The Flow has not summarised this story yet — read it at arXiv Computation and Language.
MBBQ (Multilingual Bias Benchmark for Question-answering) is a curated extension of the English BBQ dataset that includes Dutch, Spanish, and Turkish, designed to measure stereotypes across these languages while controlling for cultural differences and task accuracy. The dataset also provides a parallel control set to assess question-answering performance independently of bias. Experiments with several open-source and proprietary LLMs show that non‑English languages can exhibit greater bias than English, and cross‑lingual bias differences persist except in the most accurate models.
The paper introduces CAPRI, a dataset of conversations designed to test whether large language models (LLMs) can adapt their responses to users’ cultural backgrounds. Experiments show that while LLMs can infer a user’s cultural context and recall relevant conventions, they frequently fail to apply this knowledge when generating answers unless explicitly prompted to do so. The study also finds that LLMs’ cultural adaptations improve as more cues are provided, yet their default priors are not culture‑neutral and sometimes reflect the model’s country of origin.
CuBEs introduces culturally‑situated behavioral evaluations for large language models, adding cultural context to test scenarios and assessments. The authors built a human‑labeled dataset covering 12 cultures, revealing significant cross‑cultural differences that one‑size‑fits‑all judgments miss. Evaluating 13 LLMs shows that culturally situated tests uncover varied behaviors, such as Western political bias versus non‑Western religious or colonial biases, which standard evaluations overlook.
The paper investigates cultural biases in large language models (LLMs) by introducing the Culture-Related Open Questions (CROQ) dataset, which contains 24‑language questions about generic culture. Experiments reveal that LLMs disproportionately favor Japan in their responses, especially when prompted in high‑resource languages, while low‑resource languages tend to highlight countries where the language is official. The study also finds that these biases emerge after supervised fine‑tuning rather than during pre‑training.
The study examines how multilingual large language models (LLMs) produce outputs that differ across sociocultural contexts, highlighting that identity labels and source-language cues can mislead assessments of cultural grounding. Using a human‑validated, multi‑agent audit on 89,253 outputs from 12 LLMs in English, French, and Chinese across 18 occupations and three task conditions, the authors find that bias representation varies systematically by language and task. Removing direct identity cues reduces identity‑label prediction in English and Chinese but not in French, and the source language’s cultural context consistently receives the highest relevance score, though this signal weakens after translation or name masking. "whyItMatters":"The findings show that surface cues can obscure true cross‑cultural patterns, underscoring the need for careful audit designs to avoid misleading conclusions about bias in multilingual LLMs."
The paper investigates which demographic attributes large language models (LLMs) default to when annotating text without explicit demographic cues. By comparing non‑demographic, placebo‑conditioned, and demographic‑conditioned prompts on politeness and offensiveness tasks in the POPQUORN dataset, the authors find that LLMs exhibit notable gender, race, and age influences in their annotations. This contrasts with earlier studies that reported no such effects, highlighting the importance of considering demographic bias in LLM‑based annotation workflows.