skos-integrity

name

skos-integrity

description

Check a SKOS vocabulary against the SKOS Reference (W3C Rec. 18 Aug 2009) integrity conditions using this project's own engines — the SHACL validator (98 pass, 0 fail on the W3C core suite), SPARQL 1.1 with property paths (631 pass, 0 fail), and the generic OWL 2 RL closure. Use when asked to validate a SKOS taxonomy or thesaurus, when auditing skos:prefLabel/altLabel/hiddenLabel usage, skos:broader/narrower/related consistency, or mapping-property (exactMatch/broadMatch/narrowMatch/relatedMatch) clashes, or when the user references SKOS "integrity conditions" / S-numbers from the Reference. A prose skill rather than an F* module by design — the checks mix shapes, SPARQL, and axiom data rather than new engine logic.

SKOS integrity checking with Factoidal's own engines

The SKOS Reference (W3C Recommendation, 18 August 2009) designates six numbered paragraphs as formal Integrity Conditions — normative constraints a conformant SKOS graph is expected to satisfy, kept separate from the vocabulary's own OWL axioms. This skill checks all six with artifacts that already exist in this repository: no new F* module, no SKOS-specific engine code. Everything here is prose, SHACL shapes, SPARQL queries, and RDF data — which is why it ships as a skill instead of a .fst file.

Why these six conditions aren't OWL axioms

The Reference's own machine-readable vocabulary (skos.rdf) asserts owl:disjointWith between the three core classes (skos:ConceptScheme, skos:Concept, skos:Collection — this is S9 and S37) but does not assert owl:disjointWith between any of the properties the other four conditions talk about (skos:prefLabel/altLabel/hiddenLabel, skos:related/ broaderTransitive, skos:exactMatch/broadMatch/relatedMatch). That is deliberate, not an oversight: skos:broadMatch is a sub-property of both skos:mappingRelation and skos:broader, so an OWL axiom declaring skos:exactMatch owl:disjointWith skos:broadMatch would make a large fraction of real-world SKOS graphs OWL-inconsistent rather than merely non-conformant. The Reference calls S13/S14/S27/S46 "integrity conditions" rather than axioms for exactly this reason, and says implementations should check them separately. That separation is why SHACL and SPARQL — not the OWL-RL closure — do the checking here.

The six conditions

S#	Section	Statement	Checked by
S9	4.4	skos:ConceptScheme is disjoint with skos:Concept.	SHACL (sh:not/sh:class), SPARQL
S13	5.4	skos:prefLabel, skos:altLabel and skos:hiddenLabel are pairwise disjoint properties.	SHACL (sh:disjoint), SPARQL
S14	5.4	A resource has no more than one value of skos:prefLabel per language tag.	SHACL (sh:uniqueLang), SPARQL
S27	8.4	skos:related is disjoint with the property skos:broaderTransitive.	SPARQL only (needs closure or a property path)
S37	9.4	skos:Collection is disjoint with each of skos:Concept and skos:ConceptScheme.	SHACL (sh:not/sh:class), SPARQL
S46	10.4	skos:exactMatch is disjoint with each of the properties skos:broadMatch and skos:relatedMatch.	SHACL (sh:disjoint), SPARQL

Labelled total: 6 conditions checked. 5 catchable by SHACL (S9, S13, S14, S37, S46). All 6 catchable by SPARQL. 1 (S27) needs either a property-path traversal over skos:broader+ or an OWL-RL closure pass, because skos:broaderTransitive is a derived relation that is almost never asserted directly in real data.

Every command below was actually run against sample-vocab.ttl (valid — every check reports no violation) and sample-vocab-broken.ttl (the same vocabulary with exactly one planted violation per condition, commented VIOLATION-S<n> at the offending triple). Output below is abridged but not edited for content.

Artifacts in this directory

• skos-shapes.ttl — SHACL shapes for S9, S13, S14, S37, S46 (the five conditions expressible without transitive closure).
• queries/ — one SPARQL file per condition, including both the property-path and the closure-dependent variant of S27.
• skos-axioms.ttl — the SKOS OWL axioms (owl:TransitiveProperty, owl:SymmetricProperty, owl:inverseOf, rdfs:subPropertyOf ladders) transcribed from skos.rdf as plain data, so this project's existing generic OWL 2 RL closure derives SKOS entailments with no SKOS-specific engine code.
• sample-vocab.ttl / sample-vocab-broken.ttl — the valid/broken fixture pair every command below runs against.

