AI Inequalities: Minority Languages
The question of how minority language speakers are impacted by AI in the workplace has garnered little attention from researchers. But research into how AI might threaten the preservation of minority languages by promoting the further rise of so-called ‘super languages’, and English in particular, on a global scale suggest that minority language speakers are at risk of marginalisation and exclusion with the uptake of AI in the workplace. Although technological advancements are being made to develop AI tools that can support and even strengthen minority languages, the dominance of closed systems built on and for majority languages pose a significant barrier to the preservation of minority languages.
A central challenge for the preservation of minority languages with AI is the limited resources and data available to train AI models on. This means that it is necessary to build up datasets that are sufficiently large and diverse enough to allow for large language models to be trained on to a good enough standard that they can be used for real-world tasks. This has both cost implications and consequences for accuracy that mean that minority language speakers are more at risk of exclusion and discrimination.
A clear example of some of these limitations is the growing reliance on generative AI tools, such as ChatGPT, across workplaces. Researchers suggest that even if ChatGPT is trained on another language than English, the time and labour required may make it largely unviable. Current assessments of the performance of ChatGPT for different languages have shown that it performs worse in all tasks. Given the broad adoption of ChatGPT for English in different problems and areas, a natural question is therefore whether ChatGPT can also be applied effectively for other languages, or if it is necessary to develop more language-specific technologies. This is an important finding to help guide the proliferation of government partnerships with Gen AI chatbots, such as the recent partnership with OpenAI in Wales. At the same time, past research on the availability Welsh language applications indicate that the market for Welsh language applications is fairly robust and could therefore extent to generative AI. In fact, major projects are underway in both Wales and Scotland to improve the functioning of computer systems for minority languages such as Welsh and Gaelic, including funding for research towards the production of large language models catered specifically for such speakers.
Importantly, for the uptake of technologies catered to minority languages to be successful, there is significant research showing that this requires not just the availability of the technology, but the facilitation of access, including educational resources that can help minority language speakers adopt tools within their particular contexts. That means incorporating a culturally sensitive learning environment that supports minority language speakers that can then translate into lifelong applications in giving minoritised people access to spaces where they have previously experienced exclusion. A major concern raised by researchers is therefore the closed nature of the AI models being taken up in workplaces that limit opportunities to adapt and integrating these tools into a diverse range of settings, especially for minority groups such as minority language speakers.
In light of the issues regarding the lack of sufficient data to train AI models on different languages, significant efforts are being made to build linguistic technological infrastructures that can support the development of AI models in minority languages, including in Wales, where emphasis has also been made on building educational tools and translation tools that can support Welsh speakers across different sectors. However, some have called for more to be done to advance ‘linguistic justice’ in the context of AI that would allow for minority language speakers to be more actively involved in both the design and application of new technologies, alongside this investment into linguistic infrastructures. This is a pressing challenge if minority languages are to be preserved in an AI era, particularly as adoption of such technologies become mainstreamed within public and work life. For minority language speakers, the growing use of ChatGPT in workplaces, for example, risks creating further barriers and exclusion, especially considering the speed at which these technologies are being developed and adopted.
TUC Cymru is campaigning for all workers to be protected against the risks of AI. If you’re a worker who speaks a minority language like Welsh and are concerned about these issues, raise them at your trade union branch. TUC Cymru has successfully negotiated guidance on the use of AI in the public sector. Use it and adapt it for your workplace.
The TUC is campaigning for additional legal protections against the threats of AI and has produced a range of materials to assist reps and officers.
The full report AI Inequalities at Work is published on the Data Justice Lab’s website.
