RIF Telecon 24 July 2007 -- 24 Jul 2007

Scribe remark (AxelPolleres): Some entries in these minutes have been changed/adapted from the original notes taken during the meeting or from the irclog after clarification with the respective participants. These are marked with ^*edited*.

Admin

<Christian> clear agenda

<ChrisW> Meeting: RIF Telecon 24 July 2007

<ChrisW> Chair: Christian de Sainte-Marie

<ChrisW> Agenda: http://lists.w3.org/Archives/Public/public-rif-wg/2007Jul/0128.html

<Christian> scribe: Adrian Giurca

<Christian> scribenick: agiurca

Scribe remark (AxelPolleres): I didn't find anything either in the autogenerated minutes nor in the irclog whether the minutes of last time were approved or not.

Liaison

<johnhall> SBVR - nothing to report

<Christian> no actions pending on Liaison

F2F7

<sandro> Manhattan 26-27 option open until Thursday.

<sandro> ACTION: Christopher to double check whether Manhattan Sept 26-27 option is available. [recorded in http://www.w3.org/2007/07/24-rif-minutes.html#action01]

<rifbot> Created ACTION-329 - Double check whether Manhattan Sept 26-27 option is available. [on Christopher Welty - due 2007-07-31].

<sandro> If not, we fallback to Hawthorne Sept 27-28.

<sandro> So we are definitely meeting in New York State, with IBM Hosting, either Sept 26-27 or Sept 27-28. Exact dates and location to be announced on Thursday.

<sandro> (confirmed by chair.)

Technical Design

Sandro ACTION-323 on "Make a pass at updating the extensibility page based on the discussions and strawmen" not concluded.

Christian Shall we switch to topic: Changes in Basic Logic dialect?

Harold: suggest to postpone to the end of today's telecon, could give a summary.
... since Michael not yet here.

Christian: please post regrets if you don't attend.

<DaveReynolds> http://www.w3.org/2005/rules/wg/wiki/Arch/Data_Sets

Datasets

Dave: Original discussion: What does RIF need to say about dataset.
... first issue: dataset identification
... reference to datesets (possibly remote).
... Second issue: How to describe access to that data, format, datamodel, etc. (using metadata)

Dave: out of scope of rif of how to turn dataset Id (URI) into actual dataset.

Hassan: this is precisely what KM3 had in mind to do.

<ChrisW> there is no RDFS data model

Hassan: see http://wiki.eclipse.org/index.php/KM3

Christian: Think we had agreement to not define our own schema/metamodel language.

Hassan: But this is what asn07 is doing.

<sandro> (I'm totally baffled by what Hassan is saying....)

Hassan: we should avoid adhoc languages, and reuse existing.

Jeff: Question for Dave: I didn't see RDF data model for RIF in the document. Isthere a reason for that?

<Hassan> Sandro: what is baffling in what I said?

<JeffP> ok

<sandro> I don't see how KM3 (or asn07) solves the problem Dave is talking about.

Dave: Not particularly, could be added.

<Hassan> It allows describing the meta-metamodel

<Hassan> This allows then to parameterize the data model

Christian: We had a preliminary list of datamodels at f2f6 containing XML Schema, RDFS, OWL, etc.
... we had agreement not to require one particular data model upfront, but to require one via metadata for data models.

<sandro> okay, I have a hint of an understanding now, Hassan.

<hak> having the metametamodel is used therefore to support things like navigating through data in some so-described data-model

<hak> ok

<sandro> Dave: Data-Set-Identification is separate from the Data Models discussion

Dave: 2 issues: a) a ruleset requires a particular data model and needs to declare that b) all data in RIF have to conform to a meta data model like KM3 or MOF, these are separate issues.

<sandro> I wonder whether Axel or I understood Dave correctly, because we are disagreeing about what he said.

Christian: Do we need to define the data model or do we only need to refer to it?
... Those are different problems.

Dave: We solved the problem already for RDF.
... Should we, for instance for XML Schema define it likewise, or define a more general way to refer to different data models?