Commands

SHACL: S9, S13, S14, S37, S46

bin/linux-x86_64/factoidal validate \
  --shapes skills/skos-integrity/skos-shapes.ttl \
  skills/skos-integrity/sample-vocab.ttl

sh:conforms true

bin/linux-x86_64/factoidal validate \
  --shapes skills/skos-integrity/skos-shapes.ttl \
  skills/skos-integrity/sample-vocab-broken.ttl

sh:conforms false
  [Violation] focus=...#Cat path=...#exactMatch value=...#Cat source=_:d0__anon7
  [Violation] focus=...#PetsCollection path=(none) value=...#PetsCollection source=...#CollectionNotConceptShape
  [Violation] focus=...#scheme path=(none) value=...#scheme source=...#ConceptSchemeNotConceptShape
  [Violation] focus=...#Animal path=...#prefLabel source=_:d0__anon3
  [Violation] focus=...#Dog path=...#prefLabel value="Dog"@en source=_:d0__anon0

Five violations, one per planted SHACL-checkable condition: S46 (Cat's exactMatch/broadMatch clash), S37 (PetsCollection also typed Concept), S9 (scheme also typed Concept), S14 (Animal has two en prefLabels), S13 (Dog's prefLabel/altLabel share the value "Dog"@en). The planted S27 violation (Dog related and broader the same concept) does not appear here — it needs the SPARQL check below, as designed.

SPARQL: all six, direct or property-path

bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab.ttl \
  --query skills/skos-integrity/queries/s09-conceptscheme-concept-disjoint.rq
# false
bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab-broken.ttl \
  --query skills/skos-integrity/queries/s09-conceptscheme-concept-disjoint.rq
# true

bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab-broken.ttl \
  --query skills/skos-integrity/queries/s13-label-disjointness.rq -o csv

resource,value,p1,p2
http://example.org/skos-integrity/animals#Dog,Dog,http://www.w3.org/2004/02/skos/core#prefLabel,http://www.w3.org/2004/02/skos/core#altLabel

bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab-broken.ttl \
  --query skills/skos-integrity/queries/s14-preflabel-unique-lang.rq -o csv

resource,lang,count
http://example.org/skos-integrity/animals#Animal,en,2

bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab.ttl \
  --query skills/skos-integrity/queries/s27-related-broadertransitive-property-path.rq
# false
bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab-broken.ttl \
  --query skills/skos-integrity/queries/s27-related-broadertransitive-property-path.rq
# true

bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab.ttl \
  --query skills/skos-integrity/queries/s37-collection-disjoint.rq
# false
bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab-broken.ttl \
  --query skills/skos-integrity/queries/s37-collection-disjoint.rq
# true

bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab-broken.ttl \
  --query skills/skos-integrity/queries/s46-exactmatch-mapping-clash.rq -o csv

a,b
http://example.org/skos-integrity/animals#Cat,http://example.org/skos-integrity/external#Cat

All six ran clean (no rows / false) against sample-vocab.ttl and caught exactly the planted violation against sample-vocab-broken.ttl — no false positives, no false negatives.

S27, the materialized-closure variant

The property-path variant above needs no closure. The literal reading of S27 — skos:related disjoint with skos:broaderTransitive — only makes sense once skos:broaderTransitive has actually been derived. Load skos-axioms.ttl alongside the data and turn on the OWL-RL closure:

bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab.ttl \
  --data skills/skos-integrity/skos-axioms.ttl --entail owl-rl \
  --query skills/skos-integrity/queries/s27-related-broadertransitive-materialized.rq
# false
bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab-broken.ttl \
  --data skills/skos-integrity/skos-axioms.ttl --entail owl-rl \
  --query skills/skos-integrity/queries/s27-related-broadertransitive-materialized.rq
# true

