PROFILE

2022 was a big year for Translated.

A little more than two decades after the company first set up shop, the International Data Corporation (IDC) ranked it higher than key players in the tech industry like Google, Amazon, and Microsoft as a vendor of machine translation (MT) services. And at around the same time, the company also released what it claims to be the first research quantitatively measuring our speed toward the technological singularity — at least when it comes to MT.

MultiLingual caught up with the company’s CEO, Marco Trombetti, to learn a bit more about why the IDC ranked Translated so highly and about what the singularity could mean for professionals in the language services industry.

Editor’s note: This interview has been edited for clarity and conciseness.

MultiLingual: To start off, could you tell me a bit about your career path and how you got started in the language services industry?

Marco Trombetti: I’m in this industry, because of love — and not love for language, or for technology — but because I met Isabelle [Andrieu], my co-founder and wife, and I fell in love with her when I was studying in France. And basically, she was the linguist, and I was the computer scientist. So in 1999, we decided to do something together, and we wanted to spend more time together. So we decided to found one of the first purely internet-based translation services. And that’s how I got into it.

ML: You guys got your start in 1999, and have obviously come a long way since then. The IDC recently placed Translated as a leading vendor of machine translation (MT) on its Worldwide Machine Translation Software Vendor Assessment — how did that development come about?

MT: So first, let me say, I knew that the team was working super hard and for many years, now, they have done something incredible. But honestly, when I received the reports, I had some tears — it wasn’t expected. The competition was very big [IDC also rated the MT capabilities and strategies of companies such as Google, Microsoft, and Baidu, among others]. 

Capabilities and strategies were the main drivers. I don’t know the details of what they considered to be the winning factor for us, but I know what made their eyes sparkle a little bit. For capabilities, obviously, it’s about the number of languages you support, the quality of delivery, the cost, and latency. Nobody mentions latency, but most of the applications of MT today are latency-driven — you need very quick response time. And then there was also the amount of integrations you add in different scenarios for professional translators, consumers, enterprises, and small to medium enterprises, language service providers, the different solutions, and the different kinds of models that those customers were using. 

I think that the killing factor was adaptation. We invented this approach a long time ago — we started adaptive MT with Matecat, and we got the TAUS Game Changers Innovation Award in 2015. We’ve been working on adaptive MT for many years. The two things they really loved was the idea of being able to fix mistakes on a corporate website, without having to retrain the model and without waiting one year. So you find the big error, you go there, you fix it, and then you propagate the correction across the whole website. So I think they understood very well that adaptive MT gives no shortage of advantages. 

And for strategies, I think we were very different from the others, who all had the same simple strategy: better models, more data. That has been the strategy for the last 20 years. No more computations — basically brute force. And instead, they understood, what we have done with human translators, professionals, as leaders for the last 15 years in designing this symbiotic approach, basically, is generating the data, high quality data, from the corrections of the MTs. So the human machine will suggest to a human, the human will correct the mistake. And then basically, the model will improve.

ML: And in addition to this development with IDC, your team recently released a report on how close we are to reaching the singularity — can you tell me a little bit about what that means? 

MT: Everybody’s talking about singularity in AI — the point where machines are more intelligent than humans. We define that language singularity as the point where a professional human translator is preferring a suggestion coming from an MT, as if it’s something done by a professional translator in perfect context. 

When we hit that point, everything will be different in our industry. So the study we made basically started to predict the speed at which we are approaching singularity. We did it by collecting for over ten years, the time translators take to correct machine translations. We saw that this time was decreasing every single month, on a continuous linear trend down. It’s hard to make a prediction of when it’ll happen, but if you see the graph, you get your own idea that it will be quite soon. 

And maybe while we’re approaching it, it will be more complex, maybe it will take more years. But I think it’s something that we will see in our lives now, and I think it will be one of the biggest changes for humanity ever. Up until the singularity, humans have created wealth by time or by word. Their time converts to wealth. When we reach singularity, we’ll be able to create wealth via energy — our wealth will be connected to the amount of energy that we can put into the machine. That is a big change.

So the study is about asking, “When is this coming?” Because I want to prepare — I want to make sure that we can prepare our linguists, and so that all the people that are around us can prepare along with us. 

ML: You mentioned that it will change a lot for humanity — can you expand more on what you mean by that? And more specifically, what will it change for translators and others who work in the industry? How will that change the nature of their work?

MT: One hour of their work does not convert into 500 words — it converts into billions of words. And that is a nice change that is creating incredibly interesting opportunities for linguists. The amount of wealth per hour that can be created by helping a machine that can scale — that’s one important first step. 

And second, I think that once we’re approaching the singularity, translators will move more into cultural mediation, rather than language per se. By definition, singularity means that the machine will be able to produce language that is better than what a single human can do. But what about emotions? What about cultural differences that need to be mediated? 

That’s another big opportunity. The level of multilingual communication will generate ten times the amount of work for cultural mediators. And so our linguists — the ones that will have the biggest opportunities — understand the two cultures, they understand language, and they can help the machines. At the same time, they’ll be able to take care of creativity, emotions, and cultural mediations.