Re: R2RML Linked Data Aspects

Am 15.03.2011 11:30, schrieb Ivan Herman:
>> Am 14.03.2011 18:55, schrieb Ivan Herman:
>>>> 2. *namespaces* - I could not find any possibility to declare namespaces
>>>> and respective prefixes in the current draft. is that left out
>>>> intentionally? If not, we should add a respective declaration and make
>>>> clear, that only namespaces which are governed by the user creating the
>>>> mapping can be made de-referencable later on for Linked Data publishing.
>>>>
>>>
>>> I just want to understand: do you mean namespaces to be used within an R2RML file or namespaces and prefixes in the generated file?
>>>
>>> I am not sure I understand the need for the former; I suspect this is not what you meant.
>>>
>>> As for the latter: my understanding is that R2RML defines a way to generate RDF triples from tables. I mean: RDF triples in the abstract, conceptual sense, and not in any particular serialization. On the other hand, the usage of prefixes/namespaces is a matter of a particular serialization; I am not sure it has a place in R2RML...
>>
>> In some literal values of R2RML properties (such as e.g. rr:SQLQuery)
>> (as well as inside some tables) it would make sense to use CURIE-like
>> identifiers. Currently it looks like it is left unclear, wether this is
>> possible and what namespaces will be used. From my point of view we need
>> at least two distinguished default namespaces - one for schema elements
>> (i.e. classen and properties) which are generated by R2RML and one for
>> the actual data.
> 
> Well, that is what I am questioning, in fact. *RDF* does not have a concept of namespaces, only serializations do. And the different serializations (RDF/XML, Turtle, RDFa, soon JSON) may have different approaches. R2RML does not define which serialization is used at the end of the process...

I understand that, but this was not the issue. The issue is:

We have to generate IRIs manually sometimes with R2RML, e.g. in
SQL-based logical table definitions. The IRIs are created by using a
namespace and adding a local name component. Currently, if I look at the
examples, this is done by writing and concatenating the namespace always
in full. Also, if we want to generate various types of IRIs from
database columns it would be good to be able to use CURIEs. For both the
definition of namespaces which can be used would be beneficial.

The issue is in a way similar to that of SPARQL also there you declare
prefixes as shortcuts within the query although the resulting RDF
serialization (e.g. of a DESCRIBE query) might do that differently.


Sören

Received on Tuesday, 15 March 2011 12:31:51 UTC