W3C home > Mailing lists > Public > public-rdf-dawg@w3.org > January to March 2010

Re: entailment review - part 2

From: Axel Polleres <axel.polleres@deri.org>
Date: Wed, 13 Jan 2010 10:50:44 +0000
Cc: "SPARQL Working Group" <public-rdf-dawg@w3.org>
Message-Id: <6FFD3AA5-1BA2-4573-8826-AD2664C397CE@deri.org>
To: Birte Glimm <birte.glimm@comlab.ox.ac.uk>

On 12 Jan 2010, at 18:30, Birte Glimm wrote:

> Axel,
> I think we are getting closer (at least before you pointed it out, I
> didn't even see leaving SG undefined (having no SG) as a possibility).
> I still think though that the current entailment spec does not foresee
> that. We say in Section 1.3:
> "All entailment regimes specified here use the same definition of a
> scoping graph as given for simple entailment."
> Simple entailment defines SG as follows:
> "If DS is the dataset of a query, pattern solutions are therefore
> understood to be not from the active graph of DS itself, but from an
> RDF graph, called the scoping graph, which is graph-equivalent to the
> active graph of DS but shares no blank nodes with DS or with BGP."
> Thus, we say that there is (ALWAYS) a scoping graph and the scoping
> graph is graph-equivalent to AG.

ok, and I don't see a problem with this. I just meant to say that the extension mechanism would 
allow that entailment regimes leave SG undefined for inconsistent AG. So, let's look at this 
condition by condition:

"1 -- The scoping graph, SG, corresponding to any consistent active graph AG is uniquely specified and is E-equivalent to AG."

is not a problem, because it only applies to consistent AG, so E-equivalence is not required for inconsistent AGs (i.e. simple equivalence, 
as we define it now is a viable alternative to define SG
in that case, as much as leaving it undefined would be.) Agreed?

"2 -- For any basic graph pattern BGP and pattern solution mapping P, P(BGP) is well-formed for E."

is not a problem either, it doesn't refer to SG, Agreed?

"3 -- For any scoping graph SG and answer set {P1 ... Pn} for a basic graph pattern BGP, and where {BGP1 .... BGPn} 
is a set of basic graph patterns all equivalent to BGP, none of which share any blank nodes with any other or with SG"

I understand "equivalent" hee as simple equivalence, i.e. not E-equivalence, so this condition isn't a problem either 
in my opinion, since simple equivalence even works in case SG is E-inconsistent. 

> If we do not change that, we do violate the conditions.

see above.

> If we change
> that to say that SG is graph-equivalent to AG only if AG is consistent
> and SG is undefined otherwise, then we can say that the conditions on
> BGP matching extensions do not apply in the case of an inconsistency.

that would be just an alternative, but actually, I am happy with the current 
alternative as well.

> Such a change would, however, also mean that we have to say precisely
> what happens if there is no SG. For example C1 cannot be used to limit
> the answers as it is currently.
> 
> How about saying in the ed note:
> "The behavior for inconsistent active graphs is not specified by the
> conditions on extensions of BGP matching. A consequence of defining SG
> to always be graph equivalent to AG (link to Sec 1.3) without
> requiring a consistency check in the RDFS entailment regime means,
> however, that in the case of an inconsistency, not all conditions on
> extensions of basic graph pattern matching are satisfied. For example,
> the first condition
> 
>  The scoping graph, SG, corresponding to any consistent active graph
> AG is uniquely specified and is E-equivalent to AG.
> 
> explicitly mentions that if the scoping graph is defined, then it must
> be E-equivalent (RDFS equivalent in this case) to any consistent AG.

Reflecting what I wrote above, I would suggest to add here: 
"The condition does not apply for inconsistent AGs. 
The current understanding of the WG is that this condition does thus not 
restrict on how we define SG in case of an E-inconsistent AG and at 
current we just define SG the same way as for consistent graphs."

> One

and here add: "alternative"

> way around this would be to define SG only if AG is consistent. If
> AG is inconsistent, one then has to specify conditions independent of
> SG that prevent infinite answers (an inconsistent graph trivially RDFS
> entails any RDF triple). Another possibility would be to specify an
> entailment regime for a subset of RDFS that cannot express
> inconsistencies, e.g., by defining well-formed graphs for the regime
> as those that contain only syntactically valid rdf XML literals.
> Several other alternatives, including different,
> implementation-dependent behaviors for systems that do/do not check
> consistency are currently being under discussion in the working
> group."
> 

Does that work for you?

> Sorry for being so picky with this, but I think it is quite essential
> to get that right and I find the current discussion very useful.

No problem at all, as long as you also allow me an equivalent level of pickiness. :-)

