Ekaterina Vylomova & Raphael Merx

There’s plenty of warranted anxiety about the effect artificial intelligence (AI) will have on social and economic inequality. AI built by big tech firms is set to reward capital owners at the expense of workers. Technologically advanced countries will benefit, at the expense of low-income countries. But might AI also be driving linguistic inequality? After all, generative AI chatbots are powered by large language models. These aren’t trained only in English, but there’s a great divide between which languages get included, and which do not. Generative AI works best on majority languages, making minority languages – ones with very little representation online – less relevant. Hula, in Papua New Guinea, has around 10,000 speakers. Tetun, the lingua franca of Timor-Leste, has just over 1 million speakers. Both represent a tiny fraction of AI model training data, and as such are very likely to be misrepresented in the output, if they show up at all. On the other hand, we’ve observed another trend over the past year. AI can also be an enabler, letting local knowledge owners build the tools they envision for their community.

Community tools, with a little help from AI

One such tool is Vavanagi, a language documentation platform for the Hula language in Papua New Guinea. It was built entirely by and for the Hula community, with coding help from AI. With around 10,000 speakers, the Hula community lacks the workforce – such as linguists and software engineers – typically needed to build an online language documentation platform. Yet members of the community (led by Bri Olewale, who contributed to this article) have been able to design, build and run Vavanagi with the help of AI coding tools. The platform now has more than 80 users, who together have contributed over 12,000 English–Hula translations. Their ultimate goal is to collect enough data to build a Hula language translator app. Another example is Tulun, a platform to help health educators translate health education material into Tetun, the main language in Timor-Leste. Built in partnership with local non-governmental organisation Maluk Timor, it lets staff manage their own list of approved terms and phrases, tailoring translated material to their needs and ensuring accuracy. Tulun relies on AI models to adapt the translation to user-uploaded content, making automated translation a collaborative exercise between AI and Tetun health experts. A third example is the Dinka–English dictionary, by Alier Makoi Achuoth from South Sudan. Dinka has about 5 million speakers, yet it has very limited digital representation. The app is designed to support and expand Dinka vocabulary by allowing Dinka speakers to translate and verify word meanings. AI models helped Achuoth design and build the app, collect and organise its data. He told us he wanted to develop a practical solution instead of waiting for another person or organisation to address the problem.

Lowering the bar

These three tools serve different purposes, yet share a similarity: they leverage AI to better channel local knowledge back within their language community. For Vavanagi, this means elder-led review of a language that is losing ground (to Tok Pisin) among Hula youth. For Tulun, this means Timorese health educators adapting training material for Timorese health workers. For the Dinka–English dictionary, this means giving the Dinka community access to educational tools. AI is reducing the time and financial costs needed to put these tools together, to the point where communities can own and build them without any external funding. Minority language speakers don’t need to live in a high-income country to be able to support their needs. This shift in linguistic communities reflects similar geopolitical and socio-economic trends. For example, AI adoption in Kenya and Nigeria is as high as that in the United States, calling into question the framing of low-income countries as laggers in AI adoption. Similarly, AI models are helping smaller companies in the Global South compete with bigger ones. Small businesses can now afford quality machine translation of marketing content where they wouldn’t have been able to afford professional translation.

Communities can manage their own

Research on AI in the Global South often treats local communities as mere affected parties. The reality is more nuanced, and more positive. For researchers and funders, this is an opportunity to move away from top-down research and towards more community-led projects. This is especially important for Indigenous communities across Australia. Cat Kutay, a computer scientist of Aboriginal descent from Charles Darwin University, told us that several First Nations are now working on their own language technology: we understand the needs and the possibilities as well as the culture being presented.

(The writers are Senior Lecturers in The University of Melbourne, Australia)