Real-time AI captioning has made multilingual events more accessible, yet the tools that support professional interpreters lag behind. According to a new arXiv paper, computer-assisted interpreting (CAI) tools are typically designed for interpreters, with little input from the interpreters themselves. The paper argues for a shift: interpreters as developers, not just end users.
The study takes an autoethnographic approach, meaning the author is both the researcher and the subject. In this case, the interpreter directly participates in building the tool, working alongside a large language model. This human-LLM co-creation is presented as a way to close the gap between professional practice and tool design.
The paper is a design study rather than a controlled experiment, so it does not offer quantitative results. Instead, it offers a reflective account of what happens when the interpreter takes the lead in development. The authors suggest that this approach could lead to CAI tools that better match the real workflows and needs of interpreters.