Re: LC Comment on Entailment Regimes: Treatment of Semantic Inconsistency

Michael,

Thank you for your comment about the SPARQL Entailment Regimes document.

The SPARQL Query conditions on extending the basic graph pattern
matching mechanism [1,2] specify that the effect of a query on an
inconsistent graph must be specified by each SPARQL extension. Thus,
not talking about this is not an option. Furthermore, the SPARQL Query
specification requires that entailment regimes should provide
conditions to prevent trivial infinite solution multisets as
appropriate to the regime. Note that this is already a weakend version
of the conditions for SPARQL 1.0 [2], but changing conditions of
existing specs is always difficult and completely dropping this
requirement seems out of the question.

For the RDF-Based Semantics it is clear that a consistency check does
not necessarily terminate, which is ok with the specification (e.g.,
one could terminate with a timeout). The WG has decided, however, that
if a system detects an inconsistency at any point during query
answering, that this should be communicated to the user. It is
explicitly allowed to do this in the form of a warning, e.g., together
with or after some answers. Thus, the specification, e.g., for the
RDF-Based Semantics, allows for terminating a consistency check after
a certain timeout and a conformant implementation is possible even for
undecidable logics. The WG decided, therefore, not to make changes to
the working draft.

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

[1] http://www.w3.org/TR/sparql11-query/#sparqlBGPExtend

[2] http://www.w3.org/TR/rdf-sparql-query/#sparqlBGPExtend

On 26 July 2011 15:10, Michael Schneider <schneid@fzi.de> wrote:
> Dear all!
>
> Document: SPARQL 1.1 Entailment Regimes
> State: LCWD
>
> The tables that specify the different entailment regimes have an entry
> "Inconsistency", which defines how systems should handle the case of
> inconsistent queried graphs. For example, for the OWL 2 Direct Entailment
> Regime, it is stated:
>
>    If the queried ontology is inconsistent under
>    OWL 2 Direct Semantics, the system MUST raise
>    an error.
>
> Proposal
> --------
>
> I propose to drop any normative treatment of inconsistent graphs. Having an
> informative discussion is sufficient in my opinion. Not treating this topic
> at all would also be fine.
>
> Rational
> --------
>
> * Not for all specified entailment regimes can be guaranteed that
> consistency checking will always lead to a positive result if the queried
> graph is really consistent (the normal case): The OWL 2 RDF-Based semantics
> is undecidable, and RIF-Core + OWL combinations and RIF-BLD entailment
> (although not directly specified in the spec) are as well. System
> implementers of query engines for these languages will not be able to create
> a compliant system. Even with a "SHOULD" instead of a "MUST", this would
> still lead to user expectations and to some "pressure" on implementers that
> such systems will at least have to "try". But trying could easily be
> hazardous for such systems, so an implementer may well decide to never try
> at all.
>
> * Even for decidable regimes, the additional effort to decide consistency
> may heavily negatively affect efficiency, and implementers may not be
> willing to support this, but just tell users of their system to check
> decidability of a queried graph in a preprocessing step with a dedicated
> reasoning tool (an entailment checker) instead, which is very easily be
> done. But then, their query engine will not be compliant.
>
> * Implementers may even feel that there is no real reason to specifically
> treat inconsistent queried graphs at all, but think that it is better to
> simply use their general query answering procedure to handle the special
> case in the same way as any other case, leading then to all possible results
> ("principle of explosion"). This is, for example, the case for some
> first-order logic reasoners with query-answering support, as for E
> (http://www4.informatik.tu-muenchen.de/~schulz/E/E.html). Implementations
> that build on top of such a system will not be compliant then, although they
> may be correctly and efficiently working systems in all other respects.
>
> * I believe that many current query systems with reasoning support, at least
> the majority of "light-weight" RDF entailment-rule-based systems, such as
> those based on the Jena RDF framework (using the Jena rule-based OWL
> implementations), do not perform automatic consistency checking on every
> query, but one has to request an entailment check explicitly, if supported
> at all. So, a normative handling of inconsistency would ignore current
> practice.
>
> * In general (a bit more philosophical), it is my opinion that the reasoning
> task "(in)consistency checking" should be considered separate from the task
> "query answering", and is IMO out of scope for the specification of SPARQL
> entailment regimes.
>
> Best regards,
> Michael
>
> --
> Dipl.-Inform. Michael Schneider
> Research Scientist, Information Process Engineering (IPE)
> Tel  : +49-721-9654-726
> Fax  : +49-721-9654-727
> Email: michael.schneider@fzi.de
> WWW  : http://www.fzi.de/michael.schneider
> ==============================================================================
> FZI Forschungszentrum Informatik an der Universität Karlsruhe
> Haid-und-Neu-Str. 10-14, D-76131 Karlsruhe
> Tel.: +49-721-9654-0, Fax: +49-721-9654-959
> Stiftung des bürgerlichen Rechts
> Stiftung Az: 14-0563.1 Regierungspräsidium Karlsruhe
> Vorstand: Dipl. Wi.-Ing. Michael Flor, Prof. Dr. rer. nat. Ralf Reussner,
> Prof. Dr. rer. nat. Dr. h.c. Wolffried Stucky, Prof. Dr. rer. nat. Rudi
> Studer
> Vorsitzender des Kuratoriums: Ministerialdirigent Günther Leßnerkraus
> ==============================================================================
>
>



-- 
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 Thursday, 10 November 2011 09:17:00 UTC