Predicting the Performance of Rosetta Stone Language Learners with Individualized Models of Forgetting

Date The final copy of this thesis has been examined by the signatories, and we find that both the content and the form meet acceptable presentation standards of scholarly work in the above mentioned discipline. I explore the nature of forgetting in a corpus of 125,000 students using the Rosetta Stone R 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. Figure 1 Entity Relationship Diagram showing the organization of the Rosetta Stone R course 7 2 Screenshot example of the Rosetta Stone navigation screen. In this case, the system is making an automated suggestion that the student review the content from Unit 4 An example multiple-choice listening challenge screen from a Japanese review activity. First, the prompt sound is played. Next, the student must match the prompt sound with one of the image responses. If the student chooses correctly on his first attempt, that challenge is marked correct. This screen is composed of four such challenges , and this activity has eight screens. After the student completes a challenge, the response options are randomized to discourage obtaining an answer through viii 7 Histogram of retention intervals measured in the data set. This bimodal distribution can be attributed to two aspects of the product. One, the course allows students the freedom to repeat activities at will, to earn a better score. So, after a student completes a review activity, she is free to simply repeat it immediately after completing it to try again. The second mode of the distribution, at roughly 14 days in length is likely attributable to the design of the Adaptive Recall R. This feature will automatically schedule a review activity to be repeated two weeks after the initial attempt. Although the student has the ability to opt-out of the scheduled review, 8 Data points …