Creole languages have long been poorly served by translation technology. A report presented at WMT 2026 examines how researchers are trying to close that gap through machine translation and language identification. Teams tested different approaches, including fine-tuning models with carefully prepared language data. Smaller systems also performed well on some language-identification tasks. The researchers caution that overlapping training and test material, along with differences in spelling conventions, can make results look better than they are. The findings offer a useful account of work in progress, rather than proof that translation problems have been solved. The proceedings carry October 28–29 dates.
