RDF 1.2: Reifying Propositions

Lotico event 10 June 2025

Niklas Lindström, National Library of Sweden | RDF & SPARQL WG

RDF 1.2

Lots of quality improvements in the ~20 documents involved (including SPARQL).
New rdf:JSON datatype.
Initial text direction for language literals (rdf:dirLangString).
Triples as objects ("new-style reification").

RDF: A Simple Triple

<Alice> :bought <LennyTheLion> .

Forms a logical proposition.

A Graph is a Set of Triples

When interpreted, these are the facts of the interpretation.

Challenging Use Cases

Expressing circumstantial information, like:

Source provenance
Qualification
Suggestions or beliefs

Traditional Solution

More granular remodelling. (Classic reification. The PROV ontology.)

Labelled Property Graphs

Nodes with attributes; edges between nodes.

Edges can have properties too.

(:ALICE) -[:BOUGHT {date: "2014-06"}]-> (:LENNY_THE_LION)

RDF-star

Motivations: alignment with LPGs (properties on edges) & statements about statements.

2014: Foundations of RDF* and SPARQL* (Olaf Hartig)

 << <Alice> :bought <LennyTheLion> >> :date "2014-06" .

2020: CG formed

2021: CG report

<Alice> :bought <LennyTheLion> {| :date "2014-06" |} .

Feedback (e-mail; github)
2022: WG formed
2026: RDF-star → RDF 1.2

LPGs from an RDF Point of View

Each edge, even those with the same label,
is unique!

Not encodable as sets of triples. Rather, "multisets" of triples.

(No model-theoretic semantics—no formal entailments).

(:ALICE)
  -[:BOUGHT {date: "2014-06"}]->
      (:LENNY_THE_LION)

(:ALICE)
  -[:BOUGHT {date: "2014-12"}]->
      (:LENNY_THE_LION)

<Alice> :bought <LennyTheLion>
      {| :date "2014-06" ;
         :seller <ToyStore> |} .

<Alice :bought <LennyTheLion>
      {| :date "2014-12" ;
         :seller <Market> |} .

Multisets~ Occurrences

<Alice> :bought <LennyTheLion>
      {| :date "2014-06" ;
         :seller <ToyStore> |} .
      {| :date "2014-12" ;
         :seller <Market> |} .

A Triple Is an Abstraction

<Florens> :connectedTo <Rome>
              {| a :Route ;
                :road <A1> ;
                :distance "276"^^:KM |}
              {| a :Route ;
                :road <RaccordoAutostradaleFirenze-Siena-A1> ;
                :distance "309"^^:KM |} .

The Map is Not the Territory

Source: OpenStreetMap & Rome.info

RDF 1.2: Triples as Objects

[] rdf:reifies <<( <Alice> :bought <LennyTheLion> )>> .

Triples are now allowed as objects of other triples. (I.e., as terms alongside IRIs, bnodes and literals.)

Triple terms denote abstract logical propositions.

(Cf. how literals denote resources in the value space of the datatype.)

RDF 1.2: Propositions & Reifiers

A triple denotes an rdfs:Proposition.

(Transparent semantics. Its sense, not its "encoding".)

If the triple is in the graph, it encodes
a fact in the interpretation.

The set of propositions:

IPR(I) = { RE(x, y, z)｜
    x ∈ IR, y ∈ IP, z ∈ IR }

The set of facts:

F(I) = { RE(x, y, z)｜
           <x, z> ∈ IEXT(y) }

A reifier is anything relating to a proposition using rdf:reifies.

The reifier can be a data token, an event, circumstance, situation, context, etc.

Other Designs Considered

Classic reification (rdf:Statement "tokens of triples"; now conceptually a subset of new-style reifiers)
N-ary relationships (more on that later)
Named graphs (more on that later)
Singleton properties
Triples as subjects
Named triples

RDF 1.2: Syntactic Sugar

Annotated Triples

<Alice> :bought <LennyTheLion>
      {| a rdfs:Resource |} .

Expands to:

<Alice> :bought <LennyTheLion> .

_:b1 a rdfs:Resource .
_:b1 rdf:reifies <<( <Alice> :bought <LennyTheLion> )>> .

A Reified Statement

Reifying Triples

<< :TheEarth :revolvesAround :TheSun >> a :Statement ;
  :source :Galileo ;
  :date 1610 .

Expands to:

_:b2 rdf:reifies <<( :TheEarth :revolvesAround :TheSun )>> .
_:b2 a :Statement .
_:b2 :source :Galileo .
_:b2 :date 1610 .

Naming Reifiers

<< <Alice> :bought <LennyTheLion> ~ <r1> >> .

<r1> a :Transaction .

