,
Professor, The Academy of Public Policy and Administration under the President of the Republic of Uzbekistan, Tashkent, Uzbekistan
,
Namangan State University, Namangan, Uzbekistan
,
Lecturer, Applied Foreign Languages Department, Namangan State University, Namangan, Uzbekistan
,
Namangan State University, Namangan state university, Namangan, Uzbekistan
,
Vice Rector for International Relations, Namangan state university, Namangan, Uzbekistan
Lecturer, Department of Philology, Turan International University, Namangan, Uzbekistan
The current teaching method of foreign languages is more and more adopting the method of using intelligent software to observe the learner's behavior and adapt teaching methods accordingly, but the teaching software currently used in the field of English learning still has many limitations such as relying on rigid rule sets or single-pass learning prediction models which cannot track and adapt the learners' learning competence and motivation from one teaching moment to the next. This paper introduces a self-adaptive cognitive AI agent model framework which continuously models the learners' behaviour and then uses a perception-reason-plan-act cognitive loop to select tasks, plan tasks, and make tasks more or less difficult for the learners in reading, writing, listening, and speaking. The framework combines a continuous learner behaviour modelling module with a skill-domain model based on the CEFR framework and a module on self-evaluation which recalibrates the adaptation policy based on observed outcomes. The framework's main behaviour-modelling premise – that the signals of interaction that are continuously captured better predict learner outcomes than profile information – is tested empirically using the Open University Learning Analytics Dataset (OULAD), a real, publicly accessible, anonymized collection of 32,593 course enrollments and 173,912 graded submissions. The design premise for the framework is that early, continuously updated behavioural signals can drive timely adaptive intervention, and logistic regression and random forest classifiers based on behavioural features (completion rate, submission timing and score trajectory) exhibit an AUC of 0.958 and 0.965 respectively, predicting at-risk outcomes with materially the same accuracy as behaviour-plus-demographics, and a variant using only the first three submissions reaches AUC 0.853, supporting this premise. The results of these empirical studies are then applied to the proposed architecture for developing English skills, and a roadmap for implementing the architecture and further domain-specific validation before classroom deployment is then described.
This is an open access article distributed under the Creative Commons Attribution Non-Commercial License (CC BY-NC) License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The statements, opinions and data contained in the journal are solely those of the individual authors and contributors and not of the publisher and the editor(s). We stay neutral with regard to jurisdictional claims in published maps and institutional affiliations.
