Re: Various comments on SPARQL 1.1 Entailment Regimes

Dear Antoine,

thank you for the very detailed review.

You suggested to put Subsections 2.1 and 3.2 into a common Section. We
have taken that advice an introduced a new (informative) section after
the RDF entailment regime that contains discussions that are relevant
to all regimes. In addition to the subsections you mentioned, the new
section also contains the former subsections 2.2 to 2.4, which equally
are not specific to the RDF regime.

You asked why D-entailment is not defined as in RDF Semantics or
RIF/OWL. The definition of D-entailment does not prescribe the support
of any datatypes, which means that without a minimally required
datatype map, the D-entailment regime does not guarantee anything
beyond the RDFS regime. Thus, the working group decided to specify a
datatype map that has to be supported, which is a subset of the
datatypes that RIF requires, but which omits less frequently used
types (e.g., xsd:hexBinary or xsd:NMTOKEN). Systems can always go the
extra mile and also support the less common dataypes. Apart from
xsd:date and xsd:time, the chosen datatype map is also a subset of the
OWL RL and, hence, the OWL datatype map in general. We believe,
however, that xsd:date and xsd:time are so popular that a lack of
support for them seems problematic.

The detailed comments have mostly been addressed in the proposed way.
The point that has not been addressed exactly as suggested is the
following:

- Sec.5.3: "reasoning problems in OWL under OWL 2 RDF-Based Semantics
  are undecidable ..." -> semi decidable, which means the computation
  of the query results always finishes in finite time.
The terminology has been changed to semidecidable, which is more
precise than undecidable, but that still does not mean that the
computation of the query results always finishes in finite time.

The reasoning problems are semidecidable, i.e., positive entailments
can be enumerated whereas for negative entailments (non-consequences
of the ontology) the reasoner might not come back with the answer in
finite time (semidecidable). If we were to find the answers to { ?x
rdfs:subClassOf ex:C } w.r.t. an OWL Ful ontology, then any binding
for ?x that instantiates the triple into a non-entailed subsumption
might cause the reasoner to not come back with an answer in finite
time, whereas for bindings that instantiate the triple into an
entailed subsumption, there is a (theoretical) guarantee that the
reasoner comes back after finitely many steps with the answer. Now
what one can do in practise if of course to give a time limit to the
reasoner and only the answers that can be found in the time limit are
returned. In that case we might miss some positive entailments, which
could not be determined in the given time limit. Hence, we possibly
get incomplete answers.

We hope that this answer adresses your concerns. The updated working
draft is available at:
http://www.w3.org/2009/sparql/docs/entailment/xmlspec.xml

We would be grateful if you would acknowledge that your comment has
been answered by sending a reply to this mailing list.

Birte (on behalf of the SPARQL WG)



