Re: Feedback on RDF Graphs: Conceptual Role and Practical Use Cases

On Fri, Oct 3, 2025 at 17:10 Pierre-Antoine Champin <pierre-antoine@w3.org>
wrote:

> Hi Dan,
> On 02/10/2025 23:05, Dan Brickley wrote:
>
> (...)
>
> One reason to want to assert bundles of triples together is that when
> unknown URIs, URIs with unclear exact reference, or bnodes are in play,
>  then you don’t really know what any triple is saying about the real world.
> The idea that we build everything up from completely atomic triples is a
> bit of a foundational myth in our commumity. In practice we might want to
> take a handful of identification-related properties into account when
> trying to work out what some piece of RDF is claiming.
>
> Lots of perfectly fine RDF which can have clear meaning and utility, looks
> unintelligible at the triple level.
>
> something1 someProperty something2
>
> In the rare case when something1 and something2 are widely agreed URIs
> whose documentation and use doesn’t leave wiggleroom for understanding
> exactly what they’re referring to, great. Otherwise we need to go deeper
> into the graph to fogure out whether these entities have homepages, phone
> numbers, dnaChwcksums, DOIs, GTINs or whatever. Knowing which properties
> are functional, inverse functional, inverses of other properties etc can
> also help narrow things down.
>
> I don’t really know what it means to assert a triple that uses bnodes or
> non-famous URIs. Whereas asserting a medium sized graph seems closer to a
> much more natural social act…
>
> I don't disagree, but for me this is still a different problem than
> capturing the semantics of datasets and named graphs.
>
> Quoting the original "Named Graph" paper by Carroll et al.:
>
> Issues as to how to resolve conflicts between different graphs, and how to
> determine [which graphs are accepted], are seen as pragmatic issues, to be
> dealt with by application developers, rather than logical issues to be
> dealt with by formal semantics.
>
>

…which is why the triple-centric work rejecting use of the Named Graphs
concepts in rdf and sparql feels so off.


best
>
>