Re: Review of RIF Core entailment in SPARQL draft

Jos, I think it makes sense to keep this more conversational, and have
you directly reply with this, rather than us discuss it much internally
first.   If you agree, please send this to
public-rdf-dawg-comments@w3.org

     -- Sandro

On Mon, 2010-07-26 at 14:39 +0200, Jos de Bruijn wrote:
> I reviewed the RIF Core entailment section of the current SPARQL 1.1
> Entailment Regimes draft [1]. Find my comments below.
> 
> Major comment (section 7.2):
> I did not understand the definition of Query Answers. It is unclear
> what the combination is for which RIF-Simple entailment is checked. In
> particular, which are the RDF graphs that are considered? Is this just
> the scoping graph or also the graphs imported by the rule set (the
> text in the "Legal Graphs" row seems to sanction such imports).
> 
> Errors:
> - Sec 7, 2nd par, 1st sentence: the phrase "...satisfiability and
> entailment, each of which define how a common RIF-RDF interpretation
> is a model of a combination" is incorrect. Satisfiability defines when
> (not how) an interpretation is a model; entailment does not define
> models.
> - Sec 7, 1st par::
>    * It is claimed here that entailment is checked wrt. a
> Skolemization of the scoping graph. This contradicts the definition of
> query answers in Sec. 7.2.
>    * It is claimed that only combinations with one RDF graph are
> considered. This seems to contradict the definition of Legal Graphs in
> Sec. 7.2.
> - Sec 7.2, def query answers: rif:imports is not a predicate, it is a CURIE
> - Sec 7.2, "RIF-Core conformance is defined with respect to safe rules
> and so this entailment regime is defined with respect to combinations
> formed from safe RIF documents.": Not true. This entailment regime is
> defined with respect to safe RIF document because it is defined as
> such, in the row "Legal Graphs" of the table.
> - Sec 7.4, 2nd par: cyclic references *do not* rule out the existence
> of a unique minimal model. I believe that every RIF Core document has
> a single unique minimal Herbrand model. Strong safety can, in
> addition, guarantee finiteness (when not considering the datatypes).
> - Sec 7.4, 3rd par: RIF models are necessarily infinite, since the
> domain includes includes all integers.
> 
> Further comments:
> - Sec 7.2, 1st editorial note: I wanted to see if I can answer the
> question, but could not find any definition of "answer set" in the
> draft
> - Sec 7.3: I do not understand the relation between the text in this
> section and RIF Core entailment in SPARQL. I see implementation hints
> about RDFS and OWL 2 RL; it seems to me these hints should be included
> in the respective RDFS and OWL 2 RL sections.
> - Sec 7.4, 2nd sentence: I cannot parse '(with possibly infinite)'.
> - Sec 7.4, 2nd par: it is not clear why stratification is discussed
> here, since RIF Core has not negation. In general, I find the
> paragraph a bit confusing. I think it should be rewritten.
> 
> Discussion points:
> - I would suggest to remove the rif:imports triple from the scoping
> graph, since it may lead to unexpected inferences.
> 
> 
> [1] http://www.w3.org/TR/sparql11-entailment/#id35811294
> 
>

Received on Tuesday, 27 July 2010 15:20:22 UTC