W3C

RDF-star WG biweekly focused meeting

30 January 2025

Attendees

Present
AndyS, AZ, doerthe, Dominik_T, eBremer, enrico, fsasaki, gkellogg, gtw, james, ktk, niklasl, ora, pchampin, Souri, TallTed, tl
Regrets
olaf, pfps
Chair
ora
Scribe
gkellogg

Meeting minutes

is rdf:dirLangString a required datatype for RDF entailment? 1

ora: pfps abstained from the vote

gkellogg: I don't see how RDF could be consistent without such entailment.

pchampin: I agree; we must have such entailment.
… But, the question is not just do we need semantics, but at which level is the support of semantics required?
… For me, dirLangString is a core part of the model and should be supported in semantics.

AndyS: I don't see how we can not do it; langString is in semantics so that it has one meaning which can't be redefined. We need to do the same for dirLangString.
… They're called out because of the different lexical space.

fsasaki: With regards to CR, developers need to be able to test and report back.

ora: I'm pretty sure we'll hear from implementers if we get this wrong.

AndyS: Peter's asking about semantics, not if it is in the data model.
… Tests will relate to the data model, not the semantics.

pchampin: I think it wasn't so much of a if we put it in semantics, but where.
… What makes sense is in the RDF semantics alongside strings, and langString.

<AZ> you should say "in the RDF entailment regime" rather than "in RDF semantics" because the later is whole semantics document

AndyS: The PR goes through everywhere where langString is mentioned, which is more about data types.

<niklasl> This is the PR w3c/rdf-semantics#64 ?

AndyS: dirLangString is beside langString.

<gb> Pull Request 64 Recognize rdf:dirLangString (by afs) [spec:enhancement]

ora: to the extent that merging a PR means anything, we could do that, but is that a strong enough statement?

james: As I recall, pfps sentiment was that he wanted an explicit statement.

AndyS: Specifically, to put it in the set of recognized data types.

pchampin: D-entailment defines a set where implementations may differ. "D" must be part of the entailment.
… It's not just saying what it means to recognize it, but that it MUST be recognized.

AndyS: The term is that of a "recognized data type". I asked why langString was there in the first place, and didn't get a good answer.

ora: Is saying it is recognized, or do we need to mention D-entailment specifically?
… "recognized" vs "required".

pchampin: Maybe the term is RDF Interpretations.

<pchampin> PROPOSAL: rdf:dirLangString's semantics is defined in RDF-Semantics, and MUST be recognized in RDF interpretations

<gkellogg> +1

<ora> +1

<niklasl> +1

<pchampin> +1

<fsasaki> +1

<eBremer> +1

<Dominik_T> +1

<james> +1

<tl> +1

<gtw> +1

<TallTed> +1

<Tpt> +1

<AndyS> +1

<doerthe> +1

<AZ> +1

<ktk> +1

<Souri> +1

<enrico> +1

RESOLUTION: rdf:dirLangString's semantics is defined in RDF-Semantics, and MUST be recognized in RDF interpretations

Nature / role of triples 2

ktk: We'll time-box this discussion.

enrico: I'm sympathetic with the idea, but wonder what the impact is everywhere?
… We can explain this, but when we say "triple", what do we mean?
… We say triple terms are asserted, but that's sloppy.
… We should not change the notion of "triple".

pchampin: I would say we're not changing the notion of a triple. The only use was to be asserted in a graph.
… That's my mental model; now we need to be clearer when we talk about a triple where it is used.
… When I reviewed RDF Concepts, it didn't seem to need much change.
… The text talks about asserted triples. We need to be clear that a triple in a graph means that the triple is asserted.
… There is a discussion of "appears in the graph", which could be moved to concepts later.
… Most of the text should still work.

enrico: I think some one needs to be responsible for having a pass over the text to put it in order.

tl: We need a name for the triple term we've introduced.
… As far as I say, the spec uses "triple" for the abstract thing, and "statement" for a triple which is asserted.

gkellogg: worries about breaking a concept by over-simplifying it.

pchampin: I agree, a statement is something you make by putting a triple in a graph, it's not the triple itself.

<AndyS> +1 to text outside definitions saying "triple" can be clear by context and that's OK.

pchampin: I don't think we change the terminology, it's more how we introduce the terminology.
… We present them as two different kinds of things; I propose we go back to the CG report and define a triple as something abstract.
… When the context is not enough, we can call it "asserted", when it's used as the object, it's a "triple term".
… Most of text should work as is.

