DLP is just LP) but the former is better understood from previous research Mike: what are the advantages of restricting to LP semantics? Benj: primarily further extensibility to nonmon and procedural attachments; also congruence with existing LP rule (and DB) systems Harold: also some aspects of recursion Mike: we don't want to break OWL content discussion around incompleteness of practical systems, and expectations about that %%%% ***consensus: 1st step: OWL-DL + binary Horn FOL + warning label warning: (phrased by Benj): do not rely on the diff between FOL - LP semantics (i.e., entailments) if you want (as is common in most practical rule applications) to later extend to nonmon or (some kinds of) procedural attachments warning is to developers and also to users (tho' users are less likely to read the warning) Benj: there's a radical issue lurking however, which is that it's not well understood how to coherently restrict all of (OWL-)DL to LP semantics when adding Horn LP (rather than Horn FOL) to DL Benj: wrt warning label, we will need to give some explanation that some kinds of conclusions and some kinds of premises will likely not be supported when extending to more fully featured rule language that has nonmon and (more kinds of) procedural attachments - Mike: it will help to give users examples (for exposition) Mike: we can have different warning labels for different audiences %%%% discussion: can view a lot of this in terms of incompleteness - background: Horn LP semantics is a weakening of Horn FOL semantics. I.e., it sanctions as conclusions a subset of what FOL sanctions as conclusions; that subset is simply the set of ground atomic facts. some discussion about whether LP semantics is "syntactic" from viewpoint of FOL semantics - Pat: dislike calling LP semantics "semantics" since it is not based upon model theory in the same style that FOL semantics is - Benj: observe that LP is a coherent KR in itself, and its use of "semantics" there is standard terminology in its literature. In particular, I just dislike how LP literature defines "model"; rather, I prefer personally to speak of the "conclusion set" that is entailed from a given set of premises Sandro: the incompleteness issue means we could be talking about OWL-Full not just OWL-DL, relative to adding Horn rules Plan: Mike and Benj will draft a next version, send it out and solicit more comments from those committee members not on today's telecon, then we'll discuss this as part of the next telecon.