Game for new languages?

Teaching a disparate bunch of people an eclectic mix of foreign languages was never going to be easy, but the founders of online learning platform Duolingo were sure they’d manage. The premise was simple: Use a speaker’s native language to teach the foreign language. So, there are English lessons for Ukrainian, Turkish, and recently, Hindi, Japanese, or Mandarin speakers.

It’s coincidence that barely three months after the Hindi service was launched, the Indian government sparked off a debate over whether Hindi or English should be the official language. Duolingo’s founders, human computation scientist Luis von Ahn and chief technology officer Severin Hacker, are uninterested in the politics of the language debate. What interests them is the fact that India has over 400 million Hindi speakers, and a good proportion of that population wants to learn English.

Launched in June 2012, Duolingo initially offered only European languages. But as users from across the globe began expressing interest in learning languages, the founders expanded into teaching in languages with vastly different scripts and grammars. Today, it has 30 million registered users. Since inception, Duolingo has raised $38 million (Rs 228.6 crore) in capital.

Here’s the thing: Duolingo is entirely free (unlike other online language learning tools such as Rosetta Stone, which charge up to $1,000 for a course). So, how does it make money? By selling translations. Duolingo’s students practice by translating text from websites or blogs. This translated content is then sold to customers such as CNN. Everyone wins: the students get practice, the clients get translations, and Duolingo gets paid.

Apart from being free, Duolingo’s big differentiator is in how it teaches. Online education tends to replicate classroom models, but online users respond differently. “It’s better to come up with something that is native to the Internet,” says von Ahn. “It should feel more like a game. That’s what education on the Internet will look like in future.”

It works like this. At first, all users are given the same type of exercises. Over time, Duolingo learns what a user is good at, and figures out the best lesson for that person. It tries new things on a fraction of users to validate performance. For example, should plurals be taught first or adjectives? “We teach one half plurals before adjectives, and the other adjectives first, and track both,” says von Ahn. Depending on which sample fares better, Duolingo decides what to teach first.

Von Ahn is no stranger to the Internet. In 2000, he co-founded Captcha, now a standard response test where users type a word while logging in to prove they are human. His team went on to create reCaptcha: While users are authenticating themselves as human, they are also helping digitise books. Typically, digitisation projects scan pages of books, which a program then deciphers. It fails to recognise some words, and those are the words that reCaptcha gives Captcha users. This has helped in digitising 2 million books a year.

Now, von Ahn wants to improve the completion rate of online language courses, currently around 2%. Offline courses fare better because of parental or faculty pressures. He hopes Duolingo’s Internet-focussed approach will change things.