<niklasl> +1 to pchampin

ora: I tend to agree. Are we fixing something that isn't really broken.

enrico: We keep using "triple term" and "triple", and this could fix it.
… The notion of a triple being in a graph is improved by the definition of "appears in a graph".

Souri: Having "asserted triple" and "triple term" makes things clearer.
… It's when a triple is in a graph, or appears as a term.
… If we say triple, it would normally mean "asserted triple". The "asserted" could be added to clarify.

<niklasl> +1 to Souri, this clarifies the two roles that a triple may have (two usages).

ora: You're suggesting we say "asserted triple" only when it's not clear from the context.

tl: I talked about asserted triples in the context of rdf:states. It has become common place in the WG, but I don't know that it is outside.
… Also, what happened to "statement", no one mentions the term any longer.

<pchampin> https://www.w3.org/TR/rdf12-concepts/#resources-and-statements "This statement corresponding to an RDF triple is known as an RDF statement."

<pchampin> The statement *corresponds* to the triple, it *is* not the triple.

ora: I think that "triple" has a lot of history; if you're worried that "asserted triple" is not in the spec, we can add it.

enrico: We had a long discussion about the use of "statement". We were convinced from looking at the spec that "statement" is for asserted triple.
… The current spec says that asserted triples are statements.

AndyS: I went through semantics, and the word "statement" comes up 12 times.

<AndyS> https://www.w3.org/TR/rdf12-semantics/#extensions

AndyS: The first mention is "RDF statements of the form ...", and shows a triple.

niklasl: I agree with Enrico. We will need some work on the text to clarify things.
… Regarding this issue (replacing triple term with triple), I think it makes the specs simpler.

<james> are statements also propositions?

niklasl: I'm not too worried about this. I think this formally makes things simpler.

ora: What are we fixing here? Most of the language is already clear.

pchampin: We need to fix the definnitions of "triple" and "triple terms" in concepts. Currently, they are two different things; they need to be the same thing with different roles.
… That just affects the text where they are defined. I volunteered to make a PR for this.

ora: do we need the resolution?

ktk: I think we need it to limit future discussion.

<gb> Issue 150 Nature / role of triples (by pchampin) [needs discussion]

<gkellogg> +1

<pchampin> PROPOSAL: rephrase the definitions of triple and triple terms in RDF concepts according to w3c/rdf-concepts#150

<fsasaki> +1

<niklasl> +1

<pchampin> +1

<ora> +1

<gkellogg> +1

<tl> +1

<doerthe> +1

<AndyS> +1

<TallTed> +1

<gtw> +1

<eBremer> +1

<Souri> +1

<james> +1

<ktk> +1

<enrico> +1

<AZ> +1

<Tpt> +1

james: As niklasl and enrico suggested, there's the concept of a "proposition". Whatever changes are made should recognize that.

niklasl: I think that's a good idea. Maybe we can coordinate on this.

pchampin: No mention to mention the notion of Proposition in concepts; it's already defined in a PR against semantics.

RESOLUTION: rephrase the definitions of triple and triple terms in RDF concepts according to w3c/rdf-concepts#150

<gb> Issue 150 Nature / role of triples (by pchampin) [needs discussion]

triple terms in subject position - issues with RDF/XML? 3

<niklasl> FYI there is WIP wording in https://github.com/niklasl/rdf-concepts/tree/rdf12proposition

pchampin: Last week, we decided to discuss triple terms in the subject position, but it's not just an RDF/XML issue.
… RDF/XML does point towards the problems of supporting this.

<niklasl> This is the general issue: w3c/rdf-concepts#138

<gb> Issue 138 Triple Terms in Subject Position (by rat10) [ms:CR] [spec:substantive]

pchampin: Another argument against triple terms in the subject position is that we are encouraging people to only use triple terms as the object of rdf:reifies. Saying otherwise is sending mixed signals.

tl: I was in favor of this, but I re-thought about it, as it's been in the abstract grammar.
… If I want to talk about the type, why can't I create a reifier for that? It seems like a hack.
… In RDF, we talk about something being asserted.
… If we treat triple terms differently, we look at only a single use case.
… There are things that have meaning, and it would be bad to dis-allow this.
… It would say that the whole construct is just about a single use case. Not saying something about it.
… I think we're re-making the "seminal mistake".
… If it doesn't work in some syntax, I don't think that's a problem.

niklasl: I disagree; primarily because I disagree with the use case, as the statement is about an abstract meaning, which requires a reifier.

