Re: Review XML Data in RIF

On Tue, 19 Oct 2010 19:21:02 +0200
Christian De Sainte Marie <csma@fr.ibm.com> wrote:

> Hi Michael,
> 
> Michael Kifer <kifer@cs.stonybrook.edu> wrote on 19/10/2010 06:03:54:
> > 
> > I was tied up, so I took a rather superficial look at the new document. 
> It is
> > in a much better shape than before. In particular, I am happy that the
> > infomodel stuff is gone.
> 
> Great. Although the infomodel is not gone, of course. Only, now, you have 
> to read XDM by yourself :-)

But it is not there "in your face" now, and it does not have to be read on first
reading.

> 
> Michael Kifer <kifer@cs.stonybrook.edu> wrote on 19/10/2010 06:18:55:
> > 
> > Sorry, didn't have a chance to look at it before, but the whole purpose 
> of
> > defining the models is so that then one could define entailment. Without 
> the
> > entailment there is not much use for the notion of a model.
> > 
> > Regarding the notion of entailment, I think it is not quite correct 
> > (Definition
> > Logical entailment in a RIF BLD+XML data combination). We should be 
> > talking about one BLD+XML combo <phi,E,S> entailing another, <psi,E',S>.
> > Perhaps actually just entailing psi. Just phi entailing psi seems 
> incorrect.
> 
> Without earlier response to my email about entailment, here is what I had 
> put in the spec. I think that it goes in the direction you say:
> 
> Definition (Logical entailment in a RIF BLD+XML data combination). Let φ 
> and ψ be (document or non-document) formulas. We say that φ entails ψ in a 
> RIF BLD+XML data combination <φ, E, S> if and only if every model of the 
> RIF BLD+XML data combination <φ, E, S> is also a model of the RIF BLD+XML 
> data combination <ψ, E, S>.
> 
> I just rewrote it, as follows, to make the intent clearer (and the 
> definition less ambiguous):
> 
> Definition (Logical entailment of a RIF BLD+XML data combination). Let φ 
> and ψ be (document or non-document) formulas. We say that a RIF BLD+XML 
> data combination <φ, E, S> entails the RIF BLD+XML data combination <ψ, E, S>
> if and only if every model of the RIF BLD+XML data combination <φ, E, S>
> is also a model of the RIF BLD+XML data combination <ψ, E, S>.

I think it is right. One question is: does E (and S) have to be the same in
both cases?

 
> > My understanding is that the XML part, E, is just another form of data, 
> ie, a
> > bunch of facts.
> 
> My problem with the above definition is, indeed, that entailment is 
> defined wrt a specific set of data only, which might be too restrictive to 
> be useful.
> 
> I mean, combination with XML data adds semantics from XPath, and 
> combination with an XML schema adds semantics as well...
> 
> So, really, we could have entailment wrt a specific combination of data E 
> and schema S, entailment wrt any combinations including a specific schema 
> S, and entailment in a combination with XML data, in general.
> 
> I mean, if expr is an XPath expression, then ?x["fn:data(expr)"->?z] :- 
> ?x["expr"->?y] entails ?y["fn:data(.)"->?z] :- ?x["expr"->?y] in the 
> context of any RIF BLD+XML data combination, due to the semantics of XPath 
> expressions, but not in RIF BLD in general (which, btw, proves that the 
> semantics of a combination of RIF with empty XML data and schema is not 
> equivalent to the semantics of pure RIF: the simple fact of stating that a 
> RIF doc is interpreted as in a combination with XML data, even in the 
> absence of data and associated schema, changes its semantics. Notice, 
> however, that a RIF doc that imports no XML data nor XML schema is _not_ 
> the same as a RIF combination with empty XML data and schema).

This is an interesting example. Would be good to have it in the doc along with
the above remark.

It is also useful to say early on that RIF+XML gives a concrete semantics to
some attributes (like fn:data), which makes this semantics different from RIF
alone.

Anyway, I don't think that this example is enough to justify the notion of
entailment as in your original definition. Even for that example it is not
clear that the original notion of entailment is what one wants.

> > So, we should be definitely talking about <phi,E,S> entailing something.
> > Now, the question is what exactly do we want to entail? I think psi 
> > should be a
> > query, ie, a BLD body formula. Do we want to entail a piece of XML as 
> well?
> > I am not sure. We need to see a use case of what exactly the intent is.
> 
> See the example, above, where psi is a rule...
> 
> > So, I take the approval in my previous message back. I think we needto 
> clarify
> > this point.
> 
> Since we do not have the WG to discuss it, either it is easy to clarify, 
> and we can do that quickly among you, Adrian, Gary and myself; of we 
> publish with an editor's note.

I am ok with the revised "conservative" notion of entailment + the explanations
per the above.


> In the latter case, I propose to include the most general definition (the 
> one that accounts for the example, above), which seems more useful than 
> the most restrictive one (the one currently in the document):
> 
> Definition (Logical entailment in a RIF BLD+XML data combination). Let φ 
> and ψ be (document or non-document) formulas. We say that φ entails ψ in a 
> RIF BLD+XML data combination if and only if every model (E, ?, S) of any 
> RIF BLD+XML data combination <φ, E, S> is also a model of the RIF BLD+XML 
> data combination <ψ, E, S>.
> 
> And an editor's note to the effect that other definitions are possible, 
> and that this need be clarified before moving to last call.

I think the doc should include the revised definition (the conservative one that
is currently there as of today). We can add an editor's note with the earlier
definition and state that it is unclear whether that other notion is also
needed.

michael

Received on Tuesday, 19 October 2010 17:56:36 UTC