Automated Generation of Arabic Procedural Court Orders in the Moroccan Legal Context

Automating the drafting of legal documents can reduce repetitive judicial workload and support procedural efficiency in e-Justice systems. However, Arabic legal text generation remains underexplored, particularly for Moroccan procedural court orders, which are characterized by highly structured language and recurring legal formulations. This study compares retrieval-based and transformer-based approaches for generating Moroccan procedural court orders using a corpus of 4,098 authentic orders collected from 2020 to 2025. Three approaches are evaluated: a TF-IDF-based Template Baseline, AraBART, and AraT5. A random stratified split and a stricter template-disjoint evaluation are used to assess both performance and generalization. Under the random split, the Template Baseline achieves the highest BLEU score (87.39), followed by AraBART (77.33) and AraT5 (51.88). In contrast, under the template-disjoint setting, AraBART outperforms the Template Baseline, achieving BLEU scores of 68.00 and 63.67, respectively, with a statistically significant difference (p = 0.004). These findings demonstrate that retrieval-based generation can benefit from repeated legal templates, whereas AraBART achieves better performance under the template-disjoint setting when exact templates are absent. The results highlight the importance of evaluation design and corpus structure when selecting generation approaches for standardized legal documents. A conceptual hybrid strategy combining retrieval, neural generation, and human validation is proposed for future e-Justice applications.

@article{Yassine2026,
  title     = {Automated Generation of Arabic Procedural Court Orders in the Moroccan Legal Context},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {9},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Saliha Yassine and Mohammed Essoujaa and Mustapha Esghir and Ouafaa Ibrihich},
  doi       = {10.14569/IJACSA.2026.0170913},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170913}
}