<pchampin> For future reference, the seminal mistake: https://www.w3.org/2021/12/rdf-star.html#the-seminal-example

niklasl: Even in your explanation, you use a reifier.
… In RDF, we use names to describe things and we have a literal structural term which is abstract, denoting the value in the value space.
… Similarly, triple terms are structural, and are used to refer to the meaning of the triple, which is very abstract.
… I was more convinced that we should not allow this in concrete syntaxes because it is unsafe.
… Similar reasoning to why literals can't be subjects.
… I'm unconvinced by tl's argument. At the logical level, I see things differently.

Souri: I fully support those two points: only in the object position, and only with rdf:reifies.
… But, from a practical point of view, users do a lot of things. The minimum we give them would be preferable rather than introduce too much false choice.
… As an implementor, the implementation gets harder when we allow too much flexibility.
… I think this is good rationale to restrict to the object position.

<ora> STRAWPOLL: Disallow triple terms in the subject position

ora: I want to see if people are in favor of the restriction.

<ora> +1

<gkellogg> +1

<enrico> +0

<doerthe> -1

<tl> -1

<niklasl> +1

<AZ> -1

<pchampin> +0.5

<ktk> +1

<james> -1

<gtw> +1

<Souri> +1 for restriction

<Tpt> +0.5

<fsasaki> +1

<AndyS> +0.5

<TallTed> +0.7

<eBremer> +0

<james> does the restriction apply to reifier?

ora: we need to hear from people that don't support the position.

doerthe: For me, having it only in the object position seems totally random.
… I could live with it, as we allow it in the semantics, but I don't see the added value of the restriction.

ora: Would you be more in favor if it included a restriction on rdf:reifies?

doerthe: Not really.

james: My view of RDF is governed by a notion of symmetry; to restrict it is unnecessary, which will lead to problems.
… Similar to the restriction on rdf:reifies.
… Asymmetry in a statement is a bad idea.

fsasaki: I understood that doerthe could live with it. Maybe we can get an understanding of what people mean by "-1".

<Zakim> TallTed, you wanted to note that RDF has always been asymmetric -- literals have never been allowed in subject position except in Generalized RDF (and SPARQL)

TallTed: I don't understand how you can believe that RDF is symmetric, as it's never been so (literals).

james: This is a new term, which is in many ways similar to things which are nodes.
… If the group argued for treating it as a literal, it may make more sense.

<fsasaki> To add to my question: one could re-phrase in a binary way "could live with it" vs. "could not live it". But maybe not today since there is not enough time to think on it

TallTed: Your model is that it is symetric.

james: It's symmetric for nodes.

niklasl: In my mental model, literals are also nodes in the graph. IRIs and BNodes are nominal. I don't know what they denote until I learn more through triples.
… With literals and triple terms, as fixed point structural resources, they are wholly defined by their constituents.

<doerthe> what do they mean?

ora: We're out of time, and can't conclude the discussion.

ktk: It would be interesting to know what would be a common ground.

ora: If you voted -1, please send something to the mailing list so we can find common ground.

Summary of resolutions

  1. rdf:dirLangString's semantics is defined in RDF-Semantics, and MUST be recognized in RDF interpretations
  2. rephrase the definitions of triple and triple terms in RDF concepts according to w3c/rdf-concepts#150
Minutes manually created (not a transcript), formatted by scribe.perl version 242 (Fri Dec 20 18:32:17 2024 UTC).

Diagnostics

Succeeded: s/niklasl: In my model, IRIs and BNodes are nominal I don't know what they denote until I learn more./niklasl: In my mental model, literals are also nodes in the graph. IRIs and BNodes are nominal. I don't know what they denote until I learn more through triples.

Succeeded: s/... With literals and triple terms, they denote themselves and are wholly defined by their constituents./... With literals and triple terms, as fixed point structural resources, they are wholly defined by their constituents.

All speakers: AndyS, doerthe, enrico, fsasaki, gkellogg, james, ktk, niklasl, ora, pchampin, Souri, TallTed, tl

Active on IRC: AndyS, AZ, doerthe, Dominik_T, eBremer, enrico, fsasaki, gkellogg, gtw, james, ktk, niklasl, ora, pchampin, Souri, TallTed, tl, Tpt