Luis Von Ahn, one of the creators and founders of Duolingo, spoke at Duke University in November 2012 as part of the Provost’s Lecture Series.
When I first saw Luis Von Ahn’s highly praised TEDx talk about translating the web through a free language-learning tool, it seemed too good to be true. Could crowdsourcing and machine learning actually work to improve web translation services? I signed up for Duolingo when it started its limited release, and set out on a quest to re-learn Spanish.
Regardless of a student’s language level, the format of the site is set up with a number of ways to track learning progress. On the homepage, there is a full “skill tree” which shows the student what topics they have covered, and what they need to learn in order to progress to the next skill. This approach lends itself to the mastery-based learning format that the Khan Academy and other learners have used to help address Benjamin Bloom’s long-standing 2 Sigma Problem. However, Duolingo doesn’t follow Khan’s voice over virtual whiteboard style. Instead, Duolingo drops learners right into the fray with the necessary vocabulary, giving audio readouts of each word along the way. A student types English and Spanish translations, and can speak words into the microphone to have their recording recognized as an answer. While learners can choose to skip ahead on the skill tree by taking a comprehensive quiz, I decided to start from the very beginning.
In an effort to motivate students towards self-learning and intuition, Duolingo’s learning platform abundantly utilizes elements of Gamification. For example, personal progress is measured in “skill points,” which are visually represented as coins. When attempting to learn a skill, a student gets three hearts, which act like three strikes. Each time the student answers incorrectly, a heart is deducted. When the student’s hearts run out, they have to re-take the test. To students of the videogame generation, this approach can seem much more effective than trying to arbitrarily compute a score to determine success or failure. Real-time feedback on performance is a great way to motivate students, and there is a bonus if students complete the skill with full hearts. Duolingo employs another motivational technique in the form of a virtual owl tutor called Duo. Duo is happy when you show up to practice on the site, and sends you emails urging you to keep up with your skills when you don’t show up. It seems silly, but Duo’s emails are effective: in my inbox I would find a small picture of a sobbing owl and a caption that said, “You made the owl cry. To keep the owl happy you should practice regularly.” As I used Duolingo each day, I found myself becoming comfortable with the vocabulary and speaking full phrases.
Once I’d developed a certain level of understanding of a topic, it was time to translate a sentence from the web. This is where things get very interesting. The student takes his or her newly learned vocabulary and attempts to translate a real-world sentence. At first, the process is heavily aided by the built-in Duolingo dictionary, which gives different translations to the student when they mouse over each word. Over time, a combination of the article’s context and my comprehension of the associated vocabulary helped to improve my translations and emboldened me to tackle more advanced sentences. But wait – how could a beginner-level student translate advanced sentences? The solution that Duolingo employs uses the power of crowdsourcing, which involves many students offering their attempts at translating individual sentences. As each student submits a sentence, they can rate others’ translations, and the most highly rated translations “rise to the top.”
Over time, entire documents are translated and students gain many skill points for their language practice. It’s easy to see how the data collected from users could be useful to improve the algorithms that underly computer translation, but what learning benefits do the humans doing the translating really gain? As an individual student, I began to understand the intricacies of how language translation and comprehension happens in the “real world,” outside of textbooks and predefined phrases. Even if learners had a dictionary to pick up and translate each word in a sentence, combinations of words and phrases can be read and understood in many different ways.
By seeing many Duolingo peers make many individual efforts in translation, I gained insight into the parts of speech that deserve more of my studying. I found myself adjusting my previous translations based on suggestions from others. In this way, the exercise in peer-rating became a gateway to informal self-assessment and improvement. This is the main success of Duolingo: it shows that improving ourselves and improving the world through technology can be a mutually inclusive activity.