PERSON OF THE MONTH
Mait Müntel
Think of any CERN scientist involved in breakthrough technology… who isn’t Tim BernersLee. That leaves Mait Müntel, the brains behind language app Lingvist and the multilingual subject of this month’s Profile.
When they started building the Hadron Collider at CERN, it’s probably fair to say nobody was quite sure what would be discovered. But nobody’s shortlist of potential breakthroughs included an app that would rapidly accelerate the learning of foreign languages.
That’s precisely what Mait Müntel emerged with after his time in Geneva. The Estonian wasn’t only part of the team that discovered the Higgs boson particle; he also invented a way to teach himself French, in the scarce spare time available to someone helping to make one of the most fundamental scientific discoveries of the 21st century.
Müntel’s method is now a free-to-use app called Lingvist. But how did the Hadron Collider help him turn learning languages into a solvable mathematical problem?
Movie talk
Müntel had been working and living in Geneva for a few years, before it suddenly dawned on him how “impolite” it had been to live in the city without learning its language. At CERN, everyone spoke English – or what Müntel labels “Scientist’s English”. “It’s a bad English because there are so many different nationalities involved that have very, very weird accents,” he said.
Müntel had learnt English as a child, but it had taken him a decade to master it. He couldn’t wait another ten years to learn French, so he decided to find ways in which he could learn more quickly. “I was curious about the potential for computational networks to accelerate learning,” he said.
He decided to treat it as a maths problem, calculating the minimum amount of words he would need to learn to speak French fluently. “I wanted to know which words and phrases people are actually using,” he said. He couldn’t get that from books, “because they’re not based on actual language, they’re based on traditions, or how teachers are used to learning each language”.
Instead, Müntel went to the movies. He downloaded the French subtitles from around 30,000 films and began analysing their vocabulary. He also downloaded the English subtitles so he had a direct translation, second for second, of the French vocab. He started memorising words and phrases, and calculated that he could learn 98% of the vocabulary in less than 200 hours. That was so much faster than the ten years it took him to learn English that he “started to write a programme for myself on how to optimise the learning process”.
He began measuring the efficiency of his memory, based on filling in missing words in common French phrases, and working out which words he had to review and when. Before long, he’d built a prototype of the app that has today become Lingvist.
Turning a hobby into a company
Then came the really difficult bit: turning his pet project into a business. “I had no idea whatsoever how to build a company,” Müntel admits. “I had a couple of people to get things going, and then I realised ‘bloody hell, now I have people I have to pay a salary, and I have to raise money,’ and a hobby went out of control.” As Müntel was beginning to get the business off the ground, he had the “happy accident” of bumping into one of the founders of Skype, Jaan Tallinn. “He was learning Japanese and I was learning French through our computers and smartphones,” he said. “I started talking to him, asking him what he was using. He said that he was trying to develop some software to make people learn faster. I said ‘Oh, cool, I’ve just made something similar as well’, and then we spent the day together sharing each other’s experiences.” Tallinn suggested Müntel set up a company to develop his app, later becoming the first investor in Lingvist.
Meanwhile, Tallinn’s own app hit something of a brick wall. “Later I learnt that, on the same day [we met] he stopped development [of his own app] because he thought mine was so much more advanced.”
Lingvist wasn’t only backed by angel investors. In May 2015, the company received a €1.5 million grant from the EU’s Horizon 2020 programme, allowing it to employ more staff, refine the app’s features and test Müntel’s claims that Lingvist accelerated the speed of learning. “We did some testing with universities to prove that we can really teach faster than anybody else, and the tests were very successful,” he said. “We did two [tests] in the end, because we believe it’s possible to make people learn way faster – ten times faster – than they learn in a classroom, with the use of technology.”
Broadening horizons
An app that could only help English speakers learn French would have limited appeal, but Müntel found that the adaptive algorithm and techniques could also be applied to other language pairs. “The principles are the same,” he said. “If we take European languages, they’re actually pretty similar. We have to take into account the differences which stem from grammar and the way people use language, but the algorithmic part of the software is quite the same.”
There’s machine learning involved, based on how well you cope with different areas of grammar or vocabulary, so the system can help you learn, say, Spanish more quickly because it knows which parts of French you struggled with. “Different languages share similarities vocabularywise,” said Müntel. “Having this machine-learning-based system, which can collect data from all the other users as well, we can make the learning way more effective.”
The course now has 16 different pairs, including English to Russian, German and Spanish (and vice versa), as well as Arabic, Japanese and Müntel’s mother tongue, Estonian, to English. Lingvist has apps for Android and iOS, and can be accessed via web browsers, so you can brush up your skills from almost any device. And the mathematical principles applied to the learning process are evident in the interface: having spent over three hours brushing up on my French, I’m told I’ve learnt 352 words, or 48% of the words “found in any text”. If that sounds an awfully small number of words for half the language, bear in mind that the learning curve is not linear. I’ll need to master 5,000 words to have learnt 90% of the words in any given text. Zut alors!
Money spinner?
Three hours into my French course, I notice I’ve yet to be charged a euro for this multi-platform course, unlike other language apps I’ve tried over the years. Müntel says the company has yet to start monetising the product, which is surprising given that you make quick progress, and that the firm must be incurring some serious overheads.
Lingvist employs 37 staff, including linguists, designers, product managers, a data team and software developers, several of whom used to work for Skype. The company also has offices in London, Tallinn and Tokyo, and is working on additions to the package, including voice recognition.
The company is testing monetisation schemes with Japanese customers, where there is a huge demand to learn English. And if you add a second language course, the software warns: “You can pick as many languages as you like for free at the moment. This may change in the future.”
What you learn may also change. Müntel believes the technology might have applications beyond languages. “Machine learning is disrupting the whole world,” he said. “The big differentiator of human people is how fast they can actually learn. Artificial-intelligence applications will take a lot of jobs that people can do today. So learning gets more and more important, and the speed of learning gets more and more important.
“We developed this technology for language learning, but there are dependencies to other learning as well. The best educational materials are English-based. This is a first step to improve learning in general.”
It’s possible to make people learn faster – ten times faster – than they learn in a classroom, with the use of technology