Language data needs provenance, licensing, and human-readable context.

Configured source roles

• Manual core seed: initial project-authored coverage and schema validation.
• PanLex: multilingual translation candidates under its published license.
• Kaikki/Wiktionary and Wikidata Lexemes: dictionary and structured lexeme candidates.
• Tatoeba: sentence candidates that require per-item license tracking.
• UniMorph: morphology candidates under source-specific open-data terms.

Publication standard

A configured source is not automatically proof that every displayed item came from it. Public resources should identify the actual record provenance, preserve required attribution, and separate imported facts from project-authored explanations. Unverified generated sentences are not treated as authoritative examples.