Correlation between an Entropy Based Measure and English Language Learner Proficiency

It is important for education systems to analyze and provide an appropriate level of support to meet the needs of learners. An example of this is how the effectiveness of automatic language learner error detection and correction can vary depending on the learner's proficiency level. Covering a wide range of language complexity makes the task of error detection difficult. By predicting the learner's proficiency level, different error models can be applied for different proficiency levels. In this paper, we propose a measure based on the frequency of words in the sentences produced by learners during speaking exams to predict the learner's language proficiency. The proposed measure is compared to the learner's vocabulary size by correlation analysis. The results suggest that there is a stronger correlation between the proposed measure and the proficiency of the learner than the learner's vocabulary size.