Published in the Proceedings of the 4th Workshop on Arabic Corpus Linguistics (WACL-4), at COLING 2025, Abu Dhabi, UAE.
The paper introduces ArabicSense, a benchmark for commonsense reasoning in Arabic with three tasks: spotting statements that do not make sense, choosing the reason why, and generating an explanation. Fine-tuning on the dataset helped: Gemma and Mistral-7B reached F1 scores of 95.5% and 94.8% on the second task, and LLaMA-3 led the generation task with a BERTScore F1 of 77.3%.
Authors: Salima Lamsiyah, Kamyar Zeinalipour, Samir El amrany, Matthias Brust, Marco Maggini, Pascal Bouvry, Christoph Schommer.
Read the paper | Code | Models and data
All the lab's language releases are listed in Language models and datasets.