Official implementation of the ACL 2025 paper "Assessing Gender Bias of Pretrained Bangla Language Models in STEM and SHAPE Fields"

This repository contains code for HerWILL STEM Research team's research paper titled "Assessing Gender Bias of Pretrained Bangla Language Models in STEM and SHAPE Fields" accepted at the GeBNLP workshop under ACL 2025 (A* conference).

You can find our dataset on mendeley data.

https://data.mendeley.com/datasets/y3x569kk9t/2

Give our paper a read at https://aclanthology.org/2025.gebnlp-1.24/

Usage

It is recommended to open a virtual env. Install the requirements

pip install -r requirements.txt

Install normalizer

pip install git+https://github.com/csebuetnlp/normalizer

run the script

python  analyze_bias.py --modelname "csebuetnlp/banglabert"

Citation

If you use our code/ ideas from our paper, please cite

@inproceedings{arnob-etal-2025-assessing,
    title = "Assessing Gender Bias of Pretrained {B}angla Language Models in {STEM} and {SHAPE} Fields",
    author = "Arnob, Noor Mairukh Khan  and
      Mahmud, Saiyara  and
      Wasi, Azmine Toushik",
    editor = "Fale{\'n}ska, Agnieszka  and
      Basta, Christine  and
      Costa-juss{\`a}, Marta  and
      Sta{\'n}czak, Karolina  and
      Nozza, Debora",
    booktitle = "Proceedings of the 6th Workshop on Gender Bias in Natural Language Processing (GeBNLP)",
    month = aug,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.gebnlp-1.24/",
    pages = "268--281",
    ISBN = "979-8-89176-277-0",