Re: RDB2RDF semantics

Sören,

On 20 Jul 2010, at 18:03, Sören Auer wrote:
> As I mentioned in today's telco, from my POV we should (in order to  
> define the semantics of the mapping language) focus on the  
> translation from SPARQL algebra to relational algebra - both are  
> relatively well defined fragments of FOL. I prepared a first, still  
> very rudimentary draft of what I mean here (particularly Section 2,  
> Section 1 just describes the SPARQL algebra and evaluation):
>
> http://www.informatik.uni-leipzig.de/~auer/rdb2rdf/semantics.pdf
>
> The idea is to define a RDB2RDF mapping and then express the SPARQL  
> query evaluation using relational algebra. This way we can actually  
> standardize the semantics of an RDB2RDF mapping (without having to  
> deal with the variations of RDB query syntaxes i.e. SQL).

So if I get this right, then you propose to formally define this:

"Given a database D and a mapping M, here's how to evaluate a SPARQL  
algebra expression E."

Isn't that more than we need to do? Intuitively, it seems that all we  
need is:

"Given a database D and mapping M, here's the corresponding RDF graph  
G."

SPARQL evaluation over G is already defined in the SPARQL spec, so I  
think that just defining what triples G contains is sufficient, and  
should be simpler in terms of spec space?

I also believe that the actual definitions of these two things  
wouldn't look *that* different if you use the Perez et al formalism.

Best,
Richard


> I somehow have the impression that introducing another formalism  
> (i.e. datalog) unnecessarily complicates the problem further.

>
> Best,
>
> Sören
>

Received on Wednesday, 21 July 2010 16:06:36 UTC