Four villages in Tibet have a lot to tell us about AI and the future of linguistic diversity

Rebgong’s museum of Tibetan culture was built on lands acquired by the government from the Manegacha-speaking community of Nyantok. Photo by Gerald Roche. Used with permission.

Gerald Roche is an associate professor in politics at La Trobe University in Melbourne, Australia, whose work focuses on power, the state, colonialism and racism. His piece below was originally published by Cornell University Press and is republished here with permission. 

Artificial intelligence is not going to save the world’s languages. It’s hardly going to make any difference at all for the thousands of communities around the world currently under pressure to take up dominant languages. I know this because of lessons I learned studying four villages in Tibet.

If you get your information from elsewhere – if you listen to the prophets of Silicon Valley, for example – you might be full of enthusiasm for AI’s potential to save the world’s threatened languages. It seems like we’re making progress, fast.

In June this year, scientists working for Meta published an article in the journal Nature introducing their project “No Language Left Behind,” which aims to create a “universal translation system” that will support endangered languages. The same month, Google touted its addition of 110 new languages, including Tibetan, Cantonese and Afar, to Google Translate. In response to developments like these, a chorus of journalists, tech companies and academics has emerged to sing the praises of AI and its potential to save languages from the Amazon to the Arctic.

My new book, “The Politics of Language Oppression in Tibet,” contains a number of insights that should temper our enthusiasm. The book draws on eight years living and working on the northeast Tibetan Plateau in China, followed by several years conducting research in four villages where around 8,000 Tibetans speak a distinct language they call Manegacha.

What I learned in those four villages is that unless underlying economic, social and political relationships are drastically altered, AI is never going to reach the communities where it is most needed.

A family compound, with rooms surrounding a central courtyard, in Tojya, one of the villages featured in the book. Photo by Gerald Roche. Used with permission.

First – the economic dimension. Endangered languages are typically undermined by the state’s refusal to provide them with the material resources they need to thrive. China is home to around 300 languages, but most resources go towards promoting the national language, Mandarin. Languages like Manegacha receive zero material support from the state, so, no schools, no books, no mass media, no street signs, no medical translations or emergency services. Nothing. A state that has willfully defunded so many languages will not invest its resources in the expensive and labor-intensive process of developing AI for them. Given the global ubiquity of underfunding for minority and Indigenous languages, I expect that China will not be alone in this.

Next is the social dimension. People who use endangered languages are typically socially marginalized, chronically disrespected and systematically discriminated against. The state might play a direct hand in this, or it might simply provide impunity to perpetrators. In China, Tibetans are cast as a culturally backwards and potentially seditious population relative to the Han majority. Manegacha speakers, meanwhile, face discrimination from other Tibetans as a linguistic minority in the wider Tibetan community. Wherever they live, people who use endangered languages face some form of civilizational supremacy, racist oppression or cultural chauvinism. Artificial intelligence will do nothing to dislodge the privilege and conceit that drives language endangerment.

The city of Rongwo. The town is expanding to the north (the righthand side of this picture), toward
Ngandeghua-and Manegacha-speaking villages. The northern tip of the city is already built over lands acquired from Nyantok, a Manegacha-speaking community. Photo by Gerald Roche. Used with permission.

Finally, there is the political dimension. Communities that use endangered languages are typically unable to improve their material or social situation because they are excluded from political institutions and processes. They are denied their right to self-determination and their political and civil rights are violated. Civil society in China is highly restricted, and small, vulnerable communities like Manegacha speakers cannot mobilize to defend the rights they are systematically denied. Artificial intelligence will not change this situation in China, nor will it contribute meaningfully to defending democracy in a world where it is increasingly under threat. That is because the corporations that control AI tech are themselves fundamentally undemocratic organizations whose underlying motive is profit, not political enfranchisement.

Looking at the world from the perspective of those four villages in Tibet, it is clear that artificial intelligence cannot solve the material, social and political problems that are driving global language endangerment. In concluding my book, I suggest how we might start moving towards more effective solutions through building transnational solidarity in pursuit of meaningful systemic change. But we have a long way to go, much to do and too little time to be distracted by the empty promises of tech corporations.

Start the conversation

Authors, please log in »

Guidelines

  • Please treat others with respect. Comments containing hate speech, obscenity, and personal attacks will not be approved.