Extending the Contact Hypothesis: Cross-Linguistic Evaluation of Religion and Nationality Bias When Prompting LLMs in German and Icelandic

2026-05-20 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

The paper "Extending the Contact Hypothesis: Cross-Linguistic Evaluation of Religion and Nationality Bias When Prompting LLMs in German and Icelandic" presents research into biases within Large Language Models (LLMs). Specifically, it conducts a cross-linguistic evaluation to assess religion and nationality biases when these models are prompted using German and Icelandic. The study aims to extend the Contact Hypothesis, a well-established theory from social psychology concerning prejudice reduction through intergroup contact, into the realm of LLM interactions. By examining how biases manifest across distinct linguistic contexts, the research provides crucial insights into the fairness, ethical considerations, and potential societal impacts of deploying LLMs in diverse language environments. This work highlights the importance of understanding and mitigating specific biases related to religious and national identities.

Key takeaway

For NLP Engineers and AI Ethicists deploying LLMs in multilingual environments, understanding cross-linguistic bias is critical. Your models may exhibit distinct religion and nationality biases when prompted in languages like German or Icelandic, even if seemingly fair in English. You should implement rigorous cross-linguistic bias evaluations, extending beyond common languages, to ensure equitable and responsible AI system deployment. This proactive testing helps mitigate unintended societal harms and builds trust in your applications.

Key insights

LLMs exhibit religion and nationality biases, requiring cross-linguistic evaluation for fairness.

Principles

• LLM bias varies cross-linguistically.
• Contact Hypothesis applies to LLM interactions.
• Religion and nationality are key bias vectors.

Method

The paper employs a cross-linguistic evaluation methodology to prompt LLMs in German and Icelandic, specifically assessing religion and nationality biases.

In practice

• Test LLMs for specific identity biases.
• Evaluate LLM fairness across languages.

Topics

• LLM Bias
• Cross-Linguistic Evaluation
• Religion Bias
• Nationality Bias
• German Language
• Icelandic Language

Best for: Research Scientist, AI Scientist, AI Ethicist, NLP Engineer

Related on AIssential

• Location Not Found: Exposing Implicit Local and Global Biases in Multilingual LLMs — Multilingual LLMs exhibit implicit US-centric and population-driven biases, exacerbated by instruction tuning.
• A closer look at how large language models trust humans: patterns and biases — LLMs develop human-like trust patterns, influenced by trustworthiness and demographic biases, especially in financial contexts.
• Bias and Justice in Language Models: Evidence from a Linguistically, Socially, and Culturally Asymmetric Literature — LLM bias research is skewed towards English and binary gender, lacking intersectional and multilingual real-world analyses.
• I asked 5 different AIs to pick a number between 1 and 100… all of them said 42 😬 — LLMs exhibit cultural biases from training data, defaulting to non-random, culturally significant numbers when asked for randomness.
• Why are all LLMs Obsessed with Japanese Culture? On the Hidden Cultural and Regional Biases of LLMs — LLMs exhibit a surprising bias towards Japanese culture, emerging post-supervised fine-tuning.
• Investigating Counterfactual Unfairness in LLMs towards Identities through Humor — LLMs exhibit counterfactual unfairness, applying stricter safety policies and attributing more malicious intent based on speaker-target identity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.