See also: IRC log
<sandro> Agenda for Today:
<sandro> - F2F6
<sandro> - Architecture Document
<sandro> - Core OWL and RDF Compatibility
<sandro> - RIFRAF
<sandro> - Demos
<Hassan> (Sandro - can u pls give URL pointers to relevant documents? Thanks)
<sandro> i was just typing what was on the screen.
<Hassan> (thanks - sorry for being so demanding... :-)
<sandro> scribenick: Harold
<ChrisW> Leora, correction, we will start RIFRAF at 11:30
Jos: There are several events collocated with ESWC2007 and RR2007 in Innsbruck, June 07
<ChrisW> The agenda is now on the wiki for today
Within 2 weeks (March 13) possible other proposals, with proposed dates etc.
Decision on Tuesday after.
<Hassan> (for what it's worth - we can hear no one except Christian)
<Hassan> I can
Axel: Why an Ontology for RIFRAF
... Earlier email: OWL Ontology.
...Idea: formalize discriminators in an ontology to get a classification schema.
(more: see Axel's slides)
<Hassan> Excellent work Axel!
Axel: Summary: Base Ontology and Metamodel should be the same. Wants others to join.
<sandro> Indeed! Very nice, Axel.
Leora: Base ontology should be somehow more expressive than the Core.
Axel: Gradual refinement.
<sandro> "should base ontology be more expressive than core"?
Leora: There should be some way to distinguish Core and non-Core classes.
<sandro> Chris: we're calling it the Abstract Syntax, not "Metamodel", because metamodels often have other stuff.
Chris: Yesterday we agreed to call 'Abstract Syntax' what we do in UML, not call it 'metamodel'.
Axel and Leora: OK.
Christian: Plse clarify on RIFRAF: Isn't it to help us define new dialects, using features, classes, etc., first for Core, then next family, etc.
<AxelPolleres> I would have some adhoc comment on that.
Christian: You seem to see RIFRAF as a posteriori description of dialects, not what should go into dialect.
Axel: While, yes, this was planned, then nothing has happened, so went bottom-up instead of top-down.
<AxelPolleres> ... on hassan.
Hassan: Excellent work. Use of Prolog: don't apologize, this is great.
<AxelPolleres> operational/declarative way to check rulesets was exactly what I proposed.
Hassan: At least we have operational way of checking. From informal to formal. Make experiment with a methodology. Axel proposed operational methodology (happens to be in Prolog). Way to test.
<AxelPolleres> ie. we can formalize dialects on such "programs".
Hassan: Second: Can now try to characterize what is missing, so as to suggest new discriminators.
<sandro> "operationalizing desciminators"
<AxelPolleres> +1 to allen that the semantic discriminators will be much tougher. for these we will also need additional annotations.
Allen: Very impressed by Axel's work, agree with Hassan. What about semantic discriminators? Cannot all be operationalized (e.g. halting problem).
<LeoraMorgenstern> +1 to allen, as well, and I'm taking myself off the queue, because Allen made my point
Chris: Current focus on syntactic discriminators. Would it work for semantic ones?
Allen: Would need to formalize model theory. Finite models?
Hassan: Those who define the languages must decide that.
Axel: Agree with Allen: will need semantic annotations, since we cannot see such properties from the rules themselves.
Allen: Is RIFRAF supposed to apply to Rulesets or to Dialects?
Axel and Allen: Worthwhile exercise.
Hassan: System not to be used parser or something (feeding in lang. description and generate parser).
<LeoraMorgenstern> +1 to Hassan; but the point then is to see the relationships among different languages
Hassan: Rather map language into ontology. Someone has already written discriminators, since have already understood them.
Chris: Someone gives values to these features, then get right dialect.
<AxelPolleres> that's fine with me... It is a declarative way to specify what a rule dialect (syntactivally) defines, but also (as a side-effect you want, to check a ruleset.
Chris: Axel's examples are on rules, not on dialects.
<AxelPolleres> this (adding ruleset as argument) can be refined
Hassan: Yes, but methodology can also be used for dialects.
Chris: Can you give such as classification?
Axel: Yes. Declarative way to find out if dialect fulfills features.
Chris: Constraints on rules themselves.
Domain of discours: syntactic parts of rules.
... But we want syntactic-semantic features of a dialect, not of rules themselves.