<sandro> (data-set-identification == data-source-location)

Christian: question: Frames as defined in BLD at the moment (which could be used to carry RDF data) ...
... can values of slots be frames again?

Gary: We also need lists.

Christian: currently we have sets of values for slots.

<Harold> DaveR, would your distinction similarly apply to other (semweb) formats, specifically OWL 1.0? TBoxes corresponding to rulesets and -- external -- ABoxes corresponding to separate data sets? If yes, do you have any corresponding experience with such a distinction for OWL?

Christian: Is what we have now enough for navigating instance data?

Gary: 1) Do we have a data model? 2) If yes, how do we map to existing data models

<sandro> Christian: It's not that RIF doesn't define a new data model, it's that it doesn't define a new data modeling (schema) language. [ is that right? ]

Christian: Of course we have a data model, but we do not define any schema language. Our data model should be as simple as possible and encompass widely used data models, ie those should be easy to map.

Gary:^*edited* I think I was trying to make the following points:
- RIF has a data model (relations and frames)
- the original frame proposal with classification and typed attributes is pretty good for mapping many application data models
- we have significantly weakened frames by removing classification and types and this makes mapping harder

Dave: As soon as we have hierarchies in the data model, we have more of an issue.

Gary: Can we then write rules that talk about the class hierarchy?

Hassan:^*edited* agree with dave, unlikely that we can have something to make all those existing data models agree.
... often rules DO use the data model explicitly - e.g., for data type nagigation, aggregates, etc... How do we abstract from the data model using some classification scheme?

Christian: RIF has to know the datamodel in some way.

Hassan: Who's gonna do that? (mapping)

Christian: the receiver
... e.g. When you specify that your rules operate on the MISMO schema, the reciever has to know how to map its own format onto MISMO.

Hassan: Of course, but we should identify sets of languages which can be readily mapped.

Christian: We should have it for XML Schema, RDF, OWL, and that's it. But users need to know how to map these onto their own data model.

Hassan: Anyway, to express a data model, either we use something like KM3, which is systematic, or we do it adhoc.

<Zakim> sandro, you wanted to ask Dave how the data-set-identification problem is addressed for the RDF example.

Christian: We do not need it

sandro: I heard Dave saying that for RDF its kinda solved, but we do not have a solution for the dataset identification problem.

Dave: If we use IRIs for identification, we don't want to mandate http dereferencability.
... we should specify the minimum.
... minimum: We have an id and the posisibility to annotate it with metadata.

Scribe remark (AxelPolleres): (sandro, can you sum up your last argument yourself, pls)

<MichaelKifer> Christian, can u write the options in the chat?

<DaveReynolds> Christian: Would like opinion of group on options for data models ...

<Christian> option 1: All we need is a specification of how the RIF data model (frame) maps onto a limited number of standard format used for interchanging data model (e.g. XML-S)

<Christian> option 2: RIF should define its own meta-modelling language

<Hassan> I vote for option 2 whether KM3 or ASN06++

Christian: Option 2 includes option 1.
... if we go for option 2 we probably go way beyond what we want to achieve by RIF.

<Hassan^*edited*> Option 1 proposes to use the ad hoc data model that we have in RIF to express most other data models

Christian: hte point is not to be able to describe XML schema in RIF, but how to reference it.

<Harold> +1 to option 1 (RIF already has relations, equation-defined functions, frames, signatures, prim types)

sandro: can we move on in the agenda or give an action item to somebody to seek resolution of this?

<scribe> ACTION: Dave to update his dataset/datamodel page according to f2f6 discussion [recorded in http://www.w3.org/2007/07/24-rif-minutes.html#action02]

<rifbot> Created ACTION-330 - Update his dataset/datamodel page according to f2f6 discussion [on Dave Reynolds - due 2007-07-31].

<MichaelKifer> i am not sure i understand what either of these options really means

<scribe> ACTION: Christian to clarify the issue on mappings from existing data models vs defining a new data modelling language [recorded in http://www.w3.org/2007/07/24-rif-minutes.html#action03]

<rifbot> Sorry, couldn't find user - csma

<Harold> -1 to option 2 (we didnt want yet another datamodel)

Changes in Basic Logic dialect

<ChrisW> (the action was not created)

Michael: main thing to resolve sorts.
... first understnading signatures, second "real" sorts.
... I kept the signatures but kept the sorts.
... but only took out sorts for variables, but kept datatypes like RDF and XML literals.

