View Proposal

---

Proposer

Gavin Abercrombie

Title

Cultural Hegemony in LLMs

Goal

Develop a cross-cultural dataset that captures how cultural hegemony manifests in LLM outputs across different national, regional, linguistic, ethnic, and social contexts

Description

This project is part of a wider effort to investigate the extent to which LLM outputs exhibit cultural hegemony. The broader aim of this work is to understand how LLMs represent different communities, regions, languages, identities, and cultural practices, and whether they tend to privilege certain dominant social or cultural perspectives while marginalising others. This work differs from much of the existing research on cultural bias in LLMs by moving beyond the question of whether models make factual errors about a culture or represent one country more accurately than another. Instead, we focus on cultural hegemony: how models may normalise dominant social, linguistic, caste, class, religious, gendered, regional, or urban perspectives as if they represent the whole culture. This approach, therefore, examines internal hierarchies within societies and communities rather than focusing solely on differences between countries. We are now hoping to expand this work internationally and would be very interested in collaborating with you in collecting more data. The goal is to develop a broader cross-cultural dataset that captures how cultural hegemony manifests in LLM outputs across different national, regional, linguistic, ethnic, and social contexts. Your team’s local/cultural expertise would be extremely valuable for this task. We are looking for collaborators who can help design culturally relevant prompts, identify important social and cultural axes in their own context, review model outputs, and support the annotation or validation of the dataset. In this project we will develop a data collection pipeline and analyse the outputs of LLMS. The project requires both technical skills and deep reading.

Resources

Background

Url

Difficulty Level

Challenging

Ethical Approval

Full

Number Of Students

1

Supervisor

Gavin Abercrombie

Keywords

nlp, llms, computational social science

Degrees

Bachelor of Science in Computer Science
Master of Science in Artificial Intelligence
Master of Science in Data Science