Improved accuracy for CEFR level detection

The Common European Framework of Reference (CEFR) is a worldwide standard to measure language proficiency. The framework is organised into six levels, A1 to C2, which can be regrouped into three broad levels: Basic User, Independent User and Proficient User. The levels are defined through ‘can-do’ descriptors.

Gradingly’s AI Grading Engine is capable of determining the CEFR level based on writing submissions from English language learners. In our latest update, we have improved the accuracy of our systems to more precisely place students on the correct CEFR level using a variety of new AI algorithms. 

We trained our updated model on tens of thousands of essay submissions of various types, user levels and task types. In addition to this, we benchmarked the results against academic linguistic specialists as well as other automatic CEFR detection tools. 

The latest results show that our updated AI Grading Engine is closely aligned with human-level marking and outperformed other leading automatic detection tools on CEFR placement based on text.