On 15 June 2011 09:39, Antoine Zimmermann <antoine.zimmermann@deri.org> wrote:
> These are various comments (mostly editorial) about SPARQL 1.1 Entailment
> Regimes.  Overall, I find this document very good.
>
> Apart from the detailed comments below, my main concerns are the following:
>
>  A)  Section 2.1 and 3.2 should be put together in a separate section.  2.1
> does not belong to Section 2 "RDF entailment" and 3.2 does not belong to
> Section 3 "RDFS entailment".  They both are applicable to any entailment
> regime.
>
>  B)  D-entailment is not defined as in RDF Semantics.  Why?  The definition
> does not even correspond to an OWL 2 D-entailment, nor a RIF D-entailment.
>
>
> Detailed comments:
>  - At the end of the abstract, give a comprehensive list of the entailment
> regimes given in this document instead of "...such as RDF entailment, RDFS
> entailment, etc."
>  - Section "Changes":
>  * "forth" -> fourth
>  * "has been made explicit in for each regime" -> in OR for?
>  - Sec.1.1.3:
>  * use a different font or face (bold, italic, etc) for variables to
> differentiate them from normal words (especially "I", "a", which are English
> words too).
>  * "A triple pattern is member of the set" -> is a member
>  - Sec.1.3:
>  * certain well-formed of RDF graphs" -> well-formed RDF graphs
>  * "to be taken to to address" -> taken to address
>  - Sec.2:
>  * "forth" -> fourth
>  * condition (C2) is different from the condition given in Sec.2.2, where
> \mu(x) is replaced by sk(\mu(x)) and SG is replaced by sk(SG). Please
> harmonise.
>  - Sec.2.1:
>  * I don't see any reason why this section is in the RDF entailment regime
> section. It's clearly independent of the entailment regime. Even the
> examples inside are going beyond RDF entailment. This should be in a
> separate section
>  * there is the following example:
> ex:s ex:p "<a/>"^^rdf:XMLLiteral .
> ex:p rdfs:domain rdfs:Literal .
> I don't see why the second line was included.
>  * in the last example triples:
> _:lit rdf:type rdfs:literal .
> It should read "Literal" with a capital "L".
>  - In Sec.2.2: compare condition (C2) here and in the tables of the
> entailment regimes;
>  - Sec.3.1: "Here we can to derive the" -> we can derive?
>  - Sec.3.2: this has nothing to do in the section on RDFS regime. It is
> independent of the regime, like sec.2.1. Make a separate section, probably
> together with what's in Sec.2.1;
>  - Sec.4:
>  * "A datatype map for the D-entailment refime MUST contain at least all
> pairs of the form <prefix:dt,dt>, ..." -> this is different from the
> definition in RDF Semantics. RDF D-entailment does not require that XSDs are
> included in the datatype map.
>  * """The set Lit(SG) is the set of all literals "xx"^^dt such that
> "yy"^^pdt occurs in SG, xx is the canonical representation of yy, and pdt is
> the primitive datatype from which dt is derived""" -> dt and pdt should be
> swapped. The same error occurs in Sec.5 and Sec.6.
>  - Sec.4.1:
>  * "e.g., integer, byte, and short all based on decimal" -> ..., and short
> are all based ...
>  * "(from the first tripe)" -> from the first triple
>  - Sec.5.2: the editorial note is correct, the example is not well chosen.
> Use this example instead:
> Data: ex:x owl:sameAs "5"^^xsd:decimal .
> Query: SELECT ?l WHERE { ex:x owl:differentFrom ?l .}
>  - Sec.5.3: "reasoning problems in OWL under OWL 2 RDF-Based Semantics are
> undecidable ..." -> semi decidable, which means the computation of the query
> results always finishes in finite time.
>  - Sec.6.5.1: "OWL 2 DL can also handle ... EL and QL profiles ..." -> and
> RL profile too.
>  - Sec.6.5.4: "OWL 2 RL defines a syntactic subset of OWL 2, which ..." ->
> subset of OWL 2 DL, which ...
>  - Sec.7.1: Illegal Handling does not specify anything about handling
> illegal graphs that are syntactically correct.
>  - Sec.7.3: the sentence "The ex:hasHospitalization ... with the events.",
> though grammatically correct, is very convoluted and hardly readable. Please
> rephrase it.
>  - Sec.8: "For an example, we consider a data set with consists of an empty
> ..." -> For example, we consider a data set which consists ...
>  - References:
>  * the W3C ref should have a link to the current version, not the latest
> which may be different from what is used by the SPARQL spec.
>  * why is "OWL 2 Profiles" not normative?
> --
> Antoine Zimmermann
> Researcher at:
> Laboratoire d'InfoRmatique en Image et Systèmes d'information
> Database Group
> 7 Avenue Jean Capelle
> 69621 Villeurbanne Cedex
> France
> Tel: +33(0)4 72 43 61 74 - Fax: +33(0)4 72 43 87 13
> Lecturer at:
> Institut National des Sciences Appliquées de Lyon
> 20 Avenue Albert Einstein
> 69621 Villeurbanne Cedex
> France
> antoine.zimmermann@insa-lyon.fr
> http://zimmer.aprilfoolsreview.com/
>
>



-- 
Dr. Birte Glimm, Room 309
Department of Computer Science
University of Oxford
Parks Road
Oxford
OX1 3QD
United Kingdom
+44 (0)1865 283520

Received on Tuesday, 5 July 2011 14:24:41 UTC