Confirm the closure is doing real work, not coincidentally matching: without the axioms and --entail, zero skos:broaderTransitive triples exist in the data (skos:broaderTransitive is never asserted in the fixtures); with them, the closure derives it including the two-hop case (Dog broader Mammal broader Animal):

bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab.ttl \
  --data skills/skos-integrity/skos-axioms.ttl --entail owl-rl \
  -e 'PREFIX skos: <http://www.w3.org/2004/02/skos/core#> SELECT ?a ?b WHERE { ?a skos:broaderTransitive ?b }' -o csv

a,b
http://example.org/skos-integrity/animals#Mammal,http://example.org/skos-integrity/animals#Animal
http://example.org/skos-integrity/animals#Dog,http://example.org/skos-integrity/animals#Mammal
http://example.org/skos-integrity/animals#Dog,http://example.org/skos-integrity/animals#Animal
http://example.org/skos-integrity/animals#Cat,http://example.org/skos-integrity/animals#Mammal
http://example.org/skos-integrity/animals#Cat,http://example.org/skos-integrity/animals#Animal

bin/linux-x86_64/factoidal query --data skills/skos-integrity/sample-vocab.ttl \
  -e 'PREFIX skos: <http://www.w3.org/2004/02/skos/core#> SELECT ?a ?b WHERE { ?a skos:broaderTransitive ?b }' -o csv

a,b

The derivation runs through two generic OWL-RL rules, neither of which knows anything about SKOS: RDF.Graph.Executable.fst's rdfs_rule_subPropertyOf (rule "rdfs7": skos:broader rdfs:subPropertyOf skos:broaderTransitive plus the asserted broader edges yields broaderTransitive edges) followed by the transitive-property rule ("prp-trp": skos:broaderTransitive rdf:type owl:TransitiveProperty plus the two-hop Dog broaderTransitive Mammal / Mammal broaderTransitive Animal yields Dog broaderTransitive Animal). skos-axioms.ttl also declares skos:related owl:SymmetricProperty (rule "prp-symp") and the full mapping-property ladder (owl:inverseOf, rdfs:subPropertyOf) for the same reason — none of it is exercised by the six integrity checks directly, but it is what lets any other SPARQL query over this data see real SKOS entailments (e.g. skos:narrower derived from an asserted skos:broader) without hand-writing SKOS-aware code anywhere in the engine.

Engine notes

No engine gaps were found while building this skill. Everything needed already existed: sh:disjoint, sh:uniqueLang, sh:not + sh:class, sh:targetClass, sh:targetSubjectsOf on the SHACL side; property paths (+, ^, alternation) and --entail owl-rl with generic rdfs:subPropertyOf / owl:TransitiveProperty / owl:SymmetricProperty / owl:inverseOf rules on the SPARQL/OWL-RL side. If a future SKOS check needs something not on this list (for example, owl:AllDisjointProperties as a first-class SHACL or OWL-RL construct, which would let S13/S27/S46 be expressed as data-level axioms instead of external checks), that is an engine gap worth opening an issue for — it is not a gap this skill's checks depend on today.

Future work: skosdex as the scale corpus

github.com/danbri/skosdex is the owner's running attempt to aggregate "all the SKOS in the world" — it tracks an arbitrary number of third-party SKOS vocabularies one folder each, fetches and caches each from its canonical source, normalizes to N-Quads, and canonicalizes with RDFC-1.0. The live instance (skosdex.fly.dev) currently embeds roughly 26,000 concepts across multiple bundled, openly-licensed vocabularies.

That corpus is a natural next step for this skill, in two directions:

• Perf: the six checks here are all single-pass SHACL/SPARQL over property paths or a single OWL-RL closure — skosdex's ~26k concepts (and whatever larger pull a future fetch does) is a realistic size class to benchmark against, alongside the existing UK Parliament corpus (see the perf-benchmarking skill).
• Integrity sweep: running skos-shapes.ttl and every query in queries/ across every vocabulary skosdex bundles would surface which real-world published SKOS vocabularies actually violate S9, S13, S14, S27, S37, S46 — none of the small literature on SKOS integrity checking has done this at any scale, as far as this skill's research turned up.

This is future work, not implemented here: skosdex's data was not cloned or fetched into this repository (only its README, via WebFetch), per the task's scope.

What this skill does not cover

• SKOS-XL (the label-as-resource extension) — not modeled by any Factoidal engine yet.
• The non-integrity-condition parts of the SKOS Reference (concept scheme membership recommendations, notation datatypes, collection ordering) — out of scope for an integrity skill.
• Enforcing S13/S27/S46 as hard OWL-RL inconsistencies. As explained above, the Reference deliberately does not do this either.