Recap of VKD’s round table discussion on artificial intelligence

As promised, here comes the English version of my article published in the German conference interpreters’ association (VKD im BDÜ e.V.) bulletin VKD-Kurier in December 2023

Quo Vadis yet again?

On 17 October 2023, VKD hosted a virtual round table discussion on AI. I was invited to give a keynote speech – a dubious honour to my mind, as there are so many uncertainties around the topic of machine vs. human conference interpreting, not to mention the fact that I don’t quite like this idea of building up defences against the alleged doom of human interpreting. But in the end, I accepted, as I found the idea of discussing this topic with my dear colleagues too tempting to resist.

An association and its purpose

Looking back, the spirit of this virtual round table reminds me a bit of other Quo Vadis discussions held within VKD in the past, when major changes that were perceived as threatening were discussed in detail.

For example, the (in)famous Junior vs. Senior discussion about the enormous increase in the number of young, or rather novice, members, who were back then referred to as “candidate conference interpreters” (in German Koferenzdolmetscher-AnwärterKDA). In 2014, the number of members in our association, which was only 10 years young, had doubled within a few years, with the ratio of Senior (KD) to Junior (KDA) members changing from around 3:1 to almost 1:1. People began to worry that there might be more competition, a drop in quality, or simply too many unknown faces in the association. In the end, however, VKD was both rejuvenated and matured as a result of this development and the heated debates around this “flood of novice interpreters”. Since then, members who haven’t yet submitted their list of 200 working days are no longer called “candidate conference interpreters”, but “Junior members”. And many newcomers from those days have now become experienced interpreter personalities who are highly regarded in the association, many of whom are actively involved and hold mandates.

And then came Covid-19. Our professional lives were frozen overnight and the next thing we knew we were moving into virtual meeting rooms. The death of face-to-face interpreting was predicted by some, the feasibility of videoconferencing was discussed back and forth. Now, after the pandemic, although sound quality is still often an issue, video conferencing and face-to-face interpreting are both common business standards, and our profession has been enriched by a new technical alternative. Now, I wouldn’t want to suggest that our association’s internal affairs and videoconferencing interpreting are as relevant to the future of our profession as machine interpreting. Nevertheless, what all three topics have in common is that irreversible changes have given and are giving rise to fundamental discussions about the concept and future of the profession.

And so, on 17 October 2023 our round table discussion was held under the inspired moderation of Rafael Adam with 35 colleagues attending. Where is our profession heading? Will machine interpreting bring us support, be our competitor, or be the new member of the interpreting team?

At the beginning, a survey was conducted to gauge the general mood. In terms of business expectations, 54% of respondents expected machine interpreting to reduce their market volume. Regarding attendees’ knowledge on AI, the survey showed that 65% of participants knew about AI “what you happen to pick up”, while 35% had familiarised themselves with the topic. No one stated that they “knew nothing at all” or that they were “at the forefront and trying everything out” (the author was not allowed to respond). In any case, nobody came completely unprepared!

Apples and pears

A large part of the discussion focussed on the question of how human and machine interpreting differ and what advantages and disadvantages both have. Machines are superior to humans in particular when it comes to non-content-related factors such as speed, stamina, ease of use, availability, flexibility and price – similar to other intellectual services, not least translating, but also writing advertising copy or legal submissions.

When it comes to in-depth semantic understanding, checking the plausibility of what is being said, and taking into account world knowledge (current AI applications based on large language models are known to “hallucinate”), humans are clearly the more reliable option, at least at the current stage. As Socrates summed it up perfectly in his day: “I know that I know nothing.” Being aware that you don’t know everything is a profoundly human trait that certainly has its merits, as constant self-monitoring is an important quality assurance mechanism in interpreting. Another thing AI struggles to deal with is very specific and, in particular, novel technical jargon as well as sociolects (gang slang). By their very nature, these are constantly changing, just as the reality to which the words refer has often been recently invented. Machines are also not familiar with emotions, jokes, irony, and any kind of indirect message. There is a reason why “interpreting” is called “interpreting”. We can and must be free to detach ourselves from the source text in order to convey the message into the world of the other language.

One thing is crystal clear: a computer cannot interpret like a human being (and a human being cannot interpret like a computer). But the question is: is human interpretation the one and only real thing? Certainly not. Machine interpretation is already being used and there are use cases where it makes sense. Travel agencies, quirky libraries and your grocery store around the corner are also more charming, individual, and cosy than an internet platform. But e-commerce is faster, more convenient and offers a wider range of products on offer. In my opinion, this is not about human parity of machine interpreting. Machine interpreting, which does not come close to human interpreting and follows a fundamentally different approach anyway, can still have valuable use cases. There are situations where not every detail or not even every sentence counts, nuances are not relevant, or where a team of human interpreters is not available for the required language combinations. “Better than nothing” can be a strong point.