Axel: Dialects could be defined on top rules and rulesets.
<Hassan> +1 with Allen!
Allen: Need a tool to check feature-dialect mapping.
<Hassan> Good tool to experiment with...
Dave: How can you have such a tool:
chicken and egg problem.
... Notion of fully operationalizing: stay back from.
<AxelPolleres> +1 to dave Yes, it needs/is the RIF! That's why i said this is just a different way of writing down the metamodel, an that should be elaborated in parallel with the metamodel/language.
Leora: Rules and/or Ruleset? The whole set only makes an informal spec a theory.
Chris: Distinction is between Rules/Ruleset and Dialects.
Axel: Should serve as RIFRAF Core that can be gradually extended.
Chris: At least for Phase 2 of RIF we need identify (more completely the) family of rule languages (dialects).
Axel: We also need to top-down part?
Leora: Relative complex predicates to decide what goes into a language. Metapredicate talking about existence.
Chris: Think about RIFRAF questionnaire: "Does this language have this feature?"
<AxelPolleres> The approach s more: you need to define a discriminator in terms of its features, and not the features in terms of the discriminators... at least I got no input on that direction so far. and I suggested to start from the other end last time.
Leora: In last f2f we found that this is
not sufficient. Example: Various action languages.
... believed that was consensus then.
Allen: About chicken and egg problem.
E.g. production rules, parts of which can correspond to derivation
... Not impossible that there are such overlaps.
... E.g. using Jess for pure deduction.
... What Axel shows is that you gain something on the Ruleset level.
Chris: Yes, but in parallel we need Dialect level.
Hassan: Agree with both. No hope that
there is a complete characterization. Only approximations are possible.
Stepwise. Then draw a lin: E.g. "This is fine enough for our purpose."
... RIFRAF is a bridge, people can agree, and then exchange.
<csma> +1 to hassan
Hassan: with RIF.
Axel: Wanted to get started with Base
... Two outcomes: Define discrimintors by features. Continuing this exercise by inventing new features.
People can give feedback.
scribe: Second: Was meant to lay the
groundwork for others to build on. Iterations on top, to see what comes
... Obviously not do it alone. Re-confirm commitment?
<AxelPolleres> Shall I repeat, I think I was alking to fast :-)
Dave: Different grain sizes of dialects.
Originally small number of large dialects. Today's examples much more
... Might have a mechanism to avoid hundreds of dialects (by merging some?).
<Hassan> +1 to Dave on combining BU with TD classification (has been my point all along BTW)...
<AxelPolleres> just short: I need to drop out at 4
Chris: We still need more coarse-grained dialects. But Axel's fine-grained ones are still useful.
<Hassan> (Chris pls use your mike)
Chris: You can find that a given Ruleset fits into a dialect using fine-grained discriminators.
<AxelPolleres> no objection against what Chris said. As I said, it was one possible proposal, and I am definitly open to other proposals.
<Hassan> I agree that all these efforts should go concurrently
Chris: Hassan and Leora, do you agree with this disctinction?
Hassan and Leora: Yes.
Chris: Has obvious use for interchange.
Leora: Questionnaire made assumptions about defining classes, and only asking questions within these classes.
Hassan: Yes, the classes were already defined top-down.
<PaulVincent> Comment: is the idea that syntax analysis is part of the problem, vs semantic analysis (or rule languages covered by RIF)
Hassan: In beginning Hassan started bottom-up.
<Hassan> (Harold - I did not say that!)
Chris: Bottom-up (starting from specific languages) should still be continued.
<sandro> Hassan, you can correct the log with s/from/to/ substitution commands.
<Hassan> I originally proposed that we start BU, but everyone preferred TD and we have been doing so up to now!
Christian: PRR metamodel could be read as an ontology for production rules.
<AxelPolleres> it shall be aligned, no doubt.
<csma> Axel, the latest (but not so new) public draft of PRR is linked on the wiki front page
Paul: OMG allow great room to incorporate
future other representations. Bear in mind that PRR metamodel didn't
provide very many discriminators.
... Paul would be interested to participate in RIFRAF.
Chris: Pick a small set of dialects.
Axel: We should have some actions. How to
proceed? Still not sure what to edit.
... Need to complete this exercise. Others should help.
<AxelPolleres> dropping out!
Leora: Initially assigned ECA part, later done by Paula. Would prefer ECA.
