PR#2 for Provenance DM

[Markus Demleitner]

Hi DM,

On Tue, Sep 03, 2019 at 01:31:39PM +0000, Gerard Lemson wrote:
> > VO-DML as it stands already has a link to vocabularies through
> > SemanticConcept, which, however, requires SKOS vocabularies.  
> This is not correct. Can be any semantic vocabulary, though what
> that means we might want semantics WG to formalize further.
> For this reason we renamed <<skosconcept>> to <<semanticconcept>>.

Excellent.  I should make it a habit to check the actual RECs (rather
than my memory of early drafts) before blurting out publicly.

Looking at VO-DML REC-1.0, it says on semanticConcept/@vocabularyURI
(p. 46):

  If this attribute is given a value, it indicates that the attribute
  to which the SemanticConcept has been assigned MUST take values
  from the vocabulary identified by the URI. It may be possible to
  define a subset of its values using the topConcept attribute.

-- the topConcept part would be interesting if pulling things from,
say, the UAT (which, incidentally, might be a good idea in SimDM), but
I guess that's irrelevant here.

In what's left, there are two things that, now that we actually want
to apply it, aren't quite to my taste.

(1) That the terms MUST come from the vocabulary may be a bit too
harsh a constraint, in particular for the provenance people that so
far have language like 

  "Roles in relations to Agent are free text attributes, but if one
  of the terms in Table 9 applies, it should be used."

It *is* a bit of a step from "if applies, ... should" to "must".
Could the provenance folks perhaps give an indication if they'd go
with the "must" from VO-DML if Semantics promise to have a
lightweight process that will allow people quick and easy additions?
My current idea is in sect. 4.2 of http://docs.g-vo.org/vocinvo2.pdf,
and I'm fairly optimistic that we're ready to try something like this
by the time ProvDM becomes REC.

The "must" certainly will help clients a lot to do interesting
things, so if you can live with it, that would be great.

(2) VO-DML also says (p. 43):

  It is not necessary to indicate in the VO-DML model which of these
  options has been chosen, since the URI which is the value of the
  attribute will contain its own typing information.

This at least very strongly suggests that attributes controlled by a
SemanticConcept should contain full URIs.  This means you'd write

  http://www.g-vo.org/rdf/datalink/core#bias

when all you really want to say is

  bias

-- that's a lot of boilerplate string for a vanishing benefit in case
of SemanticConcepts that give a vocabulary URI
(see also the URI vs. fragment discussion in
http://mail.ivoa.net/pipermail/dm/2019-September/005936.html).

In that case, since VO-DML has the "must" just discussed, clients
can already know that they have to prepend "http://.../core#" to
every string.  I'd *really* like to exploit that knowledge and let
them write just "bias".

So, I wonder if something that says:

  When a semanticConcept/@vocabularyURI is in force, the concept URI
  is built as the simple string concatenation of the vocabulary URI
  and the attribute value.

could be sneaked in as an Erratum to VO-DML?  Or is there anything in
VO-DML that says there must be full URIs in such attributes?

        -- Markus