I explore the nature of forgetting in a corpus of 125,000 students using the Rosetta StoneRTM foreign-language instruction software on 48 Spanish lessons. Students are tested on a lesson after its completion and are then retested after a variable time lag. The observed power-law forgetting curves have a small temporal decay rate that varies from lesson to lesson. I obtain improved predictive accuracy of the forgetting model by augmenting it with features that encode characteristics of a student's initial study of the lesson and the activities the student engaged in between the two tests. I then analyze which features best explain individual performance, and find that using these features the augmented model can predict about 25% of the variance in an individual's score on the second test. |