Chris: Alright, this is your action now.
Allen: Some properties only apply to
language level, not ruleset level.
... Need to classify properties/discriminators.
Hassan: You as the language designer know
the properties, so you can type it into.
... You fill in values into the discriminators.
Allen: Putting in ontology.
Hassan: Yes, need to ask the right questions.
Hassan: and pragmatic.
Chris: Drop Frank's action. Ask Paula. Leora moved to section 5. Hassan more in asn06 and XML generation.
Leora: Test cases to point out usefulness, nothing else.
Chris: Will ask Paula, if she wants to
continue that action.
... What is reasonable milestone?
Leora: Useful characterization of action languages. Until some time in April.
Chris: 1st week of April.
Allen: Same here: 1st week of April.
<MichaelKifer> Discussion of the architecture document
<MichaelKifer> CSMA: time to start thinking on what should go into the arch document. Some stuff from the current core will be moved/copied there
<MichaelKifer> Dave: maybe the condition language should be in the arch doc
<Deborah> scribenick: MichaelKifer
<Deborah> zakiim, scribenick: MichaelKifer
<sandro> the first one was fine. there is no confirmation.
Harold: well, it would be sufficient if only some design principles of the condition language should go to the arch
csma: the arch should be a global view
... should have a list of basic concepts where the details will be filled in by different dialects
<Harold> Sandro, I would say the Core consists of the Condition language as a foundation and the Horn language as one extension.
DaveR: the arch will contain many features outside of the core. For instance: negation.
<Harold> Other extensions, e.g. Production Rules, can be built on top of the Condition language, also reusing aspects of the Horn language.
csma: yeah (hopefully the arch will be a repository for reusable components)
Jos: what does it mean to include a spec in the arch doc? This either makes no sense or will lead to huge duplication
<sandro> Arch and Library -- maybe different documents.
<Harold> Jos, maybe the Architecture chapter/document could be based on RIFRAF, thus give a framework to classify RIF language features, not give the complete semantics (even less syntax) of dialects..
csma: duplication might be avoided (but doesn't quite know how to achieve)
Harold: should list discriminators and help the outsider to navigate RIF waters
<allen> The title uses "The RIF", Isn't that a violation of our convention?
Dave: arch - a central repository of concepts where dialects will point to for definitions
Sandro: will this make the core a 1p doc?
csma/sandro: discussion of what is "reusable"
<Harold> 'Reusers' of a base dialect such as the Condition language who layer other dialects on top could propose refinements/extension to the base dialect that they think could also help other layerings on top of the base dialect.
Sandro: should every little thing go to the arch doc? Eg, some piece from a fuzzy rule dialect?
DaveR: no, only the main parts are of interest in the arch doc.
<allen> is the section titled "text cases" supposed to be "test cases"?
<Harold> If indeed the propose refinements/extensions are helpful synergetically across 'side by side' layers, they should move into the base dialects.
<sandro> Sandro: but components shuold be able to become entries in the library (or multiple libraries) without changing them.
<sandro> Can you hear csma?
DaveR/CSMA: semantics of some concepts (eg, AND/OR) could be defined in the arch document. Other things might be harder to make reusable (eg, negation as failure)
MK: this might make the docs unreadable (full of pointers)
DaveR: instead of pointers we could use cut/paste to include the relevant pieces into dialect descriptions
<Harold> Dave, the initial Condition language was internally layered in a way similar to what you just mentioned, namely from A1 to A.5 of the Extensible Design: http://www.w3.org/2005/rules/wg/wiki/Extensible_Design
<Harold> Michael, I agree to what you say: We should not set the bar too high, but talk about principles, eg. on how to extend FOL with NAF.
<sandro> Sandro: I'm unclear about how XML namespaces should line up with the dialects/components, etc. rif: or rifcore+rifpr, etc....?
<sandro> ACTION: Harold to experiment with modularization of Condition library [recorded in http://www.w3.org/2007/02/28-rif-minutes.html#action01]
<rifbot> Created ACTION-253 - Experiment with modularization of Condition library [on Harold Boley - due 2007-03-07].
MK: we are possibly trying to set the bar too high for the arch doc. This doc should contain the general principles that go into the design of RIF (eg, how to go from FOL semantics to nonmon semantics of LP style rule languages.
