IGF 2026 WS #252 AI for the Unheard: Governing Language-Inclusive AI

Description

This hybrid workshop will explore how Internet governance can make AI language technologies more inclusive, accountable, and responsive to the roughly four billion people who speak marginalized languages and remain locked out of digital conversations that concern them. Of 7,000+ languages spoken worldwide, fewer than 100 have enough digital presence to meaningfully shape a large language model and language exclusion is not a technical problem alone, but a redistribution of opportunity, voice, and safety.

The session pairs implementation evidence with structured multistakeholder exchange. UNICEF will share practical lessons from Laaha, its Virtual Safe Space for women and girls, including pilots of open-source text-to-speech and translation models in Swahili, Bengali, Turkish, Lingala, and Marshallese. In humanitarian contexts, where information may relate to gender-based violence, sexual and reproductive health, or access to services, inaccurate translation or culturally inappropriate synthetic speech can create serious risks. CLEAR Global and GSMA will present joint findings on language exclusion in mobile-delivered humanitarian platforms and the gender digital divide: women in low- and middle-income countries are 14% less likely than men to use mobile internet, and 885 million women remain unconnected. An African AI developer building large language models for African languages will speak to what community-led model development reveals about data ownership, consent, dialect, and sustainability.

The session is organized around three anchor questions: what is broken, what is working and at what cost, and what should be built together with participants as co-contributors rather than an audience. It will conclude by identifying practical governance principles and priorities for inclusive language AI, feeding an open repository of lessons, workflows, and ethical guidance.

Policy Question(s)

• How can communities meaningfully participate in the development, ownership, and governance of AI tools for low-resource languages including consent over voice and language data? • How should governments, mobile operators, technology providers, donors, and civil society balance openness, safety, cultural accuracy, and long-term sustainability in language AI? • What standards and accountability mechanisms are needed to reduce harm from AI translation, content moderation, and synthetic speech in sensitive and humanitarian contexts?

Expected Outcomes

Participants will gain a clearer understanding of language exclusion as an Internet governance issue spanning access, mobile infrastructure, data governance, AI development, and human rights.

The workshop will produce: • Practical governance principles for developing and deploying AI for low-resource languages. • Recommendations for meaningful community participation, consent, and responsible stewardship of language and voice data — including the question "who owns a voice once it has been modeled?" • A framework for balancing openness, cultural accuracy, safety, affordability, and sustainability. • Concrete collaboration pathways among humanitarian organizations, the mobile sector, language specialists , and African LLM developers.

Findings will feed an open repository of implementation lessons, workflows, models, and ethical guidance and inform future language technology pilots and public-interest initiatives in underserved languages.

Format

Roundtable Duration (minutes): 60 Format description: A 60-minute roundtable workshop is best suited because the session is designed for focused multi-stakeholder exchange, not a sequence of presentations. The roundtable layout reduces hierarchy and enables direct discussion among humanitarian practitioners, language specialists, technology companies, policymakers, civil society, and affected-language communities.

The session will follow three stages:

0–10 minutes: Brief evidence-based framing from UNICEF, CLEAR Global, and GSMA on language exclusion, access, and implementation challenges.
10–45 minutes: Facilitated discussion across the three policy questions, examining community participation, data governance, openness, cultural accuracy, sustainability, and tensions between AI-generated and human voice.
45–60 minutes: Participants prioritize safeguards, responsibilities, and opportunities for collaboration.

Short visual prompts will introduce each discussion block. Online participants will contribute through moderated speaking queues, live polls, and a shared board. A rapporteur will capture recommendations visibly, creating the foundation for the session’s open repository and follow-up work..

Hybrid Format

Hybrid Format: The onsite and online moderators will operate as one facilitation team, alternating between room and remote contributions so online participants are not treated as an afterthought.

A shared digital board will collect examples, risks, and proposed safeguards before and during the workshop. After brief opening interventions, participants will work through structured questions in mixed onsite-online discussion rounds. The online moderator will surface chat contributions, conduct quick polls, and invite remote participants at designated intervals. The onsite moderator will use visible timekeeping and reserve at least half the session for participant discussion.

Contributions may be submitted anonymously through Slido or Mentimeter, while a Miro or Padlet board will capture model examples, governance gaps, and recommendations. Materials and prompts will be shared in advance in accessible formats. The rapporteur will consolidate contributions from both participation channels into one outcome document.

Organizers

Organizer 1: Caroline Masboungi, UNICEF Organizer 2: Eugenie Humeau, GSMA Organizer 3: Wakesho kililo, CLEAR Global

Speakers

Speaker 1: Eugenie Humeau, Private Sector, Western European and Others Group (WEOG) Speaker 2: Wakesho kililo, Civil Society, African Group Speaker 3: STEPHEN OBONYO, Technical Community, African Group

Moderator

Caroline Masboungi, Intergovernmental Organization, Intergovernmental Organization

Online Moderator

Abeera Akhtar, Intergovernmental Organization, Asia-Pacific Group

Rapporteur