MorphoArabia at BAREC Shared Task 2025: A Hybrid Architecture with Morphological Analysis for Arabic Readability Assessment

Résumé

This paper presents MorphoArabia, a system developed for the BAREC Shared Task 2025 on Arabic Readability Assessment.The approach is centered on the hypothesis that deep morphological analysis is fundamental for modeling the complexity of the Arabic language.A regression model was fine-tuned on AraBERTv2 with morphologically-aware tokenization via CAMeL Tools.Various configurations were explored for the strict, constrained, and open tracks, including a hybrid model with seven engineered lexical features.The system demonstrated highly competitive performance, securing top-10 rankings in all six subtasks and achieving a peak Quadratic Weighted Kappa (QWK) of 84.2% on the strict sentencelevel task.All code and models are publicly available to facilitate future research.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Voir l'article sur le site de la revue

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0