The Linguist

6 The Linguist Vol/62 No/1 thelinguist.uberflip.com NEWS & EDITORIAL ChatGPT poses risk The launch of ChatGPT has taken the world by storm and everyone – from the general public to policymakers, professionals and industry insiders – is trying to get their head around how this is going to affect our lives. The Chat Generative Pre-trained Transformer is a chatbot launched by OpenAI in November 2022. According to BBC Science Focus, the model was trained using text databases from the internet, including 570GB of data obtained from books, webtexts, Wikipedia, articles and other internet-based texts. The model was 'trained' by human AI trainers, with 300 billion words fed into the system, and works on probability to predict a naturalistic way to generate text. In the words of The Guardian's Science Editor, Ian Sample: "Essentially a souped-up chatbot, the AI program can churn out answers to the biggest and smallest questions in life, and draw up college essays, fictional stories, haikus, and even job application letters." The results can be stunning. In the House of Commons, Luke Evans MP read out a speech generated by ChatGPT from his simple command to "write a Churchillian speech on the state of the United Kingdom over the last 12 months". However, as OpenAI states, "ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers" and can even "sometimes respond to harmful instructions or exhibit biased behaviour". The implications, and risks, for many sectors are varied and important – from plagiarism concerns in education to cybersecurity risks such as phishing and malware generation. Springer Nature and Science magazines published editorials setting out their positions on legitimate use of these tools, stating that no large language model (LLM) tool will be accepted as a credited author on a research paper. Science made clear that it is "important for scientists to recommit to careful and meticulous attention to detail". For languages, this is also an opportunity and a risk. FE News ran a feature on ChatGPT translations, identifying four categories of risk: accuracy, fluency, bias and legal. Richard Davies, AI Engineer on Guildhawk Tech Development Team, said: "There's a risk too that the resulting translation could include data that is subject to copyright" because the bot's dataset could include copyrighted data. The All-Party Parliamentary Group on Modern Languages recently invited CIOL and researchers in this field, including Sabine Braun and Felix do Carmo from Surrey University, and Lucas Nunes Vieira from Bristol University, to explain the current pros and cons of AI in translation. (For details see www.ciol.org.uk/machine-translation-and- implications-public-policy.) The one common thread is that, however powerful the tool, a human still needs to check the results, especially in high-stakes contexts. In the media PHILIP HARDING-ESCH Language statistics from the 2021 UK Census were published with some interesting trends appearing. Romanian, at roughly 0.8% of the population, is now the third most spoken language in the UK after English and Polish. It was interesting to see many local newspapers celebrating the '10 most spoken languages' in their area. Teacher recruitment figures made headlines for all the wrong reasons. In the TES, school leaders described the situation as "catastrophic". Recruitment of language teachers fell from 69% of the target in 2021 to just 34% (726 teachers) in 2022. Schoolsweek reported that modern languages also saw the highest rate of teachers leaving the state sector within five years of qualifying. In better news, research from the University of Portsmouth analysing graduate job adverts showed "a high demand for language skills" in all sectors. The i newspaper reported heartbreaking stories of Afghans – including many interpreters – who escaped to Pakistan and are contemplating returning to Afghanistan as help runs out, even though this will put them at risk of reprisals from the Taliban. There were positive developments for the visibility of UK languages as Netflix announced its first ever Welsh language TV show, and ITVX launched the world's first British Sign Language channel, with 10,000 hours of programming, as well as subtitles for 90% of its on-demand programmes. It was great to see some buzz in the advertising press surrounding BMW's extraordinary 'Translating Joy' campaign in China. The slickly produced three-minute advert imagines a translation error resulting in a nonsensical English slogan going viral – and invites viewers to "create and share their own Chinese to English AI-translated new year messages via social media". The idea was inspired by the fact that "one of the most popular Chinese New Year wishes happens to include BMW's Chinese name" – Bao Ma, which means 'precious horse'. Philip Harding-Esch is a freelance languages project manager and consultant. © SHUTTERSTOCK

• Cover