Answers are grounded in this paper and cite their sources.
Dagbanli is a Gur language of northern Ghana with millions of speakers and almost no everyday digital tools. We built the Dagbanli Dictionary (https://dagbanli.info) as a client of Wikidata's lexicographical data rather than as a database of its own: every word entry is a Wikidata Lexeme, and a correction made on Wikidata reaches the dictionary at the next harvest. This paper describes how the application works and what running it has taught us about the data. A chunked, resumable harvester on a serverless worker copies about 11,300 Dagbanli Lexemes into a small language pack, which a progressive web app stores on the device so that search, browsing and audio for saved words work without a connection. The same snapshot feeds one static, linked page per Lexeme for search engines. We then link 39,472 recorded sentences from Mozilla Common Voice and a University of Ghana speech dataset to the Lexemes they contain, giving over 4,000 entries spoken usage examples. Matching every sentence against every Lexeme also works as an audit of the Wikidata data: two thirds of the Lexemes have only one Form, 55% of matched tokens are ambiguous between Lexemes, and some of the most frequent words in the corpus have no Lexeme or Form at all. We report these figures, the engineering limits we ran into, and what we would ask of Wikidata tooling for languages like ours.
No takes yet. Share an insight, caveat, or question.
Abdulai et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: