– DRAFT –
RDF-star WG biweekly focused meeting

Respect the conventions on the web for use of IRIs

ktk: who could summarize the topic?

andy: There was a discussion about opacity and transparency in the semantic task force and we discussed use cases but we did not see that in the whole context of the whole web and the more basic conventions

enrico: in many cases local uris can solve the problem

<gkellogg> URI transparency can be inferred from LinkedData principles: https://www.w3.org/wiki/LinkedData

<niklasl> I agree. The value/hazard of unambiguous, precise identifiers (always transparent) is crucial *also* for citing statements.

pfps: we are getting close to social meaning. We can't build a forced agreement into the principles of RDF

TallTed: the basic idea is to talk about what someone said, I need to know who else talked about my triple, so I need to be able to clearly identify it. We need a good definition of asserted/unasserted and opaque and transparent

<pfps> asserted in a graph iff element of the graph, no?

<pfps> https://pfps.me/best-president-ever denotes the same thing everywhere?

pchampin: I would like to add a clarification to pfps statement, one iri should always mean the same in the whole web, no different meanings for the same iri

Kurt: one question: when we talk about reification, what we do is we have a graph which describes a triple but not the triple itself

<enrico> ted: the formal definition of transparent vs opaque IRIS, and of asserted triples (in the graph) and triple terms is at https://github.com/w3c/rdf-star-wg/wiki/RDF-star-%22baseline-with-IRI-opacity%22

Kurt: now opacity vs. transparency: do we have a problem to distinguish between the triple itself and its representation (components). The iri can identify the pieces of the triple.
… the key point is that we do not assert triples but describe them

<Zakim> pfps, you wanted to say that RDF absolutely does not say that the denotion of an IRI is the same everywhere, now intended (or consensus, or majority) meaning might be different

pfps: the formal basis of RDF does allow for a concept like "intended meaning"

<enrico> doerthe: we can describe the "meaning" of a triple term, or the "syntactic structure" of a triple term - that's captured by transparent vs opaque IRIs

<pfps> intended meaning is very different from denotation

<niklasl> Agree with Peter, and I'd add that the Semantic Web project is an attempt to share one interpretation of the "giant global graph".

<pfps> and RDF is a language used in the Semantic Web project, but RDF is not the Semantic Web, and this WG is about RDF

enrico: we do have the distinction between asserted and unasserted triples, we have annotated triple terms, we can annotet the triple itself or (opaque case) the structure

<TallTed> <http://example.com/ernie#agent> ex:asserted {| http://example.com/ernie#agent ex:president_rating ex:best |} . <http://example.com/bert#agent> ex:asserted {| http://example.com/ernie#agent ex:muppet_rating ex:best |} . <http://example.com/TallTed#agent> foaf:knows http://example.com/ernie#agent , http://example.com/bert#agent .

<TallTed> DESCRIBE <http://example.com/ernie#agent>

<enrico> doerthe: agree

AndyS: responding to pfps: so far we did not have the possibility of one graphs referring to another. Within one graph, iris have the same meaning.

<pfps> it is certainly the case that within an interpretation there is a functional mapping from IRIs to denotation, and that (sort of) implies that they have to have the same intended meaning

Kurt: If I refer from one graph to another, it might be the same, but it I move a triple from one graph to another, I change contexts, the triple does not actually change. I triple can be true in one graph and not in another graph.
… if I define something in graph 1 and then change to graph 2, the meaning of the iri and a triple should not change, but the graph changes
… this is even stronger if I for example use shacl
… the core assertion should not change

niklas1: if we arrive at transparency, we should mention it. If we do not need the difference between different graphs, we do not need opacity. but different meanings between graphs are out of scope

<TallTed> { ex:water ex:state ex:liquid . ex:water ex:state ex:solid . ex:solid owl:disjointWith ex:liquid . } ==> uh oh

AndyS: If you put iris in a graph, you are committing to them.

<Souri> graph :graph1960s { :kennedy :firstname "John" } graph :graph2000s { :kennedy :firstname "Edward" } ==> we have to be careful if we merge these two graphs

AndyS: I am concerned that we keep forgetting about decisions we made, we should write notes and stick to them

<gkellogg> +1 to what Andy said

<niklasl> +1 we need that Note on decisions

Kurt: I would like to talk about a proposal I have in some of the next meetings

<pchampin> Souri: IMO, your example is flawed. Using the same IRI to mean two different persons is simply wrong.

<gkellogg> We have a boilerplate "What's New in RDF 1.2" document at https://w3c.github.io/rdf-new/spec/

<Kurt> https://github.com/w3c/rdf-star-wg/wiki/Proposal:-Named-Node-Expressions

<Souri> They are independently authored. We have no control.

AndyS: can you relate your proposal to our discussions (and write them explicitely)

<ktk> q`?

