RIF Design Roadmap Draft 2006-02-27-6PM ~~~~~~~~~~~~~~~~~~ Harold Boley (NRC), Michael Kifer (Stony Brook University), Axel Polleres (DERI), Jos de Bruijn (DERI), Michael Sintek (DFKI), Giorgos Stamou (NTUA), Jeff Pan (University of Aberdeen) This preliminary design roadmap aims at a consensual system for RIF Phase 1 features distilled from the current UCR draft, the RIFRAF, and the public archive. It builds on the Design Goals (cf. http://www.w3.org/2005/rules/wg/wiki/UCR/Design_Goals). An approach to RIF Phase 2 is sketched as well. PHASE 1 ======= 1. Specify syntax and semantics of Horn Logic and sublanguages (cf. http://www.ruleml.org/modularization/#Model) 1.1. Full Horn Logic (functions, no negation) 1.2. Datalog (Horn Logic without functions, no negation) 2. Syntactic and semantic extensions of Horn Logic 2.1. Define purely syntactic extensions (cf. http://www.w3.org/Submission/SWSF-SWSL) 2.1.1. Monotonic Lloyd-Topor extensions (disjuncts in body, conjuncts in heads) 2.1.2. Named arguments (slots) in n-ary notation - Can be developed into frames with OIDs (as in F-logic) in Phase 2 2.1.3. Higher-order syntax (cf. HiLog) -> Phase 2 2.2. Support literals and datatypes (common functions and operators) 2.3. Delineate appropriate semantics for different Horn-like rules (cf. #9.2) (could be moved to Phase 2) 2.3.1. First-order (all-model) semantics (cf. #5) 2.3.2. Minimal-model semantics 2.3.3. Semantics for production rules 3. Webizing features that should be (globally) addressable 3.1. Adopting IRIs (incl. URIs, URLs) for Web-based addressing 3.2. IRI addressing of RIF constants and predicates 3.3. IRI addressing of other RIF features 4. Pure production rules with only asserts in the action part - Execution is 1-to-1 with model generation, semantics compatible with #1 - Basis for interoperation between production rules and Horn rules - Action part will then be generalized in Phase 2 5. Integrity queries, a.k.a. integrity constraints (will not require extra effort) - These constraints are considered violated if the queries have answers (an answer is a witness to an integrity-constraint violation) - The allowed queries must have the syntactic form of a rule body - Semantics of a rule body as in #1. Pragmatics of a warning or an error 6. Scope feature for the modularization/structuring of rulebases (cf. TRIPLE models/contexts, FLORA modules, named graphs) - Will enable Load-and-Query rule engines (i.e., engines that can load and then query different rulebases at once) - Units for tagging provenance etc. (cf. #9) - Basis for scoped negation as failure in Phase 2 7. Interoperability with RDF (work already ongoing) - Treat an RDF graph as a ruleset of binary or ternary facts - Treat blank nodes as existentials in rule bodies and as Skolem functions in rule heads - Accommodate SPARQL queries from the body of rules (cf. #8) 8. Interoperability with OWL (work already ongoing) - Define a hybrid combination semantics (cf. http://rewerse.net/deliverables/m12/i3-d3.pdf and ftp://ftp.cs.sunysb.edu/pub/TechReports/kifer/msa-ruleml05.pdf) At this stage, interoperation between OWL and rules will be at the level of rule bodies posing ground (after instantiation) queries to OWL 9. Metadata / semantic attributes for rule documents, scopes, rules, facts (could be moved to Phase 2) 9.1. To enable searching for rulesets, Google Directory-style 9.2. To enable tagging rulesets with intended semantics (e.g., FOL, LP/well-founded). This may not be that important in Phase 1, but will certainly be in Phase 2 (cf. #2.3) 9.3. To enable tagging rulesets to indicate syntactic features that should be supported by the recipient. This will support conformance-guided rule system interoperation. Tagging can be done by pointing to a suitable XML Schema document (cf. #1) 10. XML Serialization (cf. RuleML's serialization) - RDF can be generated via XSLT PHASE 2 ======= At this stage, allow for three subfamilies with different semantics: FOL-style, LP-style, Production rules. The specific semantics will be indicated by a semantic attribute. Define different kinds of extended RDF/OWL interoperability for the three different kinds of extended rules. The Optional extensions (cf. last sections of all subfamilies) follow Phase 1 using the extensibility notion of the Charter. These extensions could be specified in a separate RIF document working horizontally across the three subfamilies below. I. FOL-style rules 1. Based on Phase 1 2. Continued RDF Interoperability 3. SWRL-inspired rule extension of OWL-DL 4. Optional extensions - Fuzziness - Soft integrity constraints (expressing a kind of preferences) - ... II. LP-style rules 1. Scoped negations 1.1. Well-founded semantics of negation as failure 1.2. Answer-set (stable model) semantics - Can also be extended for "classical" negation Note: We might just use stable models and consider well-founded as a more tractable approximation 2. Syntactic extensions 2.1. Full Lloyd-Topor 2.2. Frame notation with object identifiers (cf. F-logic) 3. RDF and OWL-DL Interoperability - extended from Phase 1 4. Optional extensions - Disjunctive rules (using answer set semantics) - Fuzziness - ... III. Production rules 1. Allow retract and other action types as rule actions 2. Allow procedural attachments and events in rule bodies 3. Also consider actions in the body a la Transaction Logic - The latter has logical semantics as opposed to #III.1 and is compatible with LP-style semantics 4. Optional extensions - Fuzziness - ...