Artificial Intelligence as a Lifeline for Endangered Languages and Cultures

Introduction

Language and culture are intrinsic to human identity. Culture embodies the ideas, customs and social behaviour of a society, and the Dutch social psychologist Geert Hofstede famously described it as the collective “mental programming” that distinguishes members of one group from another. Across generations, the values of every society have been passed down primarily through language. When a language is lost, much of the knowledge, humour, oral literature and worldview carried within it is lost with it.

This article argues that emerging technologies, particularly artificial intelligence (AI), offer a practical means of slowing that loss, and that Zambia is well placed to use them for its children and youth, provided the tools are shaped by local communities and guided by deliberate public policy.

The scale of language endangerment

Many of the world’s languages face a serious threat of extinction. The third edition of UNESCO’s Atlas of the World’s Languages in Danger lists some 2,500 endangered languages, approaching the generally accepted estimate of around 3,000 worldwide. UNESCO now draws on data from 2,728 of the world’s roughly 6,500 languages to monitor linguistic diversity. The Atlas serves both as an advocacy tool, raising awareness among policymakers, speaker communities and the public, and as a mechanism for tracking trends in linguistic diversity.

The central mechanism of endangerment is well understood: a language is endangered when it is no longer passed on to younger generations. UNESCO’s expert framework on Language Vitality and Endangerment treats intergenerational transmission as the most important single factor in assessing a language’s vitality. Africa is heavily exposed. When UNESCO launched the online Atlas, it noted that roughly one-third of the world’s languages are spoken in sub-Saharan Africa, and that around 10 per cent of them could disappear within the next century.

What drives the decline

Language loss has several drivers. UNESCO’s vitality framework points to external pressures such as military, economic, religious, cultural and educational subjugation, alongside internal factors such as a community’s negative attitude towards its own language. Migration and conflict disperse speaker communities and weaken the everyday settings in which languages are used.

Globalisation, accelerated by the internet, adds a newer pressure. Online content is heavily concentrated in a few languages: English alone accounts for about half of all websites whose content language is known. The same imbalance extends to AI itself. Research presented at the Association for Computational Linguistics found that most of the world’s languages have very few digital resources, leaving their speakers poorly served by current language technologies. Without deliberate intervention, digital life risks reinforcing the dominance of a handful of global languages and cultures.

The Zambian context

Zambia is linguistically rich: Ethnologue lists 46 languages for the country. Under the 2013 Zambia Education Curriculum Framework, seven Zambian languages (Icibemba, Cinyanja, Kiikaonde, Lunda, Luvale, Chitonga and Silozi) serve as languages of instruction up to Grade 4, with English taking over from Grade 5. This policy reflects strong international evidence that children learn best in a language they understand. UNESCO’s Global Education Monitoring Report estimates that around 40 per cent of the world’s population lacks access to education in a language they speak or understand.

Yet policy in the classroom cannot on its own secure transmission in the home. From our country perspective, a growing number of educated, urban families choose to raise their children exclusively in English. Many of these children grow up unable to converse in their parents’ mother tongue, and those who learn it often do so only much later. This is precisely the break in intergenerational transmission that UNESCO identifies as the root of endangerment.

None of this is an argument against English. As an international language, English is necessary for social mobility and for engaging with shared global challenges. The concern is balance: English should be learned as an additional language that opens doors, not as a replacement that closes off a child’s own heritage.

How AI can help

This is where AI offers genuine promise. AI-driven learning tools can help children and youth engage with their native language and culture in new and appealing ways. Examples include bilingual and multilingual storybooks, culturally specific illustrations created with generative AI, speech tools that let children hear and practise pronunciation, and virtual or immersive experiences of traditional settings and stories.

Zambia is not starting from zero. Researchers at the University of Zambia have released Zambezi Voice, an open-source speech corpus covering Bemba, Nyanja, Tonga, Lozi and Lunda, with more than 80 hours of labelled speech and 160 hours of radio recordings, built to support speech recognition for Zambian languages. Resources of this kind are the foundation on which reading assistants, pronunciation tutors and interactive storybooks in Zambian languages can be built.

