The project
Why equip a national language
A language of radio, trade and music used every day by tens of millions of speakers in Central Africa, Lingala is still almost absent from the systems that now decide who gets access to digital services.
A quiet exclusion
Without speech recognition or translation, a Lingala speaker stays dependent on French to reach a service, to follow medical instructions or to start an administrative procedure. The barrier is not technical, it is linguistic.
A gap in the data
Despite its cultural weight, Lingala remains under-represented in datasets and in artificial-intelligence models. That gap limits speech recognition, translation, assistants and digital learning in the language.
A question of control
The whole chain is run locally: collection, storage, processing, annotation, validation by three Lingala-speaking validators, training and reproducible evaluation. Every step is documented in a public report.
Mission
Maloba is implemented by Congo Digital Services, for the WYMEE / CDS / SACLI consortium, within the UNDP Local Language Accelerator programme, under the Ministry of Posts, Telecommunications and the Digital Economy.
The project builds, with the speakers, the data and the models Lingala was missing. Everything is published, measured and open to the community, with its results and its limits.
Team
A project lead, a language lead, a technical team, annotators and three Lingala-speaking validators. People are not named: no written agreement to publish names was obtained.
Timeline
- Early 2026Launch; agreements with SIL Congo and Radio Rurale
- First half of 2026Collection: Radio Rurale archives, Health Voice-a-thon
- July–August 2026First models: Llama, Whisper, Qwen2-VL
- 25 August 2026UNDP review
- September 2026Corpus rebuild and harmonisation; human validation; NLLB-Maloba-h1 translation model; MMS adapter
- 30 September 2026Results frozen: corpus V3.2 published, five models measured, evaluation and training reports released
Partners
Each with what they actually contribute. Logos go online only once the partner has agreed in writing.
Funding, steering, technical review
Funding partnerOversight of the project
OversightDocument collection: 22 digital documents (1 008 pages) and 110 manuscripts
Data partner54 h of digital archives; about 150 h of tape still to digitise
Signed agreementHealth Voice-a-thon, at the Grandes Endémies centre
ContributorsConsortium holding the contract, delivered by Congo Digital Services
Consortium