Re: soliciting your favorite (public) SPARQL queries!

Hi Axel --

Good to see some thinking about fundamentals.

The omission of negation from the SPARQL spec (and hence the need for your
ingenious workarounds) seems to be based on a confusion that can perhaps be
explained away like this....

What some semantic web folks seem to want is that when new facts are added,
old conclusions don't go away.  They want things to be monotonic, and they
therefore deprecate SQL-style negation-as-failure (NAF).

Now suppose an old conclusion p depends on ~r  in a consistent theory, and
that an update r is pending.

We could just add r.  p would still hold, but the new theory has both r and
~r.  It's inconsistent.  That means  that a naive theorem prover can prove
absolutely anything from it.  A better theorem prover would probably refuse
to compute with it.  Neither is a desirable outcome.

But wait.  In most practical circumstances, adding r is a way of saying that
~r should be removed.  So, take the update to mean "add r and also delete
~r".   The new theory is consistent, and p no longer holds.

So, the price of keeping consistency through an update is that an old
conclusion p may no longer  be entailed.  Under consistent update, using
classical logic and using NAF lead to the *same *behavior.

If we use Clark's result [1] to view a logic program with NAF as simply
shorthand for a set of clauses in classical logic, the above starts to look
kind of obvious.

A similar argument could be advanced for the inclusion of aggregation in an
extended SPARQL spec.  Now is perhaps a good time to avoid an error that the
SQL folks made -- the results from SQL aggregations are implementation
dependent.  That's a bad idea for SQL, and a terrible one for on-the-fly
linked data and the Semantic Web.

Hope this helps.

                                         -- Adrian


[1]  http://www.doc.ic.ac.uk/~klc/NegAsFailure.pdf

Internet Business Logic
A Wiki and SOA Endpoint for Executable Open Vocabulary English over SQL and
RDF
Online at www.reengineeringllc.com    Shared use is free

Adrian Walker
Reengineering


On Thu, Aug 21, 2008 at 11:51 AM, Axel Polleres <axel.polleres@deri.org>wrote:

>
> Tackling the question from the more theoretical side,
> I like non-monotonic SPARQL queries like the ones modeling set difference.
>
> E.g.
> "Give me all persons *without* an email address" in a certain FOAF graph.
>
>
> i) It is already folklore, that you can do that with using the !bound()
> filter outside an optional, i.e.
>
> SELECT ?X
> FROM G
> WHERE { ?X a foaf:Person
>        OPTIONAL { ?X foaf:mbox ?M}
>        FILTER (! bound(?X) ) }
>
>
> ii) What some people might find surprising is that I can achieve the same
> result without using a FILTER, more generally that I can express
>
> SELECT ?X
> FROM G
> FROM NAMED <boundchecker.rdf>
> WHERE
>  {
>    { ?X a foaf:Person OPTIONAL{ ?X foaf:mbox ?M} }
>    GRAPH <boundchecker.rdf>{ ?M :is :unbound }
>  }
>
> where <boundchecker.rdf> is the graph containing the single triple
>
>   _:b :is :unbound.
>
> Maybe requires some thinking, but is a nice example :-)
>
> (Short explanation: the blanknode in Graph <boundchecker.rdf> only matches
> to unbound variables from the optional patttern. Note that non-well-designed
> OPTIONAL patterns are not commutative, see [1].
> Actually, [1] "kind of" conjectured that non-well-designed patterns are
> useless, but - as this query shows - they aren't really entirely useless.)
>
> Axel
>
> 1. http://iswc2006.semanticweb.org/items/Arenas2006bv.pdf
>
>
> p.s.: Since I didn't see a similar one before, I claim copyright for that
> one, basically, it is very easily generalizable to model arbitrary queries
>  SELECT ... P WITHOUT P'
>  ;-)
>
>
>
> Lee Feigenbaum wrote:
>
>> Hi everyone,
>>
>> I'm putting together a "SPARQL by Example" tutorial, which is, as the name
>> suggests, a step-by-step introduction to SPARQL taught almost entirely
>> through complete, runnable SPARQL queries.
>>
>> So far, I've gathered a great deal of example queries myself, but I know
>> that many subscribers to these lists probably have favorite queries of their
>> own that you might be willing to share with me.
>>
>> I'm looking for:
>>
>> 1) SPARQL queries
>> 2) ...that can be run by anyone (no private data sets)
>> 3a)...either by running the query against a public SPARQL endpoint
>> 3b)...or by using a public SPARQL endpoint that will fetch HTTP-accessible
>> RDF data (e.g. sparql.org or demo.openlinksw.com)
>> 4) ...that answers a real* question
>> 5) ...and that is fun!**
>>
>> * real is in the eye of the beholder, I imagine, but I'm not looking for
>>  "finds the predicates that relate ex:s and ex:o in this sample RDF graph"
>>
>> ** fun is also in the eye of the beholder. fun can be a query on fun data;
>> a clever query that may illustrate a particular SPARQL construct ("trick");
>> a query that integrates interesting information; a query with surprising
>> results; etc.
>>
>> thanks to anyone who is able to contribute!
>> Lee
>>
>> PS I plan to make the tutorial slides available online under an
>> appropriate CC license once they are completed.
>>
>
>
> --
> Dr. Axel Polleres, Digital Enterprise Research Institute (DERI)
> email: axel.polleres@deri.org  url: http://www.polleres.net/
>
> Everything is possible:
> rdfs:subClassOf rdfs:subPropertyOf rdfs:Resource.
> rdfs:subClassOf rdfs:subPropertyOf rdfs:subPropertyOf.
> rdf:type rdfs:subPropertyOf rdfs:subClassOf.
> rdfs:subClassOf rdf:type owl:SymmetricProperty.
>
>