Exposure matters as much as tools. Introducing children to the artistic expressions of their language, including poetry, stories, proverbs and fables, can build a deeper appreciation of its sounds, grammar and written form than basic conversation alone. Generative AI can support this by producing illustrations that bring culturally specific characters, settings and expressions to life, making heritage content as engaging as the global media children already consume.

Co-creation and ethical safeguards

AI language tools should not be something done to communities. They can include features that let children and youth co-create and co-design stories, characters and illustrations, giving them agency in the learning process alongside AI. Participatory approaches have already proved effective in African language technology: the Masakhane community showed that native speakers can build machine translation resources for dozens of African languages when involved from the outset.

Ethical safeguards are essential. Models trained on limited data can produce inaccurate grammar, misrepresent cultural practices or reproduce the biases of dominant languages, and there is a real risk of cultural appropriation when heritage content is generated or commercialised without community consent. UNESCO’s Recommendation on the Ethics of Artificial Intelligence calls for AI systems that respect and promote cultural and linguistic diversity. Speaker communities, elders, linguists and educators should therefore review AI-generated content, and communities should retain a say over how their language data is collected and used.

Implications for public policy

For developing countries, AI is becoming a viable intervention for language and cultural preservation, which makes it a legitimate subject of public policy. International bodies are active in this area, most visibly through UNESCO’s International Decade of Indigenous Languages (2022–2032), but the burden of preservation must also be borne by speaker communities and national governments.

Policy discussion in Zambia should focus on how to adapt AI to the country’s specific linguistic and cultural needs. A multisectoral approach is required, bringing together the ministries responsible for education, youth, arts, tourism and digital technology, alongside universities, traditional leaders, publishers and technology developers. Practical priorities include funding open datasets for Zambian languages, supporting local developers to build child-friendly learning applications, and embedding community review and data protection in every initiative.

Conclusion

As a vehicle of culture, collective memory and values, language is an essential component of identity and a building block of living heritage. The threat to Zambia’s languages is real, but so is the opportunity. With local data resources emerging, a supportive language-in-education policy already in place and AI tools becoming more accessible, the time to adopt AI-powered language learning for Zambia’s children and youth is now.

Sources

Official and primary sources

•      UNESCO (2010). Atlas of the World’s Languages in Danger, 3rd edition (project overview)

•      UNESCO (2023). Towards World Atlas of Languages

•      UNESCO Ad Hoc Expert Group on Endangered Languages (2003). Language Vitality and Endangerment

•      UNESCO (2016). If you don’t understand, how can you learn? GEM Report Policy Paper 24

•      UNESCO (2021). Recommendation on the Ethics of Artificial Intelligence

•      UNESCO. International Decade of Indigenous Languages 2022–2032

•      UNICEF (2017). The impact of language policy and practice on children’s learning: Country review, Zambia

•      W3Techs (2026). Usage of English as content language on websites

Research

•      Sikasote, C. et al. (2023). Zambezi Voice: a multilingual speech corpus for Zambian languages. Interspeech 2023. Dataset: unza-speech-lab/zambezi-voice

•      Mandyata, J. et al. (2023). Theory, policy, and practice: bridging the gap between teacher training and classroom practice in language of instruction in Zambia. Language and Education, 38(2)

•      Joshi, P. et al. (2020). The state and fate of linguistic diversity and inclusion in the NLP world. ACL 2020

•      Nekoto, W. et al. (2020). Participatory research for low-resourced machine translation: a case study in African languages. Findings of EMNLP 2020

•      Hofstede, G., Hofstede, G.J. and Minkov, M. (2010). Cultures and Organizations: Software of the Mind. 3rd edn. New York: McGraw-Hill (book; no online link)

News

•      UN News (2009). UNESCO on-line atlas seeks to save dying languages