What if AI could finally speak your language?

 A coalition of 60 organisations from across the artificial intelligence ecosystem has committed to helping an estimated 3.4 billion people use AI in the languages they speak and in their own voices over the next five years.

The initiative brings together AI companies, researchers, governments, philanthropic organisations, technology developers and community groups to address what the coalition describes as one of the biggest gaps in access to AI — language.

The world's roughly 7,000 languages are not equally represented in the data, tools and benchmarks used to develop and test AI systems. Only a small proportion are considered sufficiently well-resourced to support strong AI capabilities.

For speakers of languages that are poorly represented in AI systems, this can mean tools that are less accurate, less useful or less able to understand local ways of communicating.

The problem is particularly important in sectors such as health, education, agriculture, financial services and public services, where inaccurate information or poor interpretation can have consequences beyond inconvenience.

The coalition says the challenge is not simply one of translating English or other widely supported languages into local languages.

Dialect, slang, idioms and cultural context can alter meaning. Voice is also important, particularly for people who may find typing or text-based interfaces difficult or where literacy and access to written information are barriers.

The coalition's five-year goal is for an estimated 3.4 billion people who speak languages currently underrepresented in AI models to be able to use AI tools in their own language and voice.

Among the initial signatories are Amazon, Anthropic, Google, Microsoft, NVIDIA, OpenAI Foundation, Mistral, Mozilla Data Collective, UNICEF, the World Bank Group, Gates Foundation, the UK Foreign, Commonwealth and Development Office, AI Singapore, AI4Bharat, Masakhane, Data Science Nigeria and the Ministry of Telecommunications and Digital Affairs of Senegal.

The coalition says no single company, government or foundation can address the language gap alone.

Its work will focus on four areas.

The first is building an open language layer, including shared and responsibly collected language data that developers can use under open licences.

The second is developing assessments and benchmarks to measure whether AI systems are actually improving in underrepresented languages.

The third is turning language data into models and applications that can be used by AI developers, including those without the resources of the largest technology companies.

The fourth is ensuring that the expansion of AI language capabilities is carried out responsibly, with attention to privacy, consent and data sovereignty.

The coalition says local communities should play a central role in building these systems, including through local research, capacity building and ownership.

That could be particularly important in Africa, where thousands of languages and dialects are spoken but many have limited digital data available for AI development.

The statement also acknowledges that language alone will not determine whether people benefit from AI. Connectivity, electricity and access to affordable devices remain barriers to using digital technologies in many communities.

Over the coming year, the organisations say they will work together to develop the coalition's structure, governance and specific workstreams.

The initiative invites additional community organisations, researchers, technology companies, governments and funders to join.

Its stated ambition is not simply to make AI capable of translating more languages, but to make it possible for people to interact with AI in the language and voice they use in everyday life.

For health workers, farmers, teachers and communities that rely on local-language communication, the potential implications are significant. But whether the initiative can deliver useful and accurate AI in thousands of underrepresented languages will depend on the quality of the data, the involvement of communities and whether the resulting tools become available beyond the organisations building them.