Axel

> Birte
> 
> 
> 2010/1/12 Axel Polleres <axel.polleres@deri.org>:
> > Concerning today's resolution, I suggest the following rewording in the Editor's note:
> >
> > "A consequence of not requiring a consistency check is that in the case of an inconsistency, not all conditions on extensions of basic graph pattern matching are satisfied. For example, the first condition
> >
> > The scoping graph, SG, corresponding to any consistent active graph AG is uniquely specified and is E-equivalent to AG.
> > explicitly mentions that the scoping graph must be E-equivalent (RDFS equivalent in this case) to the active graph and that AG must be consistent. Clearly that is not possible if the active graph is inconsistent. One way around this would be to specify an entailment regimes for a subset of RDFS that cannot express inconsistencies, e.g., be defining well-formed graphs for the regime as those that contain only syntactically valid rdf XML literals."
> >
> > --->
> >
> > "A consequence of not requiring a consistency check is that in the case of an inconsistency, not all conditions on extensions of basic graph pattern matching apply. For example, the first condition
> >
> > The scoping graph, SG, corresponding to any consistent active graph AG is uniquely specified and is E-equivalent to AG.
> >
> > explicitly mentions that the scoping graph must be E-equivalent (RDFS equivalent in this case) to any consistent AG. The behavior for inconsistent active graphs is unspecified. One way around this would be to specify an entailment regimes for a subset of RDFS that cannot express inconsistencies, e.g., be defining well-formed graphs for the regime as those that contain only syntactically valid rdf XML literals. Several other alternatives, including different, implementation-dependent behaviors for systems that do/do not check consistency are currently being under discussion in the working group."
> >
> >
> > best,
> > Axel
> >
> >
> > On 12 Jan 2010, at 14:50, Birte Glimm wrote:
> >
> >> I think this might be used as another hack, but I can't believe that
> >> this is intended and it looks like a hack. If we do that, then an
> >> inconsistent AG implies that there is no scoping graph and, in the
> >> absence of a scoping graph, condition 3 never applies and I can just
> >> define whatever I want as query answer. I have to think how that can
> >> be done properly though. The problem is that without a consistency
> >> check, we don't know whether we are in this situation and if we are,
> >> condition C1, which limits answers, is not usable as it is now because
> >> it depends on SG. This would mean that we either have to reformulate
> >> C1 so that it is independent of SG or to have different conditions for
> >> consistent and inconsistent graphs (without necessarily being able to
> >> distinguish in which situation we are).
> >>
> >> I don't see the editorial note as wrong or too strict though because
> >> at the moment, it is assumed that there is a scoping graph and the the
> >> scoping graph can be inconsistent. The entailment regimes assumes that
> >> SG is always equal to AG apart from bnode renaming. This means that if
> >> AG is inconsistent, then SG is defined and SG is E-equivalent to AG,
> >> but SG is not consistent as it would be required. Your suggestion is
> >> to not have an SG at all.
> >>
> >> Birte
> >>
> >>
> >>
> >>
> >> 2010/1/12 Axel Polleres <axel.polleres@deri.org>:
> >> >
> >> > On 12 Jan 2010, at 12:58, Birte Glimm wrote:
> >> >
> >> >> Axel,
> >> >> we've spend some more thoughts on the inconsistency issue here.
> >> >>
> >> >> >>> > I am not sure, actually, condition 1. doesn't require consistency of SG, it only says:
> >> >> >>> > "The scoping graph, SG, corresponding to any consistent active graph AG is uniquely specified and is E-equivalent to AG."
> >> >> >>> >
> >> >> >>> > So, hmmmm, *actually*, this wording actually doesn't limit at all what the scoping graph to an
> >> >> >>> > inconsistent graph is: In fact, this even seems to let open that the SG for an inconsistent
> >> >> >>> > graph is e.g. empty, implementation dependent, etc.
> >> >>
> >> >> I now discussed that with Boris (Ian is super busy at the moment) and
> >> >> in our understanding the SG is at best undefined if AG is inconsistent
> >> >> or, rather, there is no scoping graph in that case.
> >> >
> >> > Yes indeed, but that is just what I meant: there is no condition on the scoping graph if AG is inconsistent.
> >> > To my understanding, this allows us "to do what we want in case of inconsistencies, we just have to specify it:
> >> > This understanding doesn't follows from the condition itself, but from that sentence right before the conditions:
> >> >
> >> > "The effect of a query on an inconsistent graph is not covered by this specification, but must be specified by the particular
> >> > SPARQL extension."
> >> >
> >> > In particular, this doesn't preclude that an entailment regime specifies an SG even
> >> > for inconsistent AGs, or say that it is implementation-dependent: if consistency is checked by the
> >> > implementation we raise an error, otherwise we can specify a consistent SG. This would seem to be perfectly
> >> > inline with what we currently have and make the resp. editorial note unnecessary.
> >> >
> >> >> Thus, if AG is
> >> >> inconsistent, then you could do something that does not use a scoping
> >> >> graph, but if you do that, you violate condition 3 that basically says
> >> >> that SG must entail the answers.
> >> >
> >> > Still, slight disagreement here, because I really read here that you CAN
> >> > specify a consistent SG for such cases, as long as you say how. And even if not, again the condition is void, since it only applies to "for any scoping graph SG" (= forall), so if there is no scoping graph, again the condition doesn't apply...
> >> >
> >> >> Condition 3 cannot be satisfied if
> >> >
> >> > ... ex falso quod libet.
> >> >
> >> >> you have no SG or SG is undefined and you cannot have an SG because SG
> >> >> would have to be consistent (E-equivalence).
> >> >>
> >> >> Birte
> >> >
> >> > best,
> >> > Axel
> >>
> >>
> >>
> >> --
> >> Dr. Birte Glimm, Room 306
> >> Computing Laboratory
> >> Parks Road
> >> Oxford
> >> OX1 3QD
> >> United Kingdom
> >> +44 (0)1865 283529
> >>
> >
> >
> >
> 
> 
> 
> --
> Dr. Birte Glimm, Room 306
> Computing Laboratory
> Parks Road
> Oxford
> OX1 3QD
> United Kingdom
> +44 (0)1865 283529
> 
Received on Wednesday, 13 January 2010 10:51:20 GMT

This archive was generated by hypermail 2.3.1 : Tuesday, 26 March 2013 16:15:41 GMT