Duolingo uses an AI algorithm called Half-Life Regression to predict when a learner is about to forget a word or skill, then schedules a review exercise just before that happens. It works like a smart calendar for your memory — timing reminders to keep knowledge fresh without over-drilling what you already know. The system was published as a peer-reviewed research paper in 2016 and has been in production use since.

Developed by Duolingo researcher Burr Settles and published at the Association for Computational Linguistics (ACL) conference in 2016, Half-Life Regression uses a mathematical model of memory decay (p = 2^(-Δ/h), where Δ is time since last practice and h is the estimated half-life of a memory) to predict recall probability for each word in a learner's vocabulary. It accounts for features like word frequency, morphological complexity, and whether a word is a cognate in the learner's native language. A/B testing showed a 9.5% increase in retention for practice sessions and a 12% increase in overall activity compared to the simpler Leitner card-sorting method it replaced. Duolingo open-sourced the algorithm and released a dataset of 13 million learning traces on Harvard Dataverse.