Another noticeable difference between human and machine interpreting is the voice. Machine interpreting voices are often perceived as monotonous and tiresome to listen to. On the other hand, deep fakes of real people’s voices are a reality (though not in real time simultaneous interpreting). It is probably just a matter of time and processing power before the two can be combined and the interpreting machine speaks with the same or at least an extremely similar voice to the person being interpreted. It remains to be seen whether the original speaker will be so thrilled about the interpreting machine copying their voice.

Once widely available and providing a minimum level of quality, machine simultaneous interpreting – much as videoconferencing technology has done in recent years – could also help to “democratise” the access to such functions. More people can communicate with persons from other language communities at the click of a button, everyone can speak their native language rather than having to get by in English simply because they can’t afford an expensive simultaneous interpreting team and equipment. On the other hand, developing machine interpreting for “smaller” languages is more difficult because there is much less training data available, which could make the situation less democratic or fair. But this problem is already being addressed by researchers.

AI as a colleague

Another question asked at the round table discussion was whether one would accept working in mixed teams of humans and machines. It turned out that half of the participants could imagine working in a team with machine colleagues, only 19 % would refuse it, the rest didn’t know. This question referred to a scenario where in one meeting some languages would be interpreted by humans and others by machines.

But apart from interpreting, AI can also provide support to interpreters just like booth mates do by means of computer-aided interpreting (CAI) tools, or other generic software solutions. There are various tools that can transcribe the original speech and/or display numbers, technical terms, and names in real time. We will soon be able to tell how useful this is in practice. And it is certainly no longer true to say that AI transcription only works in native Oxford (or American) English. Otter.ai, for example, only works for the English language, but is extremely versatile when it comes to understanding different accents. This includes not only American, Scottish, Irish and Indian English, but also English with a German, Italian, Russian or Chinese accent, among others. Other systems, such as Airgram.io, also transcribe German, French or Spanish with considerable quality, although they do not yet come close to the performance of English transcripts. Live transcription could therefore provide valuable support in terms of accent understanding, speed, and accuracy in the future.

Who is helping whom?

So far, so good – it is obvious that AI can be a great support to humans, taking over some of our tasks. But what about the other way round? Post-editing has been around for decades in translation – but could it also be feasible for humans to monitor a computer’s interpreting output (or rather the real-time translation) in real time and intervene to correct it? A question that was discussed at length at the round table. Especially from our experience in written interpreting, we know just how much cognitive load is involved in comparing a live transcript with what is being said and correcting it in real time – all in the same language. So regardless of technical feasibility, the human resources required for such a task between two languages would hardly be less than those needed for normal simultaneous interpreting. Still, an interesting idea …

So what now?

Finally, there is the question of what our association can do for the profession at a time of rapid technological development. At the round table, everyone seemed to agree that we want to understand the possibilities of AI in interpreting, stay informed so that we can provide the best possible support and advise our clients on the various machine and human interpreting options – true to the motto: “We are in charge of the jokes.”

In the end, I once again realised what I find so fascinating about Quo Vadis discussions. As Goethe said: “With knowledge comes doubt.“ Thinking and rethinking, researching, investigating, having doubts – even about ourselves – is all very human. And it is only logical that as knowledge workers we start thinking and questioning things when being challenged by computer systems that now even pretend to be intelligent. I found it all the more encouraging to experience this spirit of open-mindedness in which our round table discussion concluded. VKD already has live demos of AI interpreting and testing AI-based CAI tools on the agenda. Quo Vadis, which way? This way!

————————

Further reading:

https://uwaterloo.ca/news/media/new-ai-brings-power-natural-language-processing-african (on AI and small languages)

https://blog.sprachmanagement.net/smarterp/ (on Smarterp)

https://blog.sprachmanagement.net/live-transcription/ (on Otter.ai and Airgram.io)

https://blog.sprachmanagement.net/interpretbank-abm/ (on Interpretbank ABM)

————————–
About the author:
Anja Rütten is a freelance conference interpreter for German (A), Spanish (B), English (C), and French (C) based in Düsseldorf, Germany. She has specialised in tec, information and terminology management since the mid-1990s and holds a PhD on information and knowledge management in conference interpreting.

 


Posted

in

by

Comments

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.