<Hassan> +1 to Dave: the Data model and the Rule model are orthogonal...
csma: Issue: what are the deployment
models? What does it mean to implement rif? Test cases?
... extensibility: given a dialect, how to extend it?
csms: backward/forward compatibility issues
<allen> bye everyone - I need to attend another meeting. Hope you all enjoyed RIFing at MITRE!
csma: what about lub(two dialects)
<Harold> lub = least upper bound
csms: architectural principles behind the syntax. How to extend the semantics of a dialect?
Jos: where are OWL and RDF in the arch picture?
<Hassan> what was that?
<LeoraMorgenstern> That was clapping for Allen and Deborah. Thank you from us on the phone too!
csma: should we not emphasize owl but interoperation with other KB formalisms (eg, common logic)?
Harold: ontologies play a special role
(ie, OWL/RDF are more important.
... what about rule/ontology compatibility in the arch doc?
<sandro> ACTION: mkifer to start semantics section of Arch [recorded in http://www.w3.org/2007/02/28-rif-minutes.html#action02]
<rifbot> Created ACTION-254 - Start semantics section of Arch [on Michael Kifer - due 2007-03-07].
<sandro> ACTION: sandro to review extensibility section of Arch [recorded in http://www.w3.org/2007/02/28-rif-minutes.html#action03]
<rifbot> Created ACTION-255 - Review extensibility section of Arch [on Sandro Hawke - due 2007-03-07].
<sandro> ACTION: csma to revise Conformance section of Arch [recorded in http://www.w3.org/2007/02/28-rif-minutes.html#action04]
<rifbot> Sorry, couldn't find user - csma
<sandro> ACTION: christian to revise Conformance section of Arch [recorded in http://www.w3.org/2007/02/28-rif-minutes.html#action05]
<rifbot> Created ACTION-256 - Revise Conformance section of Arch [on Christian de Sainte Marie - due 2007-03-07].
<sandro> ACTION: christian to revise Conformance section of Arch [recorded in http://www.w3.org/2007/02/28-rif-minutes.html#action06]
<rifbot> Created ACTION-257 - Revise Conformance section of Arch [on Christian de Sainte Marie - due 2007-03-07].
csma: deployment of rules concept: party
A sends rule set R to party B. What data sets should be used by the
different copies of R.
... any takers to think about rule deployment?
DaveR: test cases' purpose is to document
the various decisions and to differentiate between rule sets that are
compliant with those decisions and those that do not
... example of a test case for round-tripping: should be able to test if one's application conforms to the round-tripping decision.
<sandro> ACTION: DaveReynolds to put in a short entry under Test Cases, outlining the issues [recorded in http://www.w3.org/2007/02/28-rif-minutes.html#action07]
<rifbot> Sorry, couldn't find user - DaveReynolds
<sandro> ACTION: Dave to put in a short entry under Test Cases, outlining the issues [recorded in http://www.w3.org/2007/02/28-rif-minutes.html#action08]
<rifbot> Created ACTION-258 - Put in a short entry under Test Cases, outlining the issues [on Dave Reynolds - due 2007-03-07].
<sandro> ACTION: Jos to start OWL/RDF compatibility section of Arch document [recorded in http://www.w3.org/2007/02/28-rif-minutes.html#action09]
<rifbot> Sorry, amibiguous username (more than one match) - Jos
<rifbot> Try using a different identifier, such as family name or username (eg. jdebruij, jderoo)
csma: missing section: how is RIF positions wrt OWL and other players in the semweb domain.
<sandro> ACTION: jdebruij to start OWL/RDF compatibility section of Arch document [recorded in http://www.w3.org/2007/02/28-rif-minutes.html#action10]
<rifbot> Created ACTION-259 - Start OWL/RDF compatibility section of Arch document [on Jos de Bruijn - due 2007-03-07].
<sandro> ACTION: Dave to put in a short entry under Data Sets, outlining the issues [recorded in http://www.w3.org/2007/02/28-rif-minutes.html#action11]
<rifbot> Created ACTION-260 - Put in a short entry under Data Sets, outlining the issues [on Dave Reynolds - due 2007-03-07].
<end of discussion of the arch document>
<Hassan> I need to leave - so I'll skip the demos (hopefully, Sandro will write more about his stuff). Thanks for putting up with my whining about mikes and such... Have a safe trip back. -hak
<sandro> Thanks, Hassan!
