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

Re: DELETE and blank nodes

From: Birte Glimm <birte.glimm@comlab.ox.ac.uk>
Date: Wed, 3 Mar 2010 17:45:03 +0000
Message-ID: <492f2b0b1003030945i3921973fs814204c2221c57c7@mail.gmail.com>
To: Ivan Herman <ivan@w3.org>
Cc: Andy Seaborne <andy.seaborne@talis.com>, Lee Feigenbaum <lee@thefigtrees.net>, SPARQL Working Group <public-rdf-dawg@w3.org>
On 3 March 2010 17:22, Ivan Herman <ivan@w3.org> wrote:
> Birte,
> although we agree on our choice on the original issue at hand, I have a
> clarification question... Is your mental model such that once the
> variable binding is done, based on the WHERE clause, the semantics of
> DELETE is that all triples are deleted that can be inferred by the
> DELETE template's triples?
> This is not defined and section 9 of the entailment document leaves that
> option open. That being said I agree (if this is indeed what you have in
> mind) that we should not make this approach impossible via syntactic
> means...

Yes, it is not defined how UPDATE would work with entailment regimes
yet. My main point is that I wouldn't like reasonable options to be
closed now, so that when we come to define UPDATE under non-simple
entailment, the reasonable options are excluded or UPDATE/DELETE must
be defined quite differently for non-simple entailment from how it is
defined for simple entailment.
A natural way of defining DELETE with non-simple entailment is to
apply the WHERE clause as usual (with whatever semantics the
entailment regime in use provides) and then instantiate the template
part and delete exactly the ground triples that you get from the
instantiation (no entailment for the template). Here's an example (not
very realistic): assume we have
DELETE { ex:D rdfs:subClassOf ?x } WHERE { ?x rdfs:subClassOf ex:C }
under RDFS entailment over the triples:
ex:A rdfs:subClassOf ex:B .
ex:B rdfs:subClassOf ex:C .
ex:D rdfs:subClassOf ex:B .
This would give you ?x/ex:A and ?x/ex:B from evaluation the WHERE
clause with RDFS entailment. Then the template part is instantiated
once to
ex:D rdfs:subClassOf ex:A
which is not in the data, so nothing is deleted, and once into
ex:D rdfs:subClassOf ex:B
which is in the data and gets deleted. If using OWL Direct Semantics,
the BGP might contain bnodes just to represent some OWL structures and
these bnodes cannot just be replaced by variables if I want to use OWL
Direct Semantics for evaluating the WHERE clause.

I don't suggest to define the actual semantics of updates for
non-simple entailment right now and one could also say that one uses
simple entailment to evaluate the WHERE clause no matter which
entailment regime is normally used. In the latter case, one can just
replace the bnodes with variables as Andy suggested. My main concern
is that if we forbid bnodes now, then the first possible semantics
that I described with the example above is kind of ruled out upfront
at least for OWL Direct Semantics.

I hope that clarifies my thoughts at least a bit.


> Ivan
> On 2010-3-3 18:04 , Birte Glimm wrote:
>> On 3 March 2010 15:40, Andy Seaborne <andy.seaborne@talis.com> wrote:
>>> On 03/03/2010 3:32 PM, Birte Glimm wrote:
>>>> Thanks a lot Lee for the great summary!
>>>> On 3 March 2010 06:16, Lee Feigenbaum<lee@thefigtrees.net>  wrote:
>>>> [snip]
>>>>>>  == The Proposals ==
>>>>>>  I see only two realistic proposals emerging from this.
>>>>>>  1/ We prohibit blank nodes in the DELETE template completely.
>>>> This might work nicely for simple entailment, but will not work for
>>>> OWL (maybe RIF too). E.g., if I have a statement that Peter has some
>>>> parent who's a Lawyer, I can say that in OWL functional-style syntax
>>>> with:
>>>> ClassAssertion(ObjectSomeValuesFrom(ex:hasParent ex:Lawyer) ex:Peter)
>>>> For this, the (normative) mapping to triples gives:
>>>> ex:Peter a _:x.
>>>> _:x a owl:restriction .
>>>> _:x owl:onProperty ex:hasParent .
>>>> _:x owl:someValuesFrom ex:Person .
>>> ex:Lawyer?
>> Ups, yes.
>>>> for a fresh bnode _:x. There is no way to represent this OWL statement
>>>> without bnodes, but it is always possible to use a square bracker
>>>> form.
>>> I think Lee is referring to syntactic use of bnodes in a delete template,
>>> not bnodes that might be found by a query pattern and bound to a variable.
>>> So _:x is a new bnode not the bnode with that label in the data.
>>> It would be deleted by:
>>> {
>>>  ex:Peter a ?var.
>>>  ?var a owl:restriction .
>>>  ?var owl:onProperty ex:hasParent .
>>>  ?var owl:someValuesFrom ex:Lawyer .
>>> }
>>> using variables with ?var binding to exactly the bnode that was syntactical
>>> _:x in the data.
>> Unfortunately not, because such a bnode binding will be filtered out
>> from the results by the conditions on answers. If it were not filtered
>> out, you could get infinite results because replacing _:x with :_xx or
>> _:xxx or ... would equally be entailed. Since we use entailment, using
>> the same sub-graph criterion that simple entailment uses to limit
>> answers with just different bnode labels does not really work for
>> entailments. For anonymous individuals in OWL (bnodes in individual
>> positions) we use skolemisation, but the bnodes that are just used to
>> represent some concepts are invisible to the reasoner and skolemising
>> them is quite strange for Direct Semantics. Such a BGP would, actually
>> not even be legal under Direct Semantics because it cannot really be
>> turned into an OWL structural object.
>> Birte
>>>        Andy
> --
> Ivan Herman, W3C Semantic Web Activity Lead
> Home: http://www.w3.org/People/Ivan/
> mobile: +31-641044153
> PGP Key: http://www.ivan-herman.net/pgpkey.html
> FOAF   : http://www.ivan-herman.net/foaf.rdf
> vCard  : http://www.ivan-herman.net/HermanIvan.vcf

Dr. Birte Glimm, Room 306
Computing Laboratory
Parks Road
United Kingdom
+44 (0)1865 283529
Received on Wednesday, 3 March 2010 17:45:37 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 7 January 2015 15:00:59 UTC