MarsadLab at TAQEEM 2025: Prompt-Aware Lexicon-Enhanced Transformer for Arabic Automated Essay Scoring
Year: 2025
Type: conference-paper
Abstract: We present the MarsadLab submission to TAQEEM 2025 Shared Task A on Automated Essay Scoring (AES) in Arabic.Our system extends AraBERT with a prompt-type embedding and lexicon-based features.The lexicon captures statistical associations between word usage and essay quality under each prompt type, providing prompt-aware, interpretable signals that complement semantic embeddings.Our system achieved an average QWK of 0.438, highlighting both the promise and the challenges of incorporating prompt-sensitive lexical knowledge into AES.This work represents a first attempt at leveraging a task-aware lexicon for Arabic AES, showing that lexical features provide educational value through interpretability but also require more sophisticated integration.Future improvements could combine these lexical indicators with discourse-, syntax-, and content-level features, as well as explore richer fusion strategies to better exploit their potential.
Institutions: Northwestern University in Qatar, University of Kairouan
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037583