<Deborah> ScribeNick: Deborah
<sandro> Sandro: can we solve the dichotomy between p(s,o) and rdf(s,p,o) by using the HiLog style.
Harold: The HiLog style allows one to query over predicates.
josb: Do we want to allow/require querying over predicates in the Core?
ChrisW: Would you write rules to express the semantics of rdf and want them to be expressible in RIF?
josb: You may not express the rules, but any reasoner that uses rdf rules would have to take the semantics of rdf into account.
DaveReynolds: We do have a use case for sharing rules that process rdf.
<Harold> HiLog Example: "Show all relationships between John and Mary" -- RPS: ?R(John Mary) -- XML: <Uniterm> <op><Var>R</Var></op> <arg><Const>John</Const></arg> <arg><Const>Mary</Const></arg> </Uniterm>.
DaveReynolds: May need to process over
the predicate in order to translate/operate over the rules.
... You may be using rules to do data mapping and want to carry over all properties' values.
josb: Has a different use case, where you want to make use of the RDF rules when integrating data.
ChrisW: There are two different solutions
to josb's case: embedded translation, or translating your RDF into
... It's easy to imagine doing that for a small set of rules like RDF but harder for OWL.
<sandro> "translation" (like hoolet) vs "embedding" (like surnia)
DaveReynolds: But people are writing rules for the processing of RDF data that are in effect an extension of RDF.
josb: Must also think about OWL compatibility, and the designers chose to use RDF(S).
<sandro> Chris: I am confident that some day we'll have a dialect that lets you qualitify over predicates
josb: May need different strategies for the different flavors of OWL.
mdean: In applications, the flavors are not treated as separately as you may think.
ChrisW: There is an actual incompatibility between OWL-DL and Full wrt the definition of ObjectProperty.
<Harold> Another reason for RIF to look more carefully into OWL soon is that the OWL community is moving on, too -- OWL 1.1: http://owl1_1.cs.manchester.ac.uk/
ChrisW: In the two strategies, embedding vs. translation, embeding has some advantage for round-tripping but disadvantage for size of what is exchanged.
Harold: We might state what we consider
advantages/disadvantages of embedding vs. translation.
... See also OWL 1.1 spec.
ChrisW: Not clear we have a prior position on including quantification over predicates in the Core.
MichaelKifer: The problem with including quantification over predicates is that there are languages that don't support it.
<sandro> Sandro: I'd sure love hilog in the core from an RDF perspective --- because __triple(s,p,o) is an ugly RDF ghetto, and p(s,o) doesn't support embedding.
josb: Could put constructs for RDF triples in the rules (besides the literals) so we could cover RDF without representing it as a binary predicate.
MichaelKifer: Since we're going to introduce slots, could use that.
ChrisW: Need to define 'slots'.
<sandro> "named arguments to predicates"
MichaelKifer: "Named arguments to predicates"
josb: That would be exactly equivalent to RDF triples.
josb writes on white board to illustrate: s p o 'equivalent to' s[p ->>o]
<sandro> Chris: so you're not talking about named arguments to predicates, you're talking about structured objects -- objects that have attributes or slots with values.
ChrisW: By 'slots' we are NOT talking
about keyword ('named) arguments to predicates. Rather, we are talking
about structured objects that can bind to rules.
... So if an object has something true of it, and if p is a binary predicate, then how do you write a rule about that predicate?
josb writes on board to show how, in that approach, you would write a rule that quantifies over the predicate:
<Harold> If we write _triple(s,p,o) as _triple(p,s,o) we can easily generalize it to the N-ary predicate '_ntuple' as in _ntuple(p,a1,a2,a3,...,aN). The '_ntuple' dummy predicate has been called 'apply' in Warren's classical paper "Higher-Order Extensions of Prolog -- Are they Needed?". apply(p,a1,a2,a3,...,aN) can be used to reduce the semantics of the syntactically higher-order sugared p(a1,a2,a3,...,aN) to first-order.
ChrisW: Does this create a different annotation for binary predicates than for other predicates?
josb: No, 'p' there (above) is not a predicate; it's an object.
MichaelKifer: It has an interpretation as a predicate but also as an object.
josb: In RDF, 'p' is an object, not a predicate.