<Alice> :bought <LennyTheLion> ~ <r2> .

<r2> a :Purchase .

Metaphor: Marginalia

Fact: The earth revolves around the sun. ^{[1] [2] [3]}

Galileo proposed that in 1610.
The church of old opposed that in 1633, on dogmatic grounds.
In a relativistic world, this depends on the frame of reference.

Encoded:

:TheEarth :revolvesAround :TheSun ~ <#1> ~ <#2> ~ <#3> .

<#1> a :Statement ; :source :Galileo ; :date 1610 .
<#2> a :DogmaticDenial ; :source :ChurchOfOld ; :date 1633 .
<#3> a :ReferenceFrame ; :gravitationalSystemBoundary "..." .

Kinds of Reifiers

The rdf:type of the reifier determines its nature (may be implied by predicates).

References (utterances, source provenance).
Circumstances (qualifying events or situations).

Instances seen in the wild:

rdf:Statement, prov:Influence, prov:Attribution, sdo:Role, bf:Publication, bf:Contribution...

Provenance

<Alice> :bought <LennyTheLion> {|
    a :Transaction ;
    :source :TXZ ;
    :tstamp "0x66e"
  |} .

Qualification

<Alice> :bought <LennyTheLion> {|
    a :Purchase ;
    :seller <ToyStore> ;
    :date "2024-06"
  |} .

Qualifications

<Alice> :bought <LennyTheLion> {|
    a :Purchase ;
    :seller <ToyStore> ;
    :date "2024-06"
  |} {|
    a :Purchase ;
    :seller <Market> ;
    :date "2024-12"
  |} .

Case: Wikidata

<Q59166653> a :Person ;
  :name "Malgorzata Domino"@en ;
  :alumniOf <Q681>
      {| :academicDegree <Q1862897> ;
         :startDate "2005-10-01" ;
         :endDate "2011-02-28" |}
      {| :academicDegree <Q950900> ;
         :startDate "2011-10-01" ;
         :endDate "2012-11-08" |}
      {| :academicDegree <Q752297> ;
         :startDate "2011-03-08" ;
         :endDate "2015-04-22" |} .

Continued: Wikidata

<Q59166653> sdo:alumniOf <Q681> ~ <9220f5f6> ~ <ff2f0926> ~ <c495f198> .

<9220f5f6> :academicDegree <Q1862897> ;
  :describedBy <https://pub.orcid.org/v3.0/0000-0001-9436-1074/education/6610842>
      ~ <379d8fe5> .
<ff2f0926> :academicDegree <Q950900> ;
  :describedBy <https://pub.orcid.org/v3.0/0000-0001-9436-1074/education/6610859>
      ~ <379d8fe5> .
<c495f198> :academicDegree <Q752297> ;
  :describedBy <https://pub.orcid.org/v3.0/0000-0001-9436-1074/education/6610892>
      ~ <379d8fe5> .

<379d8fe5> a :DataExtraction ;
  :statedIn <ORCIDPublicDataFile2021> ;
  :retrieved "+2021-11-05"^^xsd:date .

Varying Granularity

Use cases show reifiers of varying type, granularity and scope.

Examples include:

source tokens of granular provenance;
simplified relations sometimes needing more contextual detail.

Different modelling choices lead to different perspectives.
In many cases single propositions need to be reified.
But some may be of multiple propositions at once.

Name Forms

A simple example is the different models of person names.

Some define just name. Others, givenName and familyName. Etc.

A NameRecord may thus reify a single or multiple triples, depending on the modelling choice.

  <Alice> :name "Alice Liddell" ~ <nrx> .

  <nrx> a :NameRecord ;
    :date "1852" .

  <Alice> :givenName "Alice" ~ <nrx> ;
    :familyName "Liddell" ~ <nrx> .

  <nrx> a :NameRecord ;
    :date "1852" .

Case: Affiliated Author

<OnTheStateOfThings> a :Text ;
  :author <Alice> ;
  :dateCreated "2025-05".

<Alice> a :Person ;
  :affiliation <HardknockSchool> .

<OnTheStateOfThings> a :Text ;
  :contribution [ a :Contribution ;
      :agent <Alice> ;
      :affiliation <HardknockSchool> ;
      :date "2025-05" ] .

<Alice> a :Person .

<OnTheStateOfThings> a :Text ;
  :author <Alice> {| :affiliation <HardknockSchool> |} ;
  :dateCreated "2025-05".

<Alice> a :Person .

Continued: Annotated Affiliations

<< <Alice>
  :affiliation <HardknockSchool> ~ <ctb1> ,
    <TranscendentUniversity> ~ <ctb2> .

<Alice> a :Person ;
  :affiliation
    <TranscendentUniversity> ~ <ctb2> .

