DEVELOPMENT OF AN AI-BASED MODEL FOR DETECTING ERRORS IN HANDWRITTEN ARABIC LETTERS FOR LANGUAGE LEARNING
Keywords:
Arabic handwriting recognition, educational AI, deep learning, error detection, computer vision, second-language acquisition.Abstract
Automated evaluation of handwritten exercises is essential for scaling personalized feedback in second-language (L2) Arabic acquisition. Traditional optical character recognition (OCR) systems map handwriting directly to standard character representations while explicitly ignoring minor structural deformations. In an educational context, however, identifying these deformations is critical for diagnostic feedback. This article presents a compact, educationally grounded deep learning model designed to recognize intended Arabic letters, evaluate their orthographic correctness and classify structural writing errors in isolated handwritten characters produced by L2 learners. A formal 7-category taxonomy is introduced, comprising the correct class and six error types: shape deformations, dot placement errors, proportion imbalances, baseline misalignments, incomplete strokes and connection faults. A multi-task neural network architecture is proposed to jointly optimize character classification, binary correctness determination and multi-class error diagnosis. Furthermore, an expert-annotated dataset collection protocol and an evaluation strategy based on standard classification metrics are outlined. By bridging the gap between character transcription and pedagogical analysis, the proposed model enables real-time visual feedback within intelligent tutoring systems.
References
1. Al-Badr, B., & Suen, C. Y. (2021). Survey of Arabic handwriting recognition strategies and applications. Pattern Recognition, 112, 107730.
2. Alrobai, A., Alnuaim, A., & Zakariah, M. (2021). Deep learning-based framework for offline handwritten Arabic character recognition. IEEE Access, 9, 122105–122120.
3. Al-Sarem, M., Ahsan, M. M., & Zeshan, F. (2019). Feature extraction methods for handwritten Arabic character recognition: A comparative study. Journal of Computer Science, 15(3), 341–352.
4. Djioua, M., Cheriet, M., & Suen, C. Y. (2021). Automated evaluation of handwriting quality: A pedagogical perspective. Computers & Education, 168, 104190.
5. El-Sawy, A., Hazem, M., & Loey, M. (2017). Arabic handwritten character dataset using convolutional neural network. WSEAS Transactions on Computer Research, 5, 11–19.
6. Zhang, X. Y., Yin, F., Zhang, Y. M., Liu, C. L., & Bengio, Y. (2020). Drawing and writing with deep generative models: A survey on handwriting analysis and synthesis. IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(11), 2753–2774.
