Since the release of AI chatbots in late 2022, students and trainers of conference interpreting
have found them to be of considerable help in their training. AI chatbots can quickly generate
speeches, extract terminology and give limited feedback on transcribed interpretations. The
results of these processes are varied, depending to a certain extent to the quality of the
prompts that are used. Surprisingly, there is no systematic research on the subject and no
clear way ahead to enhance the quality and usefulness of the chatbot outputs for interpreter
training. The goal of this project is to find out whether the technique of retrieval generated
generation (RAG) can be employed to “enrich” the chatbot in such a way that it achieves
results that are more usable in pedagogical tasks. The two tasks we will focus on are the
generation of training speeches and feedback on self-study interpretation tasks.