Re: RDF is a framework, not a vocabulary

On Fri, Jul 12, 2024 at 10:19 AM Souripriya Das
<souripriya.das@oracle.com> wrote:
>
> I am just wondering if the recent wave of discussions is taking us beyond the "framework" focus of RDF over to the territory of new vocabularies that can potentially be created on top of the enhanced framework of RDF.

Yes, the discussions are indeed beyond RDF itself, going into use
thereof. It is good to be very clear on that, and to limit that to
*testing* of what we do add to RDF, which in itself must be kept down
to a required minimum.

> Since the goal of our WG is to determine the essential extensions to the framework in current RDF that will be critical for enabling and simplifying the target capabilities -- statements about statements and support for duplicate triples, concerns about issues that are more pertinent to development of interesting vocabularies on top of RDF1.2 (similar to SKOS on top of RDF1.1) should be avoided, IMHO.

I believe I fully agree. Just to be clear: we need to test this (which
is what I'm doing), including whether it affects the design of
vocabularies on top of RDF. This includes assessing whether these
additions "change the game" so to speak.

(For example, if we were to propose a new mechanism which would be at
odds with, say, the notion of `rdf:type`, that would "change
everything"; and could create confusion and divergent new designs,
which could be quite harmful. Thankfully, we're not doing that.)

From what I see, with triple terms and rdf:reifies, more *convergence*
is now possible between different forms of expression (expressions
that were already possible and in use in RDF 1.1). Some advanced
designs can make use of that, but nothing that is necessary for
every-day usage and regular ontology building. What we get mainly
empowers wider, decentralized data integration.

Specifically. as I outline in [1], with just rdf:reifies and triple
terms, we can form a bridge between annotated relationships and full
N-ary relations; something that has hitherto *required* OWL to entail
shorthand forms. And going to the full form in an existing application
either requires remodelling or complex querying; leaving users without
OWL machinery "in the dust". Additionally, with just a minimal
entailment that corresponds to the "unstar form", existing standards
such as SHACL and OWL can make use of data utilizing RDF-star without
upgrading those two standards at all.

From the other end, systems with OWL capabilities can also form a full
connection between our new reifiers to existing, full-fledged N-ary
relationships solely by adding to their ontologies (the "T-box"), and
not touching the data ("A-box") at all. So the promise of RDF-star no
not require remodelling once more details are discovered holds up. And
even better, it improves upon the integration capabilities of data
that has been differently modelled (such as between flat data using DC
Terms and more complex stuff using BIBFRAME).

> As long as RDF1.2 allows association of a term with a triple (or a block of triples, in case of many-to-many), data creators can designate such a term to belong to custom classes – :Relation, :Reification, :Myth, :Nonsense, etc. – that make sense in their domain. If there is a common set of such classes that are found to be important in many domains, enthusiasts can create vocabularies to capture those. Whether such classification determines if a "named" (put your favorite term here) triple (or block of triples) should be considered as "asserted" or not -- should be up to the vocabulary designers, not our WG.

I wholly agree. That is why I've explored how such powerful entailed
assertion capabilities are already available in OWL. This kind of
truth-generating power [3] does not belong in the simple RDF
substrate.

If you're worried about the suggestion of adding `rdf:Relationship`,
that is just the name I propose for what we've tentatively called
`rdf:Triple` or `rdf:TripleTerm` --  i.e. the range of `rdf:reifies`.
We suggested those names before we concluded that triple terms are
indeed wholly "transparent" (within the same interpretation as the
rest of the graph). Since RDF concepts [2] says that "Asserting an RDF
triple says that some relationship, indicated by the predicate, holds
between the resources denoted by the subject and object.", I think it
could be further clarified that a triple actually denotes that
relationship (both when it is a member of the graph and when it is
used as a triple term to *refer* to such a relationship, orthogonal to
its known truth value). But I can live with any reasonably sensible
name. (And I *hope* we won't need months debating any difference
between "relationship" and e.g. "attribute"; as there is no such
difference in RDF...)

> Let us focus on the "framework" improvement part only and leave the vocabulary aspects to data creators and enthusiasts.

Yes! We do need to assess that the framework improvements work for the
data creators and consumers (especially the latter, IMHO), but that is
a contained activity. And any vocabularies that may emerge as
potential shared ones would need their own community groups.

> Hoping for timely and successful completion of RDF1.2 spec,

Hear, hear!

Cheers,
Niklas

[1]: <https://lists.w3.org/Archives/Public/public-rdf-star-wg/2024Jul/0038.html>
[2]: <https://www.w3.org/TR/rdf-concepts/>
[3]: <https://gist.github.com/niklasl/cf8871f82564749153d929f747228ef3>


> Souri.
>

Received on Friday, 12 July 2024 13:49:41 UTC