Work

MarsadLab at BAREC Shared Task 2025: Strict-Track Readability Prediction with Specialized AraBERT Models on BAREC

Year: 2025

Type: conference-paper

Abstract: The BAREC 2025 Shared Task on Arabic readability targets 19 levels of ordinal prediction at the sentence and document levels under strict training.This paper describes a two stages system that basically starts with BAREC-tuned AraBERT checkpoints and then specializes on the Strict splits with Weighted Kappa Loss (WKL), an objective aligned with Quadratic Weighted Kappa (QWK).A single architecture with inputs specific to each track is utilized for both tracks.On the Strict setting, our best systems reach 0.842/0.841QWK (public/blind) at the sentence level and 0.828/0.790QWK at the document level.

Source: Arabic Natural Language Processing Workshop (WANLP)

DOI: 10.18653/v1/2025.arabicnlp-sharedtasks.39

Authors: Shimaa Ibrahim, Md. Rafiul Biswas, Mabrouka Bessghaier, Wajdi Zaghouani

Institution: Northwestern University in Qatar

Language: English

Topic: Text Readability and Simplification

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

OpenAlex ID: W4416037817

FWCI:

Cites:

Cited by:

Related to: 10