Being interpreted by AI – a speaker’s and listener’s perspective

A real first: Last week for the first time ever I spoke at an international event in German, my mother tongue. And on top of that, I was being interpreted by both a machine (in subtitles) and humans. I had the honour to be a member (the only remote one) of an illustrious panel invited and moderated by the one and only Gonzalo Celorio Morayta for the occasion of the opening of CM Idiomas’ Madrid office.

The day’s motto was

IA – Concédeme serenidad para aceptarla, valor para adaptarla, sabiduría para reconocer la diferencia or

AI – Grant me the serenity to accept it, the courage to adapt it, the wisdom to know the difference.

I was joined on the panel by

Dieter Runge, an enthusiastic tec guy I had trouble imagining as the profession’s enemy (which is a compliment indeed ;-)), and provider of the AI-based live caption translation through his company, Boostlingo, which was created to improve workflow and provide new efficiencies for language professionals like interpreters,

Dr Miguel Duro Moreno from Málaga University presenting an intriguing poetic-academic perspective of AI in translation,

Natalia Prío, freelance conference interpreter from Madrid, sharing her precious insights from twenty years of practical experience and the human factor in many different settings.

It was fascinating to compare notes on our experiences with AI in interpreting, even more so as without having coordinated our contributions before, our notion of AI in interpreting was strikingly similar. Here are the main points we all seemed to agree on:

  • Videoconferencing, the penultimate massive technical innovation in conference interpreting, is here to stay (accounting for about half of the meetings in Boostlingo’s business). Interpreters will also be IT workers in the future, as Dieter puts it.
  • Semantics and pragmatics are still an unknown dimension to AI. The same goes for tone of voice, facial expressions, body language, emotions, cultural and colloquial nuances, jokes, irony and subtext, common sense linguistic decisions. As Miguel rightly put using the words of Borges: Ningún problema tan consustancial con las letras y su modesto misterio como el que propone una traducción.
  • Formal, well-structured text in standard language is easier to handle for AI than spontaneous, casual, unstructured and/or local dialect speech.
  • AI can get things perfectly right one moment and completely wrong the next – and you won’t notice the difference. AI cannot check plausibility of what it says.
  • The quality of AI-based translation and interpretation depends on the language pairs (languages of lesser diffusion being disadvantaged).
  • Speed can be an issue for the machine interpretation even more than for humans.
  • Machines can emulate the speaker’s voice. Remains to be seen if this is a positive or negative feature.
  • Confidentiality can be an issue using machine interpreting- an issue often forgotten and rightly mentioned as very relevant in practice by Natalia.
  • Machines don’t need sleep, food, a hotel room, travel arrangements etc. They may be cheaper than human interpreters. (And they won’t run away with Cary Grant – or Audrey Hepburn – if they happen to enter the booth, says Gonzalo).

With me on Zoom was my dear colleague and AI enthusiast Magdalena Lindner-Juhnke. Unlike me, she could dedicate all her attention to the live caption translation provided by Boostlingo.  This was Magda’s experience as a listener (or reader) who, however, understood the original while she was reading the translation.

While, at the conference, multilingual captions were shown on a large screen (entertaining an audience full of language professionals), I read along on my smartphone – just one click away.  I started with a Spanish-to-Spanish transcription, which was indeed very accurate. Too accurate maybe. It gave me every word, every slip of the tongue. Many of the sentences were longer than what could fit on my phone’s screen. While I do enjoy good (predigested and concise) subtitles, this was very tiring and, after a few minutes, I found it hard to stay focused. For this to become a more enjoyable experience, the AI interpreter would need to learn what we like to call ‘salami slicing tactics’, converting a large sentence into several shorter ones. Or speakers working with AI interpretation/captioning would need to adjust their way of speaking, as Anja later pointed out in her “Eight rules to get the most out of live translation apps”.

I then switched to German subtitles … and got somewhat lost. The speaker talked about a Greek legend where 70 translators were supposed to translate the Bible in 70 days. That I got. But why did several people no longer buy a book and/or a litre of milk in Hebrew? That was when I realised how good we, as professionals, actually are at predigesting content for our listeners!

Finally, I switched to English and the quality of the captions improved. Given the extensive amount of Spanish-English translations the AI can feed off, this is probably one of the most reliable language combinations for AI translation and interpretation. Until the speaker, Miguel, deliberately challenged the AI by saying that a hot chocolate he had ordered in another country tasted like “aguachirri” (“Plörre”, as we say in German, dishwater in English). That may have ‘broken’ the AI, because this is what the captions read: “… they brought me a agua chirrié. Something like, I mean, like water, water, water, water, water, water chirri, cherry water.” Not quite. As we’ve all learned at university: You can’t just make stuff up. If you’re not sure what it is or how to say it, use interpretation strategies, generalise.

Even if AI is not (yet) able to make this kind of decision, here is what I (as a consumer of AI translation) would like to see in the future: Whenever the AI can’t find a good match and starts guessing or hallucinating, I’d like to be made aware of that. In written texts, those sentences could be marked with a red squiggly line. For oral interpretation it’s tricky indeed. Maybe the app could include a traffic light system (using green, yellow and red lights) indicating the likelihood of a sentence being correct or warning me about possible mistakes? Just thinking out loud here …

Thinking out loud, a very human skill indeed! One afterthought from the perspective of a speaker being interpreted: At one point I made an obvious mistake, a slip of the tongue really, saying “meaningful” instead of “meaningless”. Any human interpreter would have corrected this lapse, the machine of course did not. I haven’t decided yet if I want to praise AI for its honesty, or be offended, just to do justice to my role as a human being.

As both Magda and I were neither present at the event nor depended on the translation, I would also like to add the perspective of those physically present and actively using the live captions. Here’s what our host Gonzalo thought of it:

Even if AI does not have the same capabilities we interpreters and translators have, it somehow gets the gist across. Dieter, Miguel and myself were obviously able to understand what was being said e.g. in German by Anja and comment on it. Someone from the audience put it like that: The responsibility of correcting obvious mistakes,  filling in the blanks, or generalising when something is not clear are tasks that in this AI-interpreted situation are transferred from the interpreter to the final user who, in the end, is not a machine but a person who probably has enough judgment and knowledge on the topic of a conference they have chosen to attend to pick up what the machine didn’t. Whilst I am confident that we interpreters and translators will still remain part of the equation for a long time, I have no doubt that AI can be successful in some spheres. I hope that the industry gets together to help shape the right scenarios for AI instead of letting it creep up behind our backs and take over.

 So let’s hope Dieter is right in seeing the future of AI in interpreting as the Augmented, Accelerated, Advanced, Adaptable and Acclimated Interpreter – and the Advising Interpreter as well, I suppose. For as language service providers we will need to advise clients on when and how to use AI interpreting – and when to better rely on human interpreting. I’m sure there’ll be those who won’t mind correcting obvious mistakes and filling in the blanks and others who will be very happy to sit back and let us do the job for them.

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.






Leave a Reply

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