Visiting the CAI Campus

Last weekend, I had the opportunity to visit the CAI Campus at Institute of Translation and Multilingual Communication (ITMK) at the TH Köln - University of Applied Sciences together with almost 40 fellow conference interpreters participating in the VKD Mentoring Programme.

CAI stands for Computer-Assisted Interpreting, and we had the chance to put a range of CAI tools to the test in a realistic simultaneous interpreting environment inside the booth.

The tools included:

• Cymo Note from China

• InterpretBank from Germany

• SightTerp from Turkey

What made the experience particularly exciting for me was that the CAI Campus is located in the interpreting lab of my alma mater. Remarkably, the setup is still exactly the same as it was twenty years ago. The difference? Back then, we had to insert cassette tapes and actually press the record button. Today, the interpreting booths are equipped with PCs, and screen recordings are uploaded to the cloud for further review.

Under the guidance of Professor Anja Rütten, supported by Nele Kirstein and Magdalena Lindner-Juhnke, mentors and mentees alike thoroughly tested the CAI tools throughout the sessions, and explored a number of questions such as:

• Can AI support simultaneous interpreters in real time?

• How often does AI suggest incorrect terminology ?

• Does simultaneous interpreting become more demanding when live prompting is added to the cognitive workload?

As we interpreted a range of different speeches, transcripts were sent to the booth and, depending on the provider, offered highlighted names, places, and terminology. It was quite distracting, though, that the layout often shifted, the transcripts themselves were frequently incoherent, or the transcript ended mid-sentence. I often focused too much on the wording in the transcript without even listening to what was being said. And honestly, reading out numbers like 50000000 is no easy task, especially during simultaneous interpreting!

All of this consumed valuable cognitive resources. It was easy to lose your train of thought, fall behind the speaker, and many of us only realised several minutes later that there was also a presentation being shown in the room! Perhaps the biggest issue, however, was the noticeable delay compared to the original speech, meaning we had to wait for the transcript because we were actually faster than the machine.

At the end of the day, eight of us senior members took part in a controlled experiment, interpreting a 25-minute speech from English into German while testing the CAI tool Cymo Note under real working conditions. The findings will contribute to ongoing academic research.

So, what is my conclusion?

CAI tool may become more sophisticated in future, but they will not be able to fully replace professional interpreters. AI provides only a transcript. It does not interpret. It relies on statistical probabilities and provides recommendations, but ultimately, it is the human interpreter who must decide how a message is conveyed.

Compared to AI, human interpreters bring cultural knowledge, life experience, subject-matter expertise, a moral value system, and the ability to reflect critically on what they hear and say. Professional interpreters question inconsistencies, assume responsibility for their output, and are therefore inherently more trustworthy than a machine whose processes remain purely cognitive. Interpreting requires context, judgment, and an understanding of broader implications - things that machines struggle with.

You may be interested to know that a growing number of AI systems now claim to provide reliable "interpreting" using Automatic Speech Recognition (ASR) and Voice Cloning. They may sound plausible, yet their output is not checked for completeness, believability, coherence, or accuracy!

For that reason, I believe that interpreting is, and will continue to be, a domain of human expertise. It remains to be seen whether CAI tools will ever become a truly useful tool in the interpreting booth.