Year: 2025
Type: conference-paper
Abstract: Culture fundamentally shapes human perception and reasoning, while religion-often embedded within cultural contexts-provides cohesive moral frameworks and a sense of community.The PalmX 2025 shared task introduced two subtasks aimed at evaluating the capability of large language models (LLMs) to capture and represent culturally and Islamically grounded knowledge.In this paper, we present our participation in this shared task, leveraging parameter-efficient fine-tuning (PEFT) techniques in conjunction with targeted data augmentation strategies.We further conducted extensive zero-shot evaluations across a range of Arabic-centric and multilingual models to establish strong baselines and guide model selection.Our submitted system achieved competitive performance on the blind test sets, ranking 3 rd in Subtask 1 with an accuracy of 71.45% and 1 st in Subtask 2 with an accuracy of 84.22%.
Authors: Jannatul Tajrin, Bir Ballav Roy, Firoj Alam
Institution: BRAC University
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037604