<OnTheStateOfThings> a :Text ;
  :author <Alice> ~ <ctb1> .

<OnTheChangeOfPatterns> a :Text ;
  :author <Alice> ~ <ctb2> ,
    <Lenny> ~ <ctb3> .

<ctb1> a :Contribution ;
  :date "2025-05".

<ctb2> a :Contribution ;
  :date "2036-10".

<ctb3> a :Contribution ;
  :role rel:win, rel:wst ;
  :date "2036-11".

Augmented Simplicity?

Previously, a bought relationship was qualified as a Purchase with a seller.

But the initial data might have been modelled differently, with a ("too") simple sold relation already present.

Compound Reifier

<Alice>
  :bought <LennyTheLion> ~ _:r1 .

<ToyStore>
  :sold <LennyTheLion> ~ _:r1 .

_:r1 a :Purchase ;
  :date "2024-06".

N-ary Relationship?

_:r1 a :Purchase ;
  :buyer <Alice> ;
  :seller <ToyStore> ;
  :item <LennyTheLion> ;
  :date "2024-06".

www.w3.org/TR/swbp-n-aryRelations

(OWL Doesn't Force This Choice)

:bought owl:propertyChainAxiom ([owl:inverseOf :buyer] :item) .
:sold owl:propertyChainAxiom ([owl:inverseOf :seller] :item) .

If you can afford to use OWL...

(There may be a path through the form of basic encoding. Watch this space.)

Datasets

There are no defined semantics for datasets.
The name does not (necessarily) denote the graph (set of triples).

A dataset is one default graph and a set of name, graph pairs (where the names are unique).

Named Graphs Do Not Nest

You can partition data into named graphs according to different schemes. Whether you bundle reifiers in the same named graph as the triples they reify, or partition them separately, is an orthogonal concern.

Looks Like a Graph?

<Alice> :bought
    <LennyTheLion> ~ _:r1 .

:ToyStore :sold
    <LennyTheLion> ~ _:r1 .

_:r1 a :Purchase ;
  :date "2024-06".

(Note that a graph is a set of triples.)

Alice Knows a Set?

<Alice> :knows <Lenny>, <Bob> .

No, Alice knows each person.

<Alice> :knows <Lenny> .
<Alice> :knows <Bob> .

Reifies Each Proposition

<Alice> :bought
    <LennyTheLion> ~ _:r1 .

:ToyStore :sold
    <LennyTheLion> ~ _:r1 .

_:r1 a :Purchase ;
  :date "2024-06".

(Note also that a set of one triple is not a proposition.)

Practical Graph Store Management

Different Use Cases

Named graphs: data management (graph stores, LDP).
Reifiers: qualification and detailed provenance.

Provenance on the inside (the interpretation) or the outside (the dataset).

More To This Story...

The triple structure is now recursively defined. (But not cyclic.)
rdf:reifies is but one relationship relating to propositions.

RDF 1.2

Triples as objects ⇒ referencing propositions <<( :s :p :o )>>.
Related to from reifying circumstances using rdf:reifies.
Syntactic sugar for annotating asserted triples and reifying claims:
- :s :p :o {| a :Q |} . | :s :p :o ~ :r .
- << s :p :o >> a :Q . | << s :p :o ~ :r >> .
Reifiers of different rdf:types, for qualification and provenance:
rdf:Statement, prov:Attribution, prov:Influence, sdo:Role, bf:Publication, bf:Contribution...

RDF 1.2: Reifying Propositions

RDF 1.2

RDF: A Simple Triple

A Graph is a Set of Triples

Challenging Use Cases

Traditional Solution

Labelled Property Graphs

RDF-star

LPGs from an RDF Point of View

Multisets~ Occurrences

A Triple Is an Abstraction

The Map is Not the Territory

Concepts & Semantics

RDF 1.2: Triples as Objects

RDF 1.2: Propositions & Reifiers

Other Designs Considered

RDF 1.2: Syntactic Sugar

Annotated Triples

A Reified Statement

Reifying Triples

Naming Reifiers

Metaphor: Marginalia

Modelling Patterns

Kinds of Reifiers

Provenance

Qualification

Qualifications

Case: Wikidata

Advanced Patterns

Continued: Wikidata

Varying Granularity

Name Forms

Case: Affiliated Author

Continued: Annotated Affiliations

Augmented Simplicity?

Compound Reifier

N-ary Relationship?

(OWL Doesn't Force This Choice)

Named Graphs?

Datasets

Named Graphs Do Not Nest

Looks Like a Graph?

Alice Knows a Set?

No, Alice knows each person.

Reifies Each Proposition

Practical Graph Store Management

Different Use Cases

More To This Story...

Summary

RDF 1.2