Enhancing Gender Neutrality in Text using Large Language Models

Supervised by: Corina Masanti

If you are interested in this topic or have further questions, do not hesitate to contact corina.masanti@unibe.ch.

Context

As language models become increasingly integrated into applications, concerns around biases become more relevant. Many large language models (LLMs) generate text reflecting societal biases, including gender biases, which can impact the perception and representation of various groups. This thesis explores methods to enhance gender neutrality in text. On one hand, neutrality is desired, on the other, the reading process should remain smooth. This project investigates techniques to ensure gender neutrality in text.

Goal(s)

A dataset of sentences containing gender bias and their corresponding gender-neutral corrections is provided. The first step is to analyze and define the principles for applying gender neutrality. Then, some selected techniques will be explored, such as fine-tuning language models, rule-based approaches, and prompt engineering. Finally, the effectiveness of each approach will be assessed through qualitative and quantitative analyse.

Approach

• Develop and implement techniques to detect and correct gender bias in text.
• Evaluate and compare the techniques through qualitative and quantitative analysis.
• (Optional) Generate synthetic data to improve model performance.

Required Skills

• Good programming skills
• Basic understanding of machine learning concepts or interest to learn them in the process

Further Reading(s)

Sun, Tony, et al. “Mitigating gender bias in natural language processing: Literature review.” Proceedings of the 57th Conference of the Association for Computational Linguistics (2019).