Year: 2024
Type: article
Abstract: When I joined Bangor University in 1993 there was very little understanding of the huge role computers, the world-wide web and digital communications would have on everyone's lives within thirty years.English was the dominant language of these media, to be gradually joined by other large global languages.It was a very different story for small, less-resourced languages, including endangered and minoritized languages such as Welsh.To them, the advent of these new technologies and digital media were a threat, endangering their ability to survive and thrive.The new technologies represented power and excitement, economic opportunities and a prosperous future, making the languages that could not access them appear old fashioned and impoverished, and therefore less likely to be transmitted to the next generation.It was against this background that a small group of us at Bangor University decided to start developing language technologies for Welsh, partnering as far as possible with industry and the volunteer sector so that the wider community could benefit from our research.Our aim was to enable the Welsh language to obtain the necessary resources to thrive in the digital world, and that is now beginning to be realised in areas such as speech technologies, text technologies, and machine translation.But this is a fast-moving world, and even as we were publishing the first volume of Language and Technology in Wales in 2021, few of us foresaw the forthcoming revolution that would soon arrive in the wake of generative AI such as ChatGPT.This technology can work for many languages given enough text or speech data to train it, and 'enough' here means vast sums of data, involving at least a billion parameters according to latest estimates, something that very small languages struggle to achieve.Fortunately there are ways of working together across different linguistic communities to learn from each other and develop new techniques to achieve our aim.This is happening across Europe and the rest of the world, and closer to home, researchers working on Welsh and other Celtic languages have begun to collaborate, since there are many social, political and linguistic reasons for sharing our experiences and techniques.The results of some of this collaboration may be seen in the present volume, and more is yet to come.Support from the European Union, the Welsh Government and other bodies have helped Celtic language technologies mature, and I look forward to seeing future developments.I wish to thank all my colleagues at Bangor University, across Wales and beyond, and especially in Celtic language technologies for all their support and inspiration during the last thirty years.with access to the internet are able to make use of AI through chatbots such as Microsoft's Copilot and OpenAI's ChatGPT.Uses for AI are diverse and many, and not limited specifically to LT or chatbots.For instance, AI is being used to improve crop production, 2 to identify counterfeit paintings on e-bay, 3 and to upscale and improve television images.4 AI then has the potential to do great good, but if this technology is not made available to minoritized languages it could have a devastating effect in terms of the perceived status and the use of those languages.AI needs to be made available for minoritized languages, then, in order that there are no linguistic barriers to using said technology, and in order that minoritized language speakers don't turn away from their language to use AI.Effects on language use and status aside, this technology still needs further development, and that in areas which affect minoritized and majority languages alike.It can suffer from hallucinations, where the system: "perceives patterns or objects that are non-existent or imperceptible to human observers, creating outputs that are nonsensical or altogether inaccurate."[1] As has been widely reported, the energy used, and therefore environmental impact of training and using AI models, is worryingly high.5 Even a normal everyday single use (or inference) of the technology has an alarmingly high energy cost and carbon footprint.In discussing the results of their research, Luccioni et al. [2] note that:"the most efficient text generation model uses as much energy as 16% of a full smartphone charge for 1,000 inferences, whereas the least efficient image generation model uses as much energy as 950 smartphone charges (11.49kWh), or nearly 1 charge per image generation."Some are concerned about the impact of AI on jobs and the economy, indeed Georgieva [3] claims that "AI will affect almost 40 percent of jobs around the world, replacing some and complementing others", and that this figure may increase to 60 percent in advanced economies.Clearly it is true that there remains work to be done on mitigating these issues.1
Authors: Morris, Jonathan, Ezeani, Ignatius, Gruffydd, Ianto, Young, Katharine, El-Haj, Mahmoud, Knight, Dawn
Institution: Université de Pau et des Pays de l'Adour
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7011054591