Skip to Content

Getting Fluent

Can an algorithm teach us language?

Since it launched in 2012, Duolingo has been studying every move made by the 15 million–plus people who use its website and mobile app every month to study foreign languages. The company is run by technologists—it hired its first learning expert this year—and constantly tests different ideas, looking for trends in the data.

Its current goal: helping students become fluent.

As measured by the Common European Framework of Reference for Languages, Duolingo can at present get students, at best, to the point where they can express themselves in familiar situations, or the B1 level. Full mastery is three levels higher, at C2.

No software has yet been able to get students to fluency, but Duolingo believes it can achieve this goal by hiring more learning experts and studying the data, says cofounder Severin Hacker. Its language lessons are free, but the company earns money by selling $20 certificates attesting to student performance on a 20-minute English test, the results of which it says correlate with those of more established, and expensive, options including the Test of English as a Foreign Language, or TOEFL. Duolingo has raised $83 million in financing from Google Capital, Union Square Ventures, Kleiner Perkins Caufield & Byers, and others.

The company has already drawn from its constant experimentation to change many features and create new ones. A team of machine-learning specialists is fine-tuning the way its algorithm determines what students know, what they should study next, and what messages best motivate them to keep learning, delivering an adaptable, personalized learning experience.

Duolingo feels more like a game than like traditional language learning. Originally it would award students three hearts at the start of a 20-exercise lesson. Every mistake cost the user one heart. Lose all three, and you had to repeat the section.

After experimenting with three, four, and five hearts, Duolingo determined that four was the best motivator for new students. More advanced students did best with three and would still see that number. Since those experiments, the company has introduced a more sophisticated strength bar that takes into account nuances such as the relative importance of different issues. For example, it incorporates the idea that getting the gender of a word right is not as important as getting the word itself correct.

A year ago Duolingo began another experiment with a digital coach, an owl avatar that tracks goals students set for themselves and then tells them if they fall off track.

“It’s thousands of little experiments we run,” Hacker says. “Some are really good, but really it’s the combination of thousands of things that makes the difference.” The average time a student continues using Duolingo has more than doubled since its launch, says Hacker.

Greg Smolinski, a teacher who incorporates Duolingo into introductory German classes for seventh and eighth graders at Seneca Valley Middle School in Pennsylvania, says some of his most motivated students have gotten to the B1 level using the program on their own, effectively replacing hundreds of hours of classroom time with work done independently and online.

Language experts are not sold on the idea that a program like Duolingo can get a student to fluency. “I do not think that a language platform can replicate the face-to-face communicative interaction that is so important in language learning,” says Véronique Baloup-Kovalenko, who teaches French at the Convent of the Sacred Heart in Greenwich, Connecticut. Nonetheless, Baloup-Kovalenko appreciates Duolingo’s ability to tap into students’ competitive spirit and sustain their interest. The company quotes her praise on its pages for educators.

What People Study on Duolingo
Duolingo’s top markets:

1. United States: English speakers studying various foreign languages
2. Brazil: Portuguese speakers studying English
3. Mexico: Spanish speakers studying English
4. China: Chinese speakers studying English
5. Colombia: Spanish speakers studying English

Keep Reading

Most Popular

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

The problem with plug-in hybrids? Their drivers.

Plug-in hybrids are often sold as a transition to EVs, but new data from Europe shows we’re still underestimating the emissions they produce.

Google DeepMind’s new generative model makes Super Mario–like games from scratch

Genie learns how to control games by watching hours and hours of video. It could help train next-gen robots too.

How scientists traced a mysterious covid case back to six toilets

When wastewater surveillance turns into a hunt for a single infected individual, the ethics get tricky.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.