EECS500 Fall 2015 Department Colloquium

Burr Settles
Duolingo: Improving Language Learning and Assessment with Data
White 411
October 22, 2015

Duolingo is an online education platform with more than 100 million students worldwide. Our free flagship learning app is the #1 way to learn a language online, and is the most-downloaded education app for both Android and iOS devices. In this talk, I will discuss several ways we mix machine learning with computational linguistics and psychometrics to power Duolingo Test Center — a companion app launched in 2014 to make language certification accessible to anyone in the world with an Internet-enabled device. Corpus-driven machine learning allowed us to develop a state-of-the-art, interactive, computer-adaptive English exam in less than six months... with little human involvement! Recent research shows that Duolingo scores are significantly correlated with both TOEFL and IELTS exams (commonly used for admission to English-language universities, but at 10 times the cost). Duolingo Certificates are currently used by Novell, Upwork, Uber, the Government of Colombia, and many other institutions to assess English language ability.


Burr Settles is a research scientist and software engineer at Duolingo, where he has worked on numerous projects involving machine learning and natural language processing; in particular he developed Duolingo Test Center, which was named Google Play's "Best of the Best" new app in 2014. He also runs FAWM.ORG, an online collaborative songwriting experiment. Previously, Burr was a postdoc in machine learning at Carnegie Mellon University — where he worked on "interactive machine reading" as part of the Google/DARPA/NSF funded NELL project — and earned a PhD in computer sciences from the University of Wisconsin-Madison. His book Active Learning (an introduction to adaptive/curious learning algorithms) was published by Morgan & Claypool in 2012. He gets around by bike, prefers sandals to shoes, and plays guitar in the pop band Delicious Pastries.