W3C home > Mailing lists > Public > semantic-web@w3.org > July 2010

Re: Show me the money - (was Subjects as Literals)

From: Antoine Zimmermann <antoine.zimmermann@deri.org>
Date: Thu, 01 Jul 2010 20:57:58 +0100
Message-ID: <4C2CF346.6000006@deri.org>
To: Jeremy Carroll <jeremy@topquadrant.com>
CC: Linked Data community <public-lod@w3.org>, Semantic Web <semantic-web@w3.org>
Jeremy, et al.,


I think people are already showing the money but they do it 2 cents 
after 2 cents ;-)  Here is my little 2 cent contribution.

To start with, I am on the side of the people in favour of allowing 
literals in the subject position. I've read the discussion and pondered 
the arguments on each side carefully, but I'm still convinced that it 
would ultimately be the better option to allow them. I understand the 
concern of those who would have to rework their architectures. Yet I 
don't believe that the cost exceeds the benefits for those who are 
starting to implement and for future implementations and future 
developments of the standards.
As Sandro said, RIF is using triples with literals as subjects, as 
Robert Fuller said (in the LOD list), reasoners are internally inferring 
triples with literals in subject position, and other use cases (more or 
less convincing) have been proposed here. Why can't those inferences and 
facts be exposed and published in an RDF document?


Now I'd like to show some of the strange things that happen when you 
combine SPARQL with inference regimes, that are due to the inability to 
have literals (in the syntax) as subject.


Assume that you have the following data, harvested from the Web:

:www dc:creator "Tim Berners-Lee" .
:www dc:creator "Tim Berners-Lee"^^xsd:string .
:www dc:creator :timbl .
:timbl owl:sameAs "Tim Berners-Lee" .


Note that literals are commonly used with dc:creator so this example is 
fairly realistic.

Now, let us consider the following query:

SELECT ?x WHERE {
    ?x a rdfs:Resource .
}

under the RDFS-entailment regime, this would provide the following answer:
?x --> :timbl

Now, the following query:

SELECT ?y WHERE {
    ?y a rdfs:Literal .
}

would provide no answer (under RDFS-entailment) and:

SELECT ?z WHERE {
    ?z a xsd:string .
}

would provide no answer (under RDFS-entailment).

Now, imagine a SPARQL engine with an "RDFS+sameAs"-entailment regime. 
The three queries above would give the following results:

?x --> :timbl   // first query
?y --> :timbl   // second query (I can infer that :timbl is a rdfs:Literal)
and the last would give nothing.

Now consider the query:

SELECT ?t WHERE {
     ?u a rdfs:Literal .
     ?u owl:sameAs ?t .
}

It would give:

?t --> "Tim Berners-Lee"
?t --> :timbl

However, the query:

SELECT ?u WHERE {
     ?u a rdfs:Literal .
     ?u owl:sameAs ?t .
}

would give ?u -> :timbl .


This is very weird for me.



Anyway, I do not expect such a change in the near future and the spec 
are like they are at the moment, so I live with it.




Regards,
-- 
Antoine Zimmermann
Post-doctoral researcher at:
Digital Enterprise Research Institute
National University of Ireland, Galway
IDA Business Park
Lower Dangan
Galway, Ireland
antoine.zimmermann@deri.org
http://vmgal34.deri.ie/~antzim/
Received on Thursday, 1 July 2010 19:58:36 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 1 March 2016 07:42:20 UTC