- From: Birte Glimm <birte.glimm@comlab.ox.ac.uk>
- Date: Tue, 8 Feb 2011 20:07:56 +0000
- To: Matthew Perry <matthew.perry@oracle.com>
- Cc: W3C SPARQL Working Group <public-rdf-dawg@w3.org>, "Zhe (Alan) Wu" <Alan.Wu@oracle.com>
On 8 February 2011 16:33, Matthew Perry <matthew.perry@oracle.com> wrote: > Hi, > > Here are some comments from Zhe Wu. Thanks. I comment inline below. See http://www.w3.org/2009/sparql/docs/entailment/xmlspec.xml for the updated version. Birte > - Matt > > ------------------------------------------------------------------------------- > Some comments I have so far for the entailment regimes doc. > > Overall it is very well written. > > - fn: is not referenced at all. We can get rid of it. Good point, removed. > - in Section 1.3, ".... and is E-equivalent to AG" We probably want to > define E as an entailment regime. > Right now, it seems to come from nowhere. Yes, that came out of context there. I added the sentence: The SPARQL Query conditions for using a logical entailment relation E, such as RDFS-entailment, instead of sub-graph matching for the case of a consistent active graph are repeated below for clarity. An overview of how the different entailment regimes satisfy these conditions follows. ... list of conditions... > - in the paragraph before 2.3, sentence "Under the modified condition not > even rdf:_1 would be a solution and even the BGP ex:a ex:b ?x . would have > an empty solution" is a bit tricky to read (with the dot in the middle of > the sentence). > ==> > "Under the modified condition, rdf:_1 would not be a solution. Further, BGP > ex:a ex:b ?x would have no solution at all." done > - In the first table in Section 3 > ... sk(P(BGP)) are ground and RDF entailed by sk(SG) > ==> > ... sk(P(BGP)) are ground and RDFS entailed by sk(SG) Hm, I think I had corrected that, but apparently not... Done now. > - I am a bit confused about if a SPARQL query should return a blank node > (skolemized). > In 2.1, the document says "Clearly, the Skolemized blank nodes should not > occur in query results." > However, in the SELECT ... COUNT(?author) example in 3.2, we are actually > counting them. I tried to rephrase that part in 2.1 and added also another example. What that paragraph is meant to say is that the bnodes are to be returned in the results and not the Skolem constants for the bnodes. I.e., G: ex:s ex:p _:o BGP: ex:s ex:p ?o yields one solution, which is a blank node (either _:o or if the scoping graph renamed _:o we could get another bnode), but never sk:o (or whatever the Skolem function turns _:o into). For implementation purposes that means that bnodes can mostly just be treated as constants as in subgraph matching. > - The canonicalization related discussion in 4.1 is a bit odd. Basically > different vendors can return different answers. What is the problem of > enforcing a canonicalization? > Otherwise, we will get different number of results, different counting, > .... Not too good for interoperability. I am all for that, but this is a problem already relevant for SPARQL in general as some systems load ex:s ex:p "1.00"^^xsd:decimal as ex:s ex:p "1.0"^^xsd:decimal and at the last F2F it was decided that canonicalization should not be enforced. Furthermore, I think the text is not actually correct in saying that the two triples ex:s ex:p "100.0"ˆˆxsd:decimal . ex:s ex:p "100.00"ˆˆxsd:decimal . would be parsed into one triple if canonicalisation is performed. I think we would rather get two identical triples, which, however, give just one answer as opposed to two answers if no canonicalization is performed. I see choices: a) remove the D-Entailment regime entirely b) enforce canonicalization for systems that want to do D-entailment c) leave it as it is a) is my favourite since D-Entailment is anyway quite meaningless unless a datatype map is fixed. Otherwise we have one D-Entailment system that just handles xsd:integers, another one that does only xsd:string, others that do all of xsd, etc b) is my next favourite but both a) and b) will need WG consent as I understand it, which wasn't there last time we discussed it. > I have gone through the RL part as well. Seems fine. > > I haven't looked carefully into OWL 2 RDF-based semantics entailment. > > -- Dr. Birte Glimm, Room 309 Computing Laboratory Parks Road Oxford OX1 3QD United Kingdom +44 (0)1865 283520
Received on Tuesday, 8 February 2011 20:08:29 UTC