Map csvw:TableSchema and RML

[VladimirAlexiev]

https://kg-construct.github.io/rml-io/spec/docs/#logical-source-in-rml uses CSVW but only to describe the location and dialect of a Table.
In a dataspace scenario we use TableSchema to describe some csv, to then provision ingest to a timeseries database, while enabling discovery and other semantic integration approaches.
CSVW is a nice focused spec dedicated to CSV semantic description, so I think it's worth examining it, mapping it to RML, and maybe finding some features to add to RML.

The CSVW Manifest is JSON (also JSON-LD) and it has one "defect": it maps columns not to RDF props but to URL templates. So maybe we should retool to use RML instead of CSVW.

[DylanVanAssche]

Hi!

It is certainly possible to use TableSchema as data access description. RML-IO is made in such a way that you can pick your data access descriptions. However, we don't have to describe them all in the specification.

[VladimirAlexiev]

@DylanVanAssche But isn't it useful to define a CSVW "reference formulation"?

[DylanVanAssche]

@VladimirAlexiev That's indeed the plan: defining a CSVW reference formulation, but in the RML IO Registry.
The idea is to keep a simple set of reference formulations in RML Core, RML IO like JSON, CSV, XML, RDB. Any other reference formulation is also welcome, but defined in the RML IO Registry to avoid that processors must implement 20 reference formulations to comply with the RML IO specification (which will never happen of course as it becomes so complex).