Kurt: proposal at the moment: https://github.com/w3c/rdf-star-wg/wiki/Proposal:-Named-Node-Expressions

<enrico> List of various proposals and examples: https://github.com/w3c/rdf-star-wg/wiki

enrico: all proposals so far are listed here https://github.com/w3c/rdf-star-wg/wiki

Kurt: I will add my proposal

<pfps> The WG has had an example where an IRI has different denotations and has argued over it quite a bit - :superman

<pchampin> sorry to nitpick, but ":john" is not an IRI!

Souri: whenever we go for iris, we do not have any control over it... :john has different meanings in different contexts
… two different graphs are not the same thing...

<doerthe> I agree with pchampin

AndyS: I agree, you have to be careful the moment you merge graphs

<niklasl> Careful, or full of trust.

Souri: we need to limit our scope, we should sty within the graphs, do not go beyond. same holds for triples, graphs can disagree, we always talk within one graph

<Zakim> pfps, you wanted to mention that some of the examples used in the WG have depended on IRIs having different denotations

pfps: I am getting uncomfortable, the working group used to deal with different denotation. Remember the superman example where the different meanings were important
… we do not have fixed denotations

<niklasl> *All* identities are useful "fictions". Let's not go there?

<Souri> Within a single graph, two uses of :Batman must have the same intended meaning, right?

pfps: think of different batmans, or different ideas of what :europe could mean

<niklasl> Yes to Souri.

<Souri> Integrate, but be careful when merging.

tl: RDF is about integrating data from everywhere

<pfps> There was a long discussion about whether social meaning was a fundamental part of the Semantic Web. But that's not RDF, just a use of RDF. One might argue that RDF should support social meaning, whatever that is, but building social meaning into RDF does not seem viable to me.

<pfps> And anyway the result of the discussions, as I recall it, was that social meaning was not a fundamental part of the Semantic Web.

Kurt: about comic book characters, it is not about a concept having meaning but a concept in a data model. Batman of tv show is different than the batman of the movie and these differences are given through the surrounding model, a concept comes with a graph of information. We should not get too far into the semantics. Does it represent anything

before you create context? Without that we only know about similarities

ktk: what are we talking about? what is our point?

<Kurt> +1

<pchampin> +1

<Zakim> TallTed, you wanted to mention the difference between an ontology, and a set of instance data

<pfps> See https://www.w3.org/wiki/SocialMeaning and https://www.w3.org/2001/sw/meetings/tech-200303/social-meaning

TallTed: for a focussed discussion, we do not have a lot of focus. We need to agree about the terms we use in our discussion
… Kurt, did ou really do an ontology?

Kurt: I define Shacl shapes for a graph abou comic books

<pfps> See also the history of the RDF documents, particularly what happened to https://www.w3.org/TR/2003/WD-rdf-concepts-20030123/#section-Social

TallTed: I am concerned that each wrong use of a term leads to 6 week discussions, therefore we need to be careful

<tl> @pfps because the logicians took over...

enrico: what are we talking about? I think it was opacity and whether we need it. I think that the superman example is misleading. In RDF, an iri needs to denote one object. This differs for human language, we have strings and these have different meanings depending on contexts.
… this is an issue for RDF

<TallTed> URIs are often (in my world) described as "superkeys" because they work across (relational) schemas/catalogs, unlike the IDs used as "keys" within SQL-style RDBMS

enrico: superman has a fictional life and I can use an iri to talk about it, I can also choose an iri to have the movie

<niklasl> Do we need a note titled "What Wittgenstein said about the use of IRIs"? I hope not (and it would be empty anyway).

enrico: iris are different from constants

<TallTed> man has never landed on the moon. it was a movie. or was it?

enrico: iris universally mean the same for everybody and that is why opacity should not play a role here. Maybe there are provenance examples which still need it

<niklasl> And you can talk about that encoding of the triple, using strings.

enrico: we cannot refer to a triple as a syntactic structure and that is what we would win by adding opacity

pchampin: I try to get consensus: pfps is right that social meaning should not play a role in RDF, enrico is right that we have unique uris, Souri is right, that people make mistakes also using uris

<niklasl> +1 to pchampin (and by association to the meaning of the quoted statements he used)

pchampin: that should help us to get consensus

Kurt: I would like to argue against something pfps brought up, if two resources for example state different birth dates for the same person, then merging brings problems.
… So moving from one model to another can make triples wrong. I need to disambiguate before merging. Graphs are not always consistent, not even if I add constraint languages
… In my opinion, there is a deeper relation between iris and what the refer to we did not cover yet
… even though this might opens a can of worms

enrico: I will give a talk about RDF star next week, input is welcome

<TallTed> most important is to make clear that RDF-star (now RDF 1.2) is still a moving target

<niklasl> Next focused meeting, I think we should talk about if we've come to terms with reifiers being many-to-many (working as relators/N-ary)?

<Kurt> Will the Netherlands talk be recorded?

<TallTed> talk about goals, more than what we've achieved

enrico: Semantic task force: we should wrap up the transparancy/opacity discussion, in my opinion, we need use cases for that