There’s a lot of material on the web and a good portion of it is in a language that you can’t read. If you are an English speaker this is a relatively small problem but as more and more people get online, the ratios get worse–more people will be creating more content in more of languages, which means a greater chance that you are missing out.
There are two basic solutions to this problem. First, you can learn some other language, thus expanding your pool of readable content. Second, you can arrange to have the Internet translated. Neither of these are easy solutions. Learning a new language is hard and there is a lot of a data on the Internet.
Duolingo uses both these solutions at once. How it works: You pick a language that you want to learn (Duolingo offers Spanish, German, and English for now, with more to come). The program takes you through some initial training and then it pulls a sentence from somewhere on the Internet and asks you to translate it. If you do a good job, then your new sentence gets added to the Duolingo database, as a valid translation for the sentence in question.
“The whole idea was to translate the web into every language,” says Luis von Ahn, founder of Dulingo. “We thought we needed computers but it turns out that [machine translation] won’t be any good for 20 years.” So they decided they needed humans. You need a lot of humans to make it work, and you can’t pay them because of scaling issues–it’s just not affordable. The question became how to make it possible.
The payment for translating sentences is that you get to learn a language through everyday drills mixed with lessons. Duolingo has enough intelligence to figure out the rough difficulty level of a sentence, but it hands the reigns over to you to make that sentence work. It aims to give you sentences that are appropriate to your skill level. How does it know?
“It turns out that there are a few ways to do it, ” says von Ahn. “A really big indicator of how hard a sentence is, is the length of the sentence.” Beyond length, each language has its own indicators of difficulty. For example, when dative appears in Spanish, that indicates a difficult sentence. For German, the indicators are different. Duolingo discovered these indicators by classifying some sentences by hand and looking for commonalities in the data. This allows the program to rate the challenge of a particular piece of text, without needing to be able to understand or translate that text.
The interface of Duolingo takes a lot of its inspiration from the latest in theory around education and using games to motivate people. The main index page is made up of a series of unlockable skill areas. As you progress, you open up new areas for lessons and translation exercises. The branching paths allow people to pick and choose their areas of emphasis in their learning. The unlocking mechanism ensures that everyone learns the basics.
The exercises display a similar game-like feel. Each one has around 20 questions. You are given four hearts. Each wrong question costs you some life and if you fail, you must start over. “Four hearts is about right,” says von Ahn. “With three it’s too hard,” adds lead designer Marcel Uekermann. “I finish with empty hearts most of the time.” With four, users can finish most of the time, “which is what you want out of a game.” This is an element of adding game elements to your app that so many gamification experts seem to miss–for victory to taste sweet, defeat must be a possibility.
Beyond the rich interface for learning the skills, Duolingo’s featureset remains narrowly focused. “The overall goal has been trying to keep things as simple as possible. It’s tough because we’re trying to get people to do something pretty complicated,” says von Ahn. “If you look at other language learning sites, they usually boast of features,” says Uekermann, “They have forums, livechats, or whatever. We reduced it to a minimum to focus on the stuff that really matters.”
In order to ensure that they are focusing on the stuff that really matters, von Ahn says, they rely on extensive user testing. “It’s been amazing how bad we are at doing the first draft every time.” They’ve persisted, learning lessons in interface that you can only learn by helplessly watching a tester completely miss the interface element that seems unbelievably obvious to you, the creator. Over time, the feedback has improved six months ago, people were complaining that the system was a lot of work. Today, people are starting to have fun with it. Which is good, because there’s a lot of web to translate.