5 Pillars

    Pillar 1

    Data and Community Sovereignty

    Government commits to

    • Adopt FPIC protocols as the national standard for all publicly funded language data collection
    • Conduct a national audit of government-held language data assets and publish a release roadmap
    • Release public sector language datasets under appropriate licences, with a public report on what was released and why
    • Support establishment of community data cooperatives with legal recognition, benefit-sharing rules, and contributor compensation standards
    • Update public procurement standards to require multilingual capability declarations and independent evaluation benchmarks

    What success looks like

    • Mandates harmonised: a single cross-ministry language data policy replaces fragmented ministerial siloes, with a named inter-ministerial owner
    • Data released: government-held language corpora published with full provenance documentation and an annual public transparency report
    • Community ownership real: at least one legally recognised data cooperative is operational, with documented benefit-sharing payments
    • Procurement reformed: procurement uses multilingual evaluation benchmarks, creating a domestic market for local-language AI products
    Pillar 2

    AI Infrastructure and Model Commons

    Government commits to

    • Designate or fund a national node in the open language model repository, committing hosting, compute budget and public access terms
    • Adopt common metadata and interoperability standards for language datasets and models across all government agencies
    • Allocate a compute budget for fine-tuning foundation models on national languages, with a target and published scaling plan
    • Mandate that all publicly funded AI tools use open-source evaluation frameworks and publish benchmark results in a shared accessible registry

    What success looks like

    • Shared infrastructure live: a national open model repository is operational, with multiple agencies actively contributing and reusing — no agency commissioning equivalent models separately
    • Compute accessible: local research teams and registered startups can apply for and receive government compute allocation
    • Benchmarks public: evaluation results for all publicly funded language AI tools are published in a shared registry, independently verifiable
    • Duplication eliminated: ministries demonstrate they drew on shared infrastructure, not separate contracts
    Pillar 3

    Ecosystem and Capacity Building

    Government commits to

    • Develop revenue models for local language AI startups — procurement set-asides, innovation challenge funds, and regulatory sandboxes
    • Fund graduate programmes or university research tracks with a low-resource language AI specialisation and documented employment pathways
    • Invest in speech (ASR/TTS) and multimodal AI data collection and tooling alongside text — voice-first AI as equal priority, not future extension
    • Establish a peer mentorship network connecting language AI researchers across LLA countries, with funded secondments

    What success looks like

    • Startup revenue: local language AI companies winning government contracts or generating revenue without grant dependency
    • Graduate pipeline: a cohort of low-resource language AI graduates completing formal programmes with documented employment outcomes
    • Voice-first working: at least one deployed public service uses ASR or TTS as its primary interface for citizens in the target language
    Pillar 4

    Public Service Deployment

    Government commits to

    • Commission 1–2 high-impact AI use cases in local languages, with a named ministry owner and dedicated operational budget line
    • Integrate at least one language AI tool into an existing government service — health information, agricultural advisory, legal aid or civic registration
    • Support local innovators to build and scale: publish open APIs and language datasets; create a procurement set-aside or challenge fund; partner with local institutions under revenue contracts
    • Train community-level operators — health workers, teachers, agricultural extension workers — to use, maintain and provide structured feedback on deployed AI tools

    What success looks like

    • Use case live: at least one AI service operational in a local language, used by real citizens in a government service context, with published usage data
    • Local innovators contracted: local companies and institutions have won government contracts or challenge fund awards — a domestic AI economy, not grant dependency
    • Open APIs live: a public developer portal with language datasets and APIs is operational, with documented active users
      Operational budget secured: the owning ministry has a recurring budget line — the tool survives beyond the initial project funding cycle
    Pillar 5

    Measurement and Accountability

    Government commits to

    • Define and publish a measurement framework covering AI service uptake, language coverage, equity of access and value-return to communities
    • Establish a documented baseline against all KPIs so future progress is verifiable and comparative, not asserted
    • Publish an annual ELM progress report including benefit-sharing payments, community satisfaction data, and a transparent account of what is not yet working
    • Fund and enable an independent review mechanism with a mandate to publish findings publicly, including critical findings, without government interference

    What success looks like

    • Baseline published: all KPIs have a documented baseline — future progress claims are verifiable, not anecdotal
    • Annual report on time: publicly accessible, covers both progress and shortfalls — communities can access a version in their language
    • Independent review done: an external body has reviewed programme outcomes and published findings without government interference
    • Value returned: benefit-sharing payments to communities are documented, attributed and reported publicly each year