<DaveReynolds> ACTION: Christian to clarify the issue on mappings from existing data models vs defining a new data modelling language [recorded in http://www.w3.org/2007/07/24-rif-minutes.html#action04]

Michael: the basic logic now does distinguish between predicate, function and constant symbols, more restrictive than before (also not allowing same symbol with different arities).

<rifbot> Created ACTION-331 - Clarify the issue on mappings from existing data models vs defining a new data modelling language [on Christian de Sainte Marie - due 2007-07-31].

Michael: close to classic first-order logic.

Christian: Wasn't aware that removing sorts would remove typed variables.

Hassan:^*edited* By saying that you now have separation for predicates, functions, ... doesn't that amount just to hiding the sorts in the names? For example, 2-ary predicate 'foo' and 3-ary function 'foo' can always be renamed apart using a hidden sorted-signature naming scheme such as, e.g., 'foo/2#predicate' and 'foo/3#function'.

Michael: That was a requirement for the basic dialect, it is not precluded for other dialects.

Christian: You can still say a variable belongs to a class?

Michael: The different is with typed variables you can do syntactic checks at compile time, which is in general undecidable. Now you can do dynamic checks for the values of variables only.
... With sorts you can only tie variables to a sort (for constants) whereas with classification terms you can tie to e.g. a class which is more general, but not chackable statically.

Christian: (Scribe remark (AxelPolleres): question about declaration of variables)

Michael: You (Christian) mean something like "for every X in class" ... This is bounded quantification, wich is syntactic sugar only.
... the question is whether you want it as part of the language.

Gary: What that really ammount to for me is strong typing where runtime checking is not really an option.

Michael: I think we can still accomodate strong type checking, by classification terms like isInteger(x). Translating back into a language with strong checking is still possible to do.
... to some extent, but can be improved with real sorted variables.
... functionalitywise it is not different.

Christian: What else than bounded quantification are Classification terms for?

Michael: It is necessary for conceptual modelling

Christian: don't we suppose conceptual modelling in OWL, UML, etc.

ChrisW: Why don't a subclass relation just corresponds to a rule?

Michael: it does

<DaveReynolds> +1 to ChrisW, the subclass relation is a problem

ChrisW: But as soon as we do that it looks like we do another conceptual modelling language

<GaryHallmark> +1 to Michael, need something to map to

<Harold^*edited*> > We have the signature names bool, i, f0, f1, p0, p1, etc. Of these, bool and i can be seen as the primitive types of boolean values and individual values. The Basic Logic Dialect or an extension dialect could allow refining i, the type of individual values, into our primitive types long, string, decimal, time, dateTime (and iri).

Michael: The exercise of Frames is exactly the same thing.

<Harold^*edited*> So, long subTypeOf decimal subTypeOf i, string subTypeOf i, time subTypeOf i, dateTime subTypeOf i (and iri subTypeOf i).

Axel: We need rules also to *express* datamodels. We have a use case for e.g. advancing OWL expressivity by rules

Dave:^*edited* Mapping RDF et al to isa/sub gives us some problems. First, having isa/sub separate from slots and this seems likely to cause problems for the RDF metamodelling where they are the same. Second, the semantics RDF would want for this and the semantics Java classes would want for these are different so you couldn't intermix the two.

Michael: If you translate RDFS, you should put it in a scope.

<scribe> ACTION: Dave to state his argument in an email. [recorded in http://www.w3.org/2007/07/24-rif-minutes.html#action06]

<rifbot> Created ACTION-332 - State his argument in an email. [on Dave Reynolds - due 2007-07-31].

Christian: time to adjourn

<Hassan> +1

<PaulaP> +1

<PaulaP> bye

- DRAFT -

RIF Telecon 24 July 2007

24 Jul 2007

Attendees

Contents