Authors: Aeshah Alsughayyir 1, *, Abdullah Alshanqiti 2
Affiliations:
1College of Computer Science and Engineering, Taibah University, Madinah 42353, Saudi Arabia
2Faculty of Computer and Information Systems, Islamic University of Madinah, Madinah 42351, Saudi Arabia
Simplifying Arabic text remains a significant challenge in the field of Natural Language Understanding (NLU), making it difficult for current models to perform well. Recent studies have focused on simplifying texts with complex language structures to improve readability for both human users and other Natural Language Processing (NLP) tasks. This study addresses the challenge in the context of low-resource Arabic NLP by introducing a split-and-rephrase approach using a sequence-to-sequence switch transformer model, called ATSimST. Experiments using the ATSC dataset show that ATSimST performs better than existing advanced text generation models for Arabic. The improvements in SARI, BLEU, METEOR, and ROUGE scores demonstrate that ATSimST produces high-quality simplifications that are both semantically accurate and similar to human-written texts. These results confirm the model’s effectiveness and highlight its potential to significantly advance Arabic text simplification.
Arabic text simplification, Natural language understanding, Low-resource NLP, Transformer model, Text generation
https://doi.org/10.21833/ijaas.2025.04.009
Alsughayyir, A., & Alshanqiti, A. (2025). A light but efficient switch transformer model for Arabic text simplification. International Journal of Advanced and Applied Sciences, 12(4), 71–78. https://doi.org/10.21833/ijaas.2025.04.009