A Systematic Literature Review on Automatic Correction for Arabic Text Using Language Models and Deep Learning

Abstract

Automatic correction for Arabic text is challenging due to Arabic's complex morphology and orthographic rules, similar to the difficulties encountered in low resource languages. This systematic literature review investigates developments in Arabic text correction from 2015 to 2025, with a particular emphasis on methods leveraging language models and deep learning. The present advancements in Arabic grammatical error correction is represented by the AraBART model. It achieved an F1 score of 92.1 % when evaluated on the QALB-2014 dataset. For grammatical error detection (GED), a Bi-LSTM-based system attained an Fmeasure of 95.19%, while spelling correction systems reached an auto-correction rate of 96 % using a hybrid approach combining edit distance and probabilistic language models. In addition to benchmarking performance, this review highlights key limitations such as data scarcity, the need for high-quality annotated corpora, and the handling of contextually ambiguous errors. Expected outcomes include the identification of research gaps and recommendations for future work, particularly in developing more robust models and richer datasets to advance automatic Arabic text correction.