How to Become Massively Multilingual
When shopping in Amsterdam as a native Dutch person and addressing the cashier in Dutch language, don’t be surprised if you hear a bitchy “English please” as a response. This is what happened to me. Is that bad? I don’t know, but it does make me think that I don’t want to go back to that shop.?
It also makes me think of the ‘Great Debate’ at a TAUS Forum in 2014 between Nicholas Ostler and Lane Greene on the question whether English will survive as a lingua franca or be replaced by MT (AI these days). It’s only a matter of time, argued Nick Ostler, for MT to replace English as ‘the last lingua franca’. As a technology optimist pur sang I voted for his side at this debate. But now, ten years later, I am not so sure anymore.
A mismatch of 20th century thinking with 21st century technology
Yes, the technology is there and it’s doing miracles. What is lacking though is the basic idea about how we should apply the technology. After fifty years of evolution in the translation industry we still act with a twentieth century export mentality. Everything always starts from English. Every decade we have seen shifts in tooling and innovations in processes. But our basic thinking never changed: translation, localization, globalization - whatever we called it - remained one-directional.?
Even now that we have arrived at the ultimate stage in this evolution and start imagining a world in which every piece of content is ubiquitously available in everyone’s native language, we can’t help but look through an English filter. All the popular Large Language Models have been trained almost solely, for more than 90%, on English language data. That’s a fundamental failure: a mismatch of 20th century thinking with 21st century technology.
(The TAUS infographic below is copied from the article in MultiLingual Media on Translation Economics of the 2020s.)
领英推荐
The real revelation of the AI revolution is that all humans potentially get access to all information, and yes of course in their own language. But not as long as we keep operating with a last century English-centric view of the world. The way we train?
models now makes AI arrogant and turns people off. With the state-of-the-art of AI technology these days we do not even have to confine our thinking to a sequential and one-dimensional process of translation anymore. AI technology enables us to generate content in any language from scratch. What happened is that technology jumped ahead of us before we could get our minds around it.
How to become massively multilingual
Two tips for everyone who wants to master AI in every language:
Written by Jaap Van Der Meer
Senior editor at The Economist
11 个月I am watching the development of AI with great interest, and following it as well as I can. I am amazed by its progress. But (without wanting to shoot my fox on any further contributions to this debate!) I'd repeat a question I asked rhetorically at that 2014 event: is anyone ready yet to go on a date mediated by AI? Imagine translation, speech-to-text and text-to-speech working near-flawlessly in both directions. All these technologies are already excellent. But is it even *thinkable* to have a high-stakes conversation (and if not a date, imagine a bet-the-company business deal, or a farewell to a dying loved one, or whatever you like) yet? This remains my point of departure, the reason for language-learning.