Incorporating prior practice difficulty into performance factors analysis to model Mandarin tone learning
Determining how to select items for practice is an important task for any adaptive training system. Prior work has tried to create systems for second language learners (L2) to learn Mandarin tones, but such work does not often have a well-determined algorithm to choose items for practice. In order to develop a model for adaptive practice selection, we designed an experiment and asked L2 learners on Amazon Turk to finish a series of tone learning trials. Using this data we ranked the difficulty level of the stimuli and incorporated a quadratic function of difficulty for prior successes and failures into the Performance Factors Analysis (PFA) model. Results of logistic regression were used to compute the optimal difficulty level for each tone. For the four Mandarin tones, the optimal difficulty scores were 0.86, 0.75, 0.54 and 0.60, respectively. Crossvalidated results showed that the new PFA-Difficulty model had better performance than the original PFA model. While the advantage was not large, the new model allows for clear inferences about optimal item difficulty that can be used by an adaptive training system to select items for practice.
EDM 2019 - Proceedings of the 12th International Conference on Educational Data Mining
Cao, M., Pavlik, P., & Bidelman, G. (2019). Incorporating prior practice difficulty into performance factors analysis to model Mandarin tone learning. EDM 2019 - Proceedings of the 12th International Conference on Educational Data Mining, 516-519. Retrieved from https://digitalcommons.memphis.edu/facpubs/15398