Unlocking the Potential of Adaptive Learning for Spelling Acquisition: A Machine Learning Based Approach in a Large-Scale Experiment

BackgroundOnline learning in school contexts increased, partly due to the Covid-19 pandemic. Further, adaptive technologies that adapt a learning environment dynamically to the users’ needs have shown to be effective. However, adaptive learning comes in many shapes and sizes and depends on the specific implementation, learning context, and user group. ObjectivesThis article presents a large-scale online-controlled experiment that compares different machine-learning based adaptive interventions in an online learning environment specifically for German spelling.MethodsThe randomized online experiment included more than 11,000 students and more than 959,000 answered spelling tasks. During the experimental period, all users were grouped into one of five intervention groups or the control group. The effectiveness of the interventions was evaluated with regard to the error rate, dropouts, and user competency.Results and ConclusionsBoth the student-facing interventions and the task sequencing interventions have a positive impact on the error rate. The evaluation of the number of session dropouts has shown poor fit and yielded mixed results. In the development of spelling competence, measured with the Rasch model, the task sequencing interventions show to double students’ skill growth compared to the control group.Major takeawaysAdaptive learning can improve users’ learning experience in online first language learning of the German language. Task sequencing interventions show the biggest improvements in terms of error rate and development over time. The evaluation of the number of dropouts yielded mixed results and the suitability of dropout as a measure needs to be discussed.