Re: Question about literals in subject position

On 24 Sep 2009, at 11:52, Steve Harris wrote:

> On 24 Sep 2009, at 10:50, Bijan Parsia wrote:
> [snip]
>
>>> SPARQL/Query 1.0 is restricted to legal RDF graphs though, as per  
>>> (my reading at least) of section 12.6.
>>
>> Oh, right. Because entailment regimes are constrained to be  
>> subsets of RDF graphs.
>>
>> That seems like pointless restriction frankly. I mean, let's say I  
>> violate it (by supporting the natural, but pointless, queries  
>> above) in my entailment regime. What happens? Well, nothing  
>> really. It doesn't hurt interop. It just means that we have to  
>> revisit this spec if we want such an entailment regime to be  
>> compliant.
>> [snip]
>
> Well, not really, because the entailment regime can just say that  
> such triples are legal in it's data

I don't think it can.

> (you can't write that in RDF, but that's not critical), and then a  
> SPARQL query, using that regime can return such bindings. Again, by  
> my reading.

Here's what seems to be the controlling text:

"""An entailment regime specifies

	1.a subset of RDF graphs called well-formed for the regime
	2. an entailment relation between subsets of well-formed graphs and  
well-formed graphs."""

Thus, every entailment regime's entailment relation must go from RDF  
Graphs to RDF graphs. Which means you cannot have either an  
entailment regime with literal subjects, nor can you have an  
entailment regime that supports non-empty queries for BGPs binding to  
variables with literal subjects (e.g., via entailment).

This seems to be a minor quirk of the wording rather than a strong  
intended constraint.  But it does seem that if we stick to the spec  
text, we cannot define such entailment regimes.


>>> Certainly there are RDF-like systems which permit literals in the  
>>> subject position, so for some people at least it is significant.
>>
>> I guess the question is does it make sense to have the *extension*  
>> point forbid those as possible extensions? I guess I don't see the  
>> value.
>
> I think the extension point is allowed to change the behaviour there.

It seems to forbid such changes.

> The spec doesn't explicitly mandate RDFs restrictions, it's just  
> implied by the fact that SPARQL is defined in terms of RDF.
[snip]

See above :) It does so explicitly.

Cheers,
Bijan.

Received on Thursday, 24 September 2009 11:35:34 UTC