- From: Peter F. Patel-Schneider <pfpschneider@gmail.com>
- Date: Wed, 10 Jul 2024 08:02:05 -0400
- To: Franconi Enrico <franconi@inf.unibz.it>
- Cc: RDF-star Working Group <public-rdf-star-wg@w3.org>
The proposal is not about changing the syntax of RDF but instead about fundamentally changing the nature of RDF graphs (which will, in turn, require a new syntax, but that's not the important part). Here is a quick stab at the required definition, done for generalized RDF-star graphs as that is the simplest version to do. Generalized RDF-star triples is the smallest set of triples of the form subject, predicate, object where a subject, predicate, or object is an IRI, a blank node, or a literal, optionally plus a generalized RDF-star triple. The optional triple might have to instead be a set of triples. A generalized RDF-star graph is a set of generalized RDF-star triples. peter On 7/10/24 04:21, Franconi Enrico wrote: > If I understand you well, you propose that RDF has the following syntax: > > |graph ::= triple* triple ::= subject predicate object subject ::= > NoLiteralTerm predicate ::= iri object ::= term NoLiteralAtomicTerm ::= iri | > BlankNode atomicTerm ::= NoLiteralAtomicTerm | literal NoLiteralTerm ::= > NoLiteralAtomicTerm | tripleTerm term ::= NoLiteralTerm | literal tripleTerm ::= |NoLiteralAtomicTerm triple > > > Am I correct? > —e. > >> On 9 Jul 2024, at 23:09, Peter F. Patel-Schneider <pfpschneider@gmail.com> >> wrote: >> >> The point of the proposal is to require that (some) nodes in RDF graphs are >> of the form IRI x triple or BNOde x triple. >> >> Yes, Turtle should be as compact as possible but it is not the thing that >> most users should see why they view RDF graphs. >> >> peter >> >> >> On 7/9/24 15:12, Niklas Lindström wrote: >>> Hi Peter, >>> I agree with your initial reply to Thomas. And I agree that your >>> (strawman) proposal here probably won't hold up. >>> This form looks like named triples (RDFn). I don't think it would >>> work. unless RDF graphs are redefined to be `(triple* | (name, >>> triple))*`. It also imposes some troubling limitations, such as the >>> impossibility of referring to the relationship between the "name" and >>> the triple (not only in other triple terms, which may be an edge case; >>> but, crucially, in vocabulary design; which is needed, as I show in >>> [4]). And it may lead to the named graphs problem all over again -- >>> what do the names mean in relation to their triple(s)? And indeed, >>> naming multiple triples like that appears very problematic. (Problems >>> which the explicit reification of multiple triples by linking them >>> does not suffer from.) >>> I suspect that some ongoing confusion is a residual effect of the >>> original proposal to add triples as subjects. Adding triples as >>> subjects was *not* reification "done right". It was, IMO, reification >>> done more wrong. Triples as subjects didn't work at all for real world >>> LPG uses of many-to-one. With some hyperbole, it was akin to using >>> literals as subjects naively, with `"20" :currency :USD` to solve the >>> problem of values with units (structured values), but "with some >>> limitations" (saying that the integer 20 is in US-dollar currency in >>> the entire model). But to be more fair, the RDF-star error was far >>> more subtle. >>> We've finally all but expunged this error. Now, triples as *objects* >>> (triple terms) of an appropriate relation on the other hand, have >>> shown promise of some really powerful benefits. >>> There is some residue left though, one being some insistence on >>> allowing it even in non-generalized abstract syntax. But another >>> problem is sticking to this syntax: >>> << <Alice> :bought <SomeComputer> >> :date "2014" . >>> Which is now a shorthand for: >>> _:r1 rdf:reifies <<( <Alice> :bought <SomeComputer> )>> . >>> _:r1 :date "2024" . >>> _:r1 :cost 20 . >>> _:r1 :currency :USD . >>> and totally fails to make this: >>> _:r1 rdf:reifies <<( <Alice> :shoppedAt <ComputerStore> )>> . >>> _:r1 rdf:reifies <<( <Alice> :bought <SomeComputer> )>> . >>> _:r1 :date "2024" . >>> _:r1 :cost 20 . >>> _:r1 :currency :USD . >>> shorten to anything like Turtle, or even legible at all: >>> << _:r1 | <Alice> :bought <SomeComputer> >> :date "2014" . >>> << _:r1 | <Alice> :shoppedAt <ComputerStore> >> :cost 20 . >>> _:r1 :currency :USD . >>> (In case anyone wants to object to my model design choice here ("use >>> `_:r1 :seller <ComputerStore>`"!), please read my follow-up to Thomas >>> [1].) >>> If we're *serious* about the minimal baseline [2], with `rdf:reifies` >>> working *equally* well for many-to-one and many-to-many (proper N-ary >>> relationships, relators, general reification), we need to revisit that >>> in earnest, as I wrote in [3]. >>> That proposal could shorten the above--if the purchase alluded to is >>> not also true--along the lines of: >>> <Alice> << :bought <SomeComputer> >> ^{_:r1} ; >>> << :shoppedAt <ComputerStore> >> ^{_:r1} . >>> _:r1 :cost 20 ; >>> :currency :USD ; >>> :date "2014" . >>> Which might not be *beautiful* (and could be tinkered with some more), >>> but is at least more "Turtle" (once you get used to reading the quotes >>> as being for predicate+object). For the possibly (much) more common >>> case, remove the quotes to have the regular assertions with >>> annotations: >>> <Alice> :bought <SomeComputer> ^{_:r1} ; >>> :shoppedAt <ComputerStore> ^{_:r1} . >>> _:r1 :cost 20 ; >>> :currency :USD ; >>> :date "2014" . >>> This "extra resource" is *crucial*. And it isn't anything mysterious. >>> Here, it should be typed: `_:r1 a :Purchase`. In other cases, we have >>> Marriages, Publications, Pipe connections, or good old Statements, >>> Snaks, Observations, Utterances, Data Sources or Ingests, or whatever >>> the nature is of the reifying circumstance of one or more abstract >>> relationships. Regardless of their type, they relate to these >>> relationships, uniformly, with `rdf:reifies`. And this is what we >>> should convey. >>> I very much value what you wrote regarding "the limited sensory and >>> cognitive capabilities of humans". Even if my proposed form here is >>> deemed unsatisfactory, this is the condition for which I think Turtle >>> should cater. Making wikidata more readable is of great interest to me >>> too [4]. Again, the detailed polish has to wait until we have a solid, >>> agreed upon baseline. (There is some interaction though, unless >>> someone can transmit the pure qualia of the RDF abstract syntax...) >>> Best regards, >>> Niklas >>> [1]: >>> <https://lists.w3.org/Archives/Public/public-rdf-star-wg/2024Jul/0038.html> >>> [2]: <https://github.com/w3c/rdf-star-wg/wiki/RDF-star-%22minimal-baseline%22> >>> [3]: >>> <https://lists.w3.org/Archives/Public/public-rdf-star-wg/2024Jul/0011.html> >>> [4]: <https://github.com/Kungbib/wikidatalab/> >>> On Tue, Jul 9, 2024 at 5:02 PM Peter F. Patel-Schneider >>> <pfpschneider@gmail.com> wrote: >>>> >>>> Here is a proposal that I don't think will go anywhere, and I might not >>>> totally believe in, but does connect to the working group's activities. >>>> >>>> THESIS: embedded triples are not a good solution to the use cases of the >>>> working group >>>> >>>> EVIDENCE: >>>> >>>> The use cases of the working group do not use embedded triples directly but >>>> instead require a separate resource that is connected to a triple. These >>>> separate resources are needed because the information about an embedded triple >>>> from one use of it has to be separated from the information from other uses. >>>> Otherwise there is a mix-and-match problem, as shown in representing >>>> provenance where source from one provenance cannot be combined with time or >>>> access from another. This problem affects the "seminal example", all kinds of >>>> provenance, and nearly all uses of embedded triples in the enoding of n-ary >>>> predicates. The need for this extra resource and new linking predicate add to >>>> the complexity of just about any use of embedded triples in RDF and require >>>> extra shorthands in Turtle to partly hide this complexity from users. >>>> >>>> SOLUTION: >>>> >>>> The solution is to do away with the uniqueness of embedded triples and base >>>> the extension of RDF proposed by the working group instead on non-unique >>>> occurrences of triples. If we leave the proposed syntax alone, we get an >>>> extension of RDF where >>>> << :a :b :c >> :d :e , :f :g . >>>> << :a :b :c >> :h :i , :j :k . >>>> does *not* entail >>>> << :a :b :c >> :d :e , :h :i . >>>> >>>> There are problems with this version of occurrences of triples. Without some >>>> way of referencing a particular occurrence of a triple it is not possible to >>>> represent the above graphs in N-triples and all information about the >>>> occurence has to use a shorthand syntax in Turtle, making what used to be a >>>> convenience a necessity. The solution to this problem is to in effect give >>>> these resources an identifier, so that a particular occurrence of a triple is >>>> no longer "anonymous" and can be referred to. >>>> >>>> The way to do this is to allow IRIs and blank nodes in RDF to also be a triple >>>> occurence, with syntax something like (this syntax probably not good at all >>>> but you should get the idea) >>>> <:x< :a :b :c >> :d :e . >>>> <_:x< :a :b :c >> :d :e . >>>> in both N-triples and Turtle. This is a varation of a recent syntax proposal >>>> but is not just syntax and instead is the extension to the RDF data model to >>>> support quoted triples. >>>> >>>> A big problem (and one reason that I don't totally believe this proposal) is >>>> using the same IRI or blank node for multiple triple occurrences as in >>>> <:x< :a :b :c >> :d :e . >>>> <:x< :f :g :h >> :d :e . >>>> has to be handled by either forbidding it or allowing a node to have multiple >>>> triple occurrences. >>>> >>>> peter >>>> >> >
Received on Wednesday, 10 July 2024 12:02:12 UTC