Supplementing CEFR-graded vocabulary lists for language learners by leveraging information on dictionary views, corpus frequency, part-of-speech, and polysemy

The study explores an approach to supplementing existing CEFR-graded vocabulary lists, which are often incomplete, by imputing CEFR levels for additional vocabulary items. This is achieved by analysing word-level data such as dictionary views, corpus frequency, part-of-speech, and polysemy. Using English as a test case, the study employs a variety of machine-learning models to predict CEFR levels for words not included in the initial set. The models significantly outperform a random baseline, indicating their effectiveness. The findings suggest that corpus frequency is the most influential predictor, followed by dictionary views and polysemy. The study reveals the potential of this semi-automatic approach to expand CEFR-graded word lists, making them more comprehensive and accessible for language learners. At the same time, human oversight is recommended to ensure the appropriateness of the imputed words for language learners, such as regarding the inclusion of potentially offensive terms. Future research may extend this methodology to other languages, provided that sufficient linguistic data is available.