Work

ANLPers at BAREC Shared Task 2025: Readability of Embeddings Training Neural Readability Classifiers on the BAREC Corpus

Year: 2025

Type: conference-paper

Abstract: This paper presents a neural approach to Arabic readability assessment using the BAREC corpus for fine-grained classification across 19 readability levels.Our two-stage system combines embeddings from multiple pre-trained Arabic transformer models (ARBERTv2, MARBERTv2, AraBERT) with a Multi-Layer Perceptron classifier.We achieve competitive performance with Quadratic Weighted Kappa scores of 73.00-76.35,accuracy of 44.73%, and adjacent accuracy of 61.40%, within 8% of baseline models.The system offers significant practical advantages including rapid training time (10 minutes per experiment), compact architecture (12-15 million parameters), and efficient inference, making it suitable for resourceconstrained deployment.Our analysis identifies dataset quality challenges including inconsistent diacritization and annotation issues that impact performance.This work provides a foundation for practical Arabic readability assessment tools in educational applications.

Source: Arabic Natural Language Processing Workshop (WANLP)

DOI: 10.18653/v1/2025.arabicnlp-sharedtasks.38

Authors: Serry Sibaee, Omer Nacar, Yasser Alhabashi, Adel Ammar, Wadii Boulila

Institutions: Prince Sultan University, Tuwaiq Academy

Language: English

Topic: Text Readability and Simplification

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Funder: Prince Sultan University

OpenAlex ID: W4416037796