Annotation properties declared with xsd:string/xsd:dateTime range but asserted with language-tagged values — crashes Protégé editor, undetectable by ELK

[jclerman]

Symptom

Editing an RID:Synonym value in Protégé Desktop's annotation editor throws IllegalStateException: Attempt to mutate in notification and crashes the edit. This matches a previously-filed upstream Protégé bug (protegeproject/protege#1111 — still open/unfixed there).

Root cause

RadLex.owl declares several annotation properties with AnnotationPropertyRange(... xsd:string) (or, in one case, xsd:dateTime), but many of their actual asserted values are language-tagged literals (e.g. "inferior lingular artery"@en). Per OWL 2, the correct datatype for language-tagged annotation values is rdf:PlainLiteral, not xsd:string (and not rdf:langString, which isn't part of the OWL 2 datatype map). This mismatch is what triggers the Protégé crash: Protégé sets an "overriding datatype" from the declared range and then chokes when the actual value carries a language tag.

This class of error is not detectable by a DL reasoner (ELK or otherwise) — OWL 2 Direct Semantics explicitly excludes annotations ("annotations are treated as not being present"). Only a SPARQL-based check (e.g. robot report/robot query) can catch it.

Affected properties

Verified via robot query against RadLex.owl directly:

Property	Total literal values	Language-tagged
RID:Synonym	24,013	23,763
RID:Acronym	139	107
RID:Unsanctioned_Term	93	87
RID:Misspelling_of_term	36	35
RID:Created	989	16
RID:Radlex_version_of_class_change	2,435	8 (all "4.1"@en)
RID:Source	3,116	7
RID:Definition	3,840	4

RID:Created's case is worth flagging separately: its range is xsd:dateTime, not xsd:string, so a language tag there isn't a range-design gap, it's a straightforward tagging mistake (dates aren't language-dependent).

Same declared xsd:string range, but no language-tagged values found: RID:Comment (1,931 total), RID:Preferred_Name_for_Obsolete (940 total).

Related but distinct finding: RID:Replaced_by

RID:Replaced_by is declared with range xsd:string, but of its ~442 assertions, only 4 are actual string literals — the other ~438 are IRI references to other RID terms, e.g.:

AnnotationAssertion(RID:Replaced_by RID:RID10009 RID:RID9510)

This is a different (arguably more severe) range violation: most values aren't literals at all. Worth tracking, but should be fixed independently of the language-tag issue above.

Proposed fix

1. Add a ROBOT SPARQL check (tests/sparql/robot/, registered in tests/robot-report-profile.txt, following the existing ?entity ?property ?value convention — see docs/TESTING.md) that flags any annotation property whose declared range disallows language tags but which has language-tagged assertion values, so this can't silently regress.
2. Fix the existing data, either by:
   • broadening the affected AnnotationPropertyRange declarations to rdf:PlainLiteral, or
   • stripping the language tags from the offending values (e.g. RID:Created, where the tag is simply wrong).
3. Separately, investigate and fix RID:Replaced_by's range/value-type mismatch (IRI values vs. declared xsd:string range).

[hoodcm]

I agree. We keep the language tags on names and definitions, and change their declared type to allow one. We remove the tags from IDs, dates, version codes, and source names. After PR #9 every Replaced_by value is a link, so we can drop its text type. I can put the fix and the check in one PR.