Reproduce “Social Bias in Multilingual Language Models: A Survey” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lance Calvin Lim Gamboa, Yue Feng, Mark Lee
This survey systematically reviews social bias in multilingual and non-English language models, identifying significant gaps such as an Anglocentric linguistic focus, limited cultural awareness in benchmark development, and a scarcity of bias mitigation studies. It documents common challenges in adapting bias benchmarks across languages and cultures, alongside implemented solutions.
Key insights
- Multilingual bias research exhibits a pronounced linguistic bias, over-focusing on Indo-European and highly-resourced languages while neglecting many widely spoken languages in less developed regions.
- A significant portion of multilingual bias evaluation lacks explicit documentation of cultural considerations during benchmark creation, raising questions about the cultural appropriateness and validity of adapted benchmarks.
- The field shows limited focus on bias mitigation, with most studies (69.81%) not conducting mitigation experiments, and those that do often rely on older, embedding-based techniques rather than state-of-the-art approaches.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.