• Nguyen Thi Dan Ta

  Author

Automated language assessment solutions are progressively being integrated into various language tests and instructional structures. The majority of these solutions rest on the deep learning framework, which provides high predictive accuracy, but often regarded as "black boxes," giving very little explanation concerning how a certain score has been obtained or a certain judgment has been made. Consequently, this may create problems for students, teachers, and developers of assessments, especially when a result obtained is of considerable impact on the learner's placement, certification, or academic progress. Explainable AI (XAI) presents a new approach aimed at solving the problem. XAI provides a set of techniques which allow to make the process of obtaining the score understandable for humans without affecting the prediction accuracy. This paper deals with the topic of the interrelation of XAI and automated language assessment. The author presents a classification of the available methods, based on the methods used, their location, and the audiences. In addition, the paper provides an overview of the existing techniques used in essay evaluation, short answer grading, and speech assessment. The aspect of validity, trust, and involvement of students in receiving feedback is also considered. The paper proposes his vision for further research in the field of construction of a reliable, transparent, and pedagogically effective automated language assessment system.

• PDF