Traditional cognitive diagnostic models in language learning are typically static and primarily designed for evaluative scenarios. In contrast, the Bayesian Knowledge Tracing (BKT) model is suitable for dynamic prediction, offering a deeper understanding of learners’ cognitive development. Recognized as one of the pioneering student modeling approaches based on machine learning, BKT is extensively studied due to its interpretability and ability to infer latent student knowledge states. We have constructed a BKT model explicitly tailored for language learning on the Duolingo dataset, which includes six languages: German, English, Spanish, French, Italian, and Portuguese. Our study evaluates the BKT model's cognitive diagnostic capabilities across these languages, focusing on its predictive accuracy. The results demonstrate that BKT provides meaningful insights into the challenges faced by learners, highlighting its potential to support personalized language instruction and improve learning outcomes.