Child Rights by Design in Gen AI era

Children's media follows global trends of localisation, which means high and growing demand for translation, dubbing and subbing. The translation, subbing and dubbing market is set to grow 50% in the next four years, according to Kidscreen reporting. While children speaking high resource languages like English, Arabic, Mandarin are well served, several low-resource languages including Welsh, Bengali, Urdu prove costly to translate. This project takes advantage of state-of-the-art translation through generative AI platforms and a reinforcement learning game that children and their families can engage with to tune AI and reduce error rates of low-resource language translation. Work will start in collaboration with an enterprise media company before developing the commercial strategy to reach more content producers. The goal is to make rich learning content within children's media available to more families in cost accessible ways. Medium term, the ambition can be bigger. Building blocks of language acquisition means the way children learn human languages with their parents, referred to as parentese, breaks down individual rules of that language in accessible ways for the child. This project's training data of parentese offers potential to reduce low resource language error rates beyond children's media to support multi-media translation needs longer term.