MichaelKifer: Slotted syntax is disjoint from predicate syntax, but they can be used together in rules.
josb: May depend on restrictions of the dialect.
We return to the question of whether RIF will have support for RDF in the Core.
josb: s[p->>o] is equivalent to _triple(s,p,o).
sandro: What would we call it in the abstract syntax?
josb: It's called a 'molecule'.
Harold: It's called a 'molecule' in F-Logic.
csma: We don't need to import all of F-Logic into the Core.
Confirmed by MichaelKifer.
MichaelKifer: But he had raised an issue about the typing of the components of these molecules.
josb (illustrating in writing): S[ Af.type ->>P ]. (missing symbols) would be needed, also.
ChrisW: With the exception of quantification over predicates, there is no problem with handling RDF.
MichaelKifer: With the [molecular] approach, the quantification problem goes away.
josb: And with signatures for typing.
DaveReynolds: You don't want an [RDF] implemeter to have had to have written RDFS in order to make use of RIF.
<sandro> Sandro: so would RIF Core include RDFS entailment?
DaveReynolds: Want RIF to handle rules that work over RDF data without having the RDF semantics with their data.
Next topic: RIF Support for OWL
Harold: Of the OWL features Jos mentions as not included in RIF Core (include classical negation, disjunction, and existential quantification) actually disjunction and existential quantification are partially present in the condition language and the body of rules.
josb: For OWL Full, which is a direct extension of RDF, you can make use of the approach already described.
csma: See Hassan's paper, section on OWL-DL.
josb: Might be tricky to combine the [molecular] approach with OWL DL.
because OWL DL puts restrictions on the kinds of triples you can write. Also, the unary and binary predicates are reallly predicates and not objects, in OWL DL.
josb: For OWL DL, you would use a RIF dialect where you had disjointness bewteen predicates of different arity; then have different approaches for each.
Harold: How does this fit with multisorted approach.
josb: The molecule approach seems to give us a good approach to OWL.
There is a problem with having RDF data sets combined together with OWL DL.
We don't want to break the interoperability by having to treat binary predicates and then rdf properties differently.
<Harold> s/Harold: Features of OWL not included in RIF Core include classical negation, disjunction, and existential quantification/Harold: Of the OWL features Jos mentions as not included in RIF Core (include classical negation, disjunction, and existential quantification) actually disjunction and existential quantification are partially present in the condition language and the body of rules/
This is scribe.perl Revision: 1.128 of Date: 2007/02/23 21:38:13[End of scribe.perl diagnostic output]
Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/ Guessing input format: RRSAgent_Text_Format (score 1.00) Succeeded: s/If the Core can't express something, then why it should be modeled in RIFRAF?/Base ontology should be somehow more expressive than the Core./ Succeeded: s/supposed to/supposed to apply to/ Succeeded: s/no, only some/well, it would be sufficient if only some/ Succeeded: s/their complete semantics (even less their syntax)/the complete semantics (even less syntax) of dialects./ Succeeded: s/N-ary predicate versuib/N-ary predicate '_ntuple' as in/ Succeeded: s/Harold: Features of OWL not included in RIF Core include classical negation, disjunction, and existential quantification/Harold: Of the OWL features Jos mentions as not included in RIF Core (include classical negation, disjunction, and existential quantification) actually disjunction and existential quantification are partially present in the condition language and the body of rules/ FAILED: s/Harold: Features of OWL not included in RIF Core include classical negation, disjunction, and existential quantification/Harold: Of the OWL features Jos mentions as not included in RIF Core (include classical negation, disjunction, and existential quantification) actually disjunction and existential quantification are partially present in the condition language and the body of rules/ Found ScribeNick: Harold Found ScribeNick: MichaelKifer Found ScribeNick: Deborah Inferring Scribes: Harold, MichaelKifer, Deborah Scribes: Harold, MichaelKifer, Deborah ScribeNicks: Harold, MichaelKifer, Deborah Default Present: meeting_room, Hassan_Ait-Kaci, Leora_Morgenstern, Axel_Polleres Present: meeting_room Hassan_Ait-Kaci Leora_Morgenstern Axel_Polleres Got date from IRC log name: 28 Feb 2007 Guessing minutes URL: http://www.w3.org/2007/02/28-rif-minutes.html People with action items: christian csma dave davereynolds harold jdebruij jos mkifer sandro