The study evaluates how well large language models (LLMs) understand geographic lexical variation in Spanish across 21 countries, using an expert-curated database of 934 terms. Ten LLMs were tested with yes-no and multiple-choice questions, generating 10,164 and 16,637 items respectively, and performance was measured with F1 and adjusted Jaccard scores. Results show notable differences by model, country, and question format, with Spain scoring highest and Chile lowest, and suggest that digital resource availability and lexical distance from Spain influence performance.
arXiv:2606. 07422v1 Announce Type: cross Abstract: Large language models are