Artificial Intelligence in Intra- and Interlingual Translation : (Investigating the Potential of the Large Language Model GPT-4o in Plain Language and Low-Resource Translation Contexts)

This study sets out to investigate the potential of the large language model (LLM) GPT-4o in Plain Language and low-resource translation contexts. We do so in the wider context of an artificial intelligence (AI) literacy framework for translation, interpreting and specialised communication, which attempts to model a comprehensive set of competences required in light of new high-performing AI technologies such as LLMs. For each of the two use cases, we explore different prompting strategies and compare GPT-4o’s performance with the performance of a dedicated expert machine translation (MT) system. For intralingual translation, we are unable to conclude that prompting led to a significantly better output. In addition, our analyses showed that the dedicated expert MT system, which was specifically developed, trained and fine-tuned for Plain Language translations, outperformed GPT-4o in all criteria. In terms of low-resource translation, our results point in the opposite direction. While specifically highlighting diasystematically marked elements in the prompt does not seem to have a significant impact on translation, providing GPT-4o with a detailed task description and examples significantly improved the output compared to non domain-specific MT.