Adapting to Triple Terms and Reifiers

[gkellogg]

The direction for RDF 1.2 now abandons quoted triples for triple terms which are intended to be used indirectly through a reifier that relates an identifier to the triple term through rdf:reifies. This likely affects how we might represent this in JSON-LD. Rather than using a recursive value on @id, we may instead do something using additional keywords:

The bob example might look more like the following:

{
  "@context": {
    "@base": "http://example.org/",
    "@vocab": "http://example.org/",
    "rdf": "http://www.w3.org/1999/02/22-rdf-syntax-ns#",
  },
  "rdf:reifies": {
    "@triple": {
      "@id": "bob",
      "age": 42
    }
  },
  "certainty": 0.8
}

This could be reduced by also introducing a @reifies keyword (at the expense of adding yet another keyword) as follows:

{
  "@context": {
    "@base": "http://example.org/",
    "@vocab": "http://example.org/",
    "rdf": "http://www.w3.org/1999/02/22-rdf-syntax-ns#",
  },
  "@reifies": {
    "@id": "bob",
    "age": 42
  },
  "certainty": 0.8
}

The annotation syntax could also allow for an @id entry for assigning the reifier identifier:

{
  "@context": {
    "@base": "http://example.org/",
    "@vocab": "http://example.org/"
  },
  "@id": "bob",
  "age": {
    "@value": 42,
    "@annotation": {
      "@id": "_:anno",
      "certainty": 0.8
    }
  }
}

[pchampin]

• Re. the representation of triple terms, I think I prefer the introduction of the @triple keyword over the overriding of the @id keyword. I believe that there are strong expectations that the value of @id is a string. Note also that ignoring the @triple keyword and its value would amount to replace triple-terms with blank nodes, which is a good fallback, IMO.

• Exposing the rdf:reifies predicate and its triple-term value in JSON-LD feels against the trend to "hide" them in Turtle, and to encourage well-formed uses of triple terms. So I lean towards having a separate @reifies keyword, even if the fact that it exposes the word "reifies" (contrarily to the syntactic sugar in Turtle) is not entirely satisfactory, IMO...

• +1 to allow @id in annotations to convey the reifier, that seems natural enough

[niklasl]

• Re. the representation of triple terms, I think I prefer the introduction of the @triple keyword over the overriding of the @id keyword. I believe that there are strong expectations that the value of @id is a string. Note also that ignoring the @triple keyword and its value would amount to replace triple-terms with blank nodes, which is a good fallback, IMO.

I definitely agree on not overriding @id (due to those probable strong expectations for it being a string).

• Exposing the rdf:reifies predicate and its triple-term value in JSON-LD feels against the trend to "hide" them in Turtle, and to encourage well-formed uses of triple terms. So I lean towards having a separate @reifies keyword, even if the fact that it exposes the word "reifies" (contrarily to the syntactic sugar in Turtle) is not entirely satisfactory, IMO...

I also agree on avoiding exposing rdf:reifies (assuming we can stay on track with the well-formed baseline, and have that as the basis for JSON-LD).

I also lean towards a @reifies keyword (it could be named differently; though I am personally comfortable with "reifies", if we stay with that nomenclature). More on that in a follow-up comment.

• +1 to allow @id in annotations to convey the reifier, that seems natural enough

Yes, definitely agree on that too; it comes naturally.

[niklasl]

Interestingly, @reifies might work for all of these:

<< <x> :relation <y> ~ <ref-1> >> :p :o .

<ref-2> rdf:reifies <<( <x> :relation <y> )>> .

<z> :says << <x> :relation <y> ~ <ref-3> >> .

Which in JSON-LD could be this:

[
  {
    "@id": "ref-1",
    "@reifies": {
      "@id": "x",
      "relation": {"@id": "y"}
    },
    "p": {"@id": "o"}
  },
  {
    "@id": "ref-2",
    "@reifies": {
      "@id": "x",
      "relation": {"@id": "y"}
    }
  },
  {
    "@id": "z",
    "says": {
      "@id": "ref-3",
      "@reifies": {
        "@id": "x",
        "relation": {"@id": "y"}
      }
    }
  }
]

One upshot is that, IMHO, JSON-LD isn't as "triple-oriented" as Turtle; and the above reifications are more like a "neutral" singleton objects if you will (or even "quoted" data, meant without any opacity connotation).

Note though that this relies on all data being within the well-formed constraints for triple terms, i.e. only <<( ... )>> as objects of rdf:reifies. I think that is a good thing for JSON-LD, but it needs to be kept in mind.

I even think it can be beneficial, and simple enough, to allow @reifies to contain multiple "triples". For example, this:

{
  "@id": "ref-4",
  "@reifies": {
    "@id": "x",
    "firstName": "X",
    "lastName": "XY"
  },
  "p": {"@id": "o"}
}

Would be, in Turtle:

<< <x> :firstName "X" ~ <ref-4> >> .
<< <x> :lastName "XY" ~ <ref-4> >> .
<ref-4> :p :o .