Re: AW: Problem with RIF-Core specification [was Re: Urgent: Issue with RIF-Core EBNF Grammar?]

In case we'd be still allowed to make changes on non-normative parts (EBNF) 
and add clarifications to normative text, there is no problem to go either way.

I agree with Jos that the current normative part (Section 2.3 of Core)
is ambiguous and allows (at least) these following readings for the EBNF:

a) No equality or class membership in either rule conclusions nor facts:

CLAUSE         ::= Implies | ATOMIC

b) No equality or class membership in rule conclusions but in (including universal) facts: 

CLAUSE         ::= Implies | ATOMICFACT
ATOMICFACT     ::= IRIMETA? (Atom | Frame | Member | Equal)




What I had proposed in my mail was my understanding of the intention of core, i.e. 

c) to allow Class membership only 
   for *ground* facts and equality neither in facts nor in conclusions:

CLAUSE         ::= Implies | ATOMICFACT
ATOMICFACT     ::= IRIMETA? (Atom | Frame | GROUNDTERM '#' GROUNDTERM )

d) Another alternative would be to allow also universal class membership:

CLAUSE         ::= Implies | ATOMICFACT
ATOMICFACT     ::= IRIMETA? (Atom | Frame | Member )


Probably c) and d) are out of the game at this point, but a) or b) should indeed be clarified in the text in 2.3.

Axel


On 12 May 2010, at 09:44, Adrian Paschke wrote:

> Hi,
> 
>  
> 
> We discussed that in RIF Core class membership cannot occur in rule conclusions. Since facts can be seen as rules without premises, we probably did not include class membership in ATOMIC terms in the CORE EBNF.
> 
>  
> 
> However, I see the benefit of explicitly adding class membership facts in Core in addition to the not materialized facts which you can import from external data sources.
> 
> If we allow class membership facts to be represented in a Core document, we should clearly distinguish rules from facts in the Core spec. as Jos proposes it.
> 
>  
> 
> But this would mean a change of the Core document which we should avoid at this point in time.
> 
> So, probably we have to live with it – only imports of external class membership facts are possible in Core.
> 
>  
> 
> Best, Adrian
> 
>  
> 
> Von: public-rif-wg-request@w3.org [mailto:public-rif-wg-request@w3.org] Im Auftrag von Jos de Bruijn
> Gesendet: Mittwoch, 12. Mai 2010 10:32
> An: Axel Polleres
> Cc: RIF
> Betreff: Problem with RIF-Core specification [was Re: Urgent: Issue with RIF-Core EBNF Grammar?]
> 
>  
> 
> Axel, all,
> 
> I think there are a number of basic problems in the specification of RIF Core formulas. In particular:
> 
> 1- the notion of "rule conclusion" is never defined. In fact, neither the notion "rule" nor "conclusion" is defined anywhere. This leads to several ambiguities: e.g., is a variable-free rule implication a rule? perhaps. Is a variable-free atomic formula a rule? there is no wording in BLD that would suggest this.
> 
> 2- if we were to assume that "rule" means "RIF-BLD rule", which is the assumption I would naturally make from the BLD spec, then I read the restriction
> "Equality terms and class membership terms cannot occur in rule conclusions -- they are allowed only in rule premises."
> 
> in [1] as saying that equality terms and class membership terms are not allowing the the conclusions of RIF-BLD rules. Full-stop.
> This means they are allowed in variable-free rule implications, universal facts (although some text in BLD may suggest these are a kind of RIF-BLD rules), and variable-free atomic formulas.
> I am quite sure we decided not to allow the assertion of equality. I do not recall exactly what we decided about facts concerning class membership (i.e., a#b). Does anybody recall what we decided here?
> 
> In any case, this ambiguity needs to be resolved. Notice that the EBNF grammar does not help us here, since it is non-normative.
> 
> 
> Best, Jos
> 
> [1] http://www.w3.org/TR/2010/PR-rif-core-20100511/#Formulas_of_RIF-Core
> 
> On Tue, May 11, 2010 at 6:15 PM, Axel Polleres <axel.polleres@deri.org> wrote:
> 
> Hi folks,
> 
> We are working on a parser with some students and I am afraid my student found something awkward in the RIF Core grammar, see mail below.
> 
> Indeed, I think he poked into a quite weird issue:
> It doesn't make sense to allow class membership terms in rule bodies, if they can't appear at all in *any* facts.
> The current grammar and the restrictions in Section 2.3 though only allows uniterms and frames as facts.
> 
> 
> To repair this
> 
> 1) We'd need to change in Section 2.3 Formulas of RIF-Core:
> 
>  * Equality terms and class membership terms cannot occur in rule conclusions -- they are allowed only in rule premises.
> -->
>  * Equality terms cannot occur in rule conclusions -- they are allowed only in rule premises.
>  * Class membership terms can only occur in rule premises or as ground facts.
> 
> 2) a proposal to fix the grammar in Section 2.6 would  be:
> 
> In the Rule Language grammar:
> 
>  CLAUSE         ::= Implies | ATOMIC
>  -->
>  CLAUSE         ::= Implies | ATOMIC | GROUNDTERM '#' GROUNDTERM
> 
> 
> 
> sorry for spotting this now only, but I am afraid this is severe.
> the fix is not very problematic, though.
> 
> Axel
> 
> 
> 
> 
> Begin forwarded message:
> 
> > From: "Obermeier, Philipp" <philipp.obermeier@deri.org>
> > Date: 11 May 2010 16:26:50 GMT+01:00
> > To: "Polleres, Axel" <axel.polleres@deri.org>
> > Cc: "Marco Marano" <marcomarano83@gmail.com>
> > Subject: RIF-Core: EBNF, equality/memberhip facts
> >
> > Hi Axel,
> >
> > I found a minor error in the EBNF grammar [1] for RIF-Core (Altough,
> > this grammar is informative due to the lack of well-formedness checks,
> > it is also defined as strict superset of RIF-Core.).  Within this
> > grammar you cannot derive Equality or Membership (ground) facts since
> > the ATOMIC rule's rhs is restricted to atomic formulas excluding
> > Equality/Membership formulas. Apparently, this restriction is well
> > justified since ATOMIC may appear in rule heads (cf. IMPLIES rule's
> > rhs), for which Core forbids Equality and Membership formulas. In
> > conclusion, an introduction of a new ATOMIC_FACTS grammar rule extending
> > ATOMIC by Membership/Equality  would solve this issue w/o breaking the
> > restriction for atoms in rule heads.
> >
> > Best
> > Philipp
> >
> > [1]
> > http://www.w3.org/2005/rules/wiki/Core#EBNF_Grammar_for_the_Presentation_Syntax_of_RIF-Core
> >
> > --
> > Philipp Obermeier
> > Digital Enterprise Research Institute, National University of Ireland, Galway
> > email: philipp.obermeier@deri.org
> 
> 
> 
> 
> -- 
> Jos de Bruijn
>  Web:          http://www.debruijn.net/
>  LinkedIn:     http://at.linkedin.com/in/josdebruijn
>

Received on Wednesday, 12 May 2010 09:06:14 UTC