Locations
Modular Approach for Governments
No government needs to commit to all five pillars at once. Every commitment sits within the same shared framework — global consistency, local flexibility. Click a value chain stage to see how a country can enter the ELM framework there.
Countries entering at Stage 1 typically have active language communities and recording capacity but lack governance frameworks. The priority commitment is FPIC protocols and establishing community data cooperatives. Releasing government-held language data can create an immediate, transformative national asset.
Suggested first steps
- Adopt FPIC protocols as the national standard for all publicly funded collection
- Conduct a national audit of government-held language data assets and publish a release roadmap
- Establish at least one pilot community data cooperative with legal recognition
- Update procurement to require multilingual capability declarations from AI vendors
Countries entering at Stage 2 have existing datasets but lack governance and shared infrastructure. The priority is curation standards, provenance requirements, and designating a national node in the open model repository.
Suggested first steps:
- Adopt common metadata and interoperability standards across all agencies
- Mandate community review rights for all curated language corpora
- Designate a national open model repository node with committed hosting and compute
- Fund annotation capacity with documented fair compensation standards
Countries entering at Stage 3 have curated datasets and need compute access and shared model infrastructure. The priority is allocating a compute budget for fine-tuning and securing language AI talent into civil service.
Suggested first steps:
- Allocate a compute budget for fine-tuning foundation models on national languages
- Hire language AI practitioners into key ministries with career progression structures
- Fund a graduate programme with a low-resource language AI specialisation
- Mandate open-source evaluation frameworks for all publicly funded AI tools
Countries entering at Stage 4 have working models and need a deployment pathway. The priority is commissioning a named use case with a ministry owner, an operational budget line, and open APIs for local innovators to build on.
Suggested first steps:
- Commission 1–2 high-impact use cases with a named ministry owner and operational budget
- Create a procurement set-aside or challenge fund for local-language AI products
- Publish open APIs and language datasets for local developers to build on
- Train community operators to use and provide structured feedback on deployed tools
Countries entering at Stage 5 have deployed services and need accountability frameworks. The priority is establishing a documented baseline, committing to annual public reporting, and funding independent review.
Suggested first steps
- Define and publish a KPI measurement framework covering uptake, coverage, equity and value-return
- Establish a documented baseline against all KPIs
- Commit to publishing an annual ELM progress report publicly
- Fund an independent review mechanism with a mandate to publish all findings
What Stays Constant Regardless of Entry Point
The Local Language Accelerator (LLA) is UNDP’s global programme supporting countries to develop inclusive language AI by strengthening the digitalisation of low-resource languages. Working with governments, universities, technical partners, startups, and communities across 11 countries in Asia, Africa, Latin America, Europe, and the Arab region, the LLA supports the creation of language datasets, local AI ecosystems, and public-interest applications. Through two years of implementation, recurring bottlenecks have emerged across very different country contexts — in data collection and stewardship, model development, infrastructure, deployment, institutional uptake, financing, governance, and reuse. Here is what we are seeing in two active implementations.
Speech and multimodal AI
ASR and TTS are equal priorities to text across all pillars. Voice-first AI is an equal priority from the outset.
Community governance
FPIC, benefit-sharing, and community data rights are non-negotiable at every stage.
International visibility
All participating governments gain access to international forums and peer learning from day one.
Open to all initiatives
Ground-level initiatives independent of UNDP are full ELM participants. ELM welcomes all ground-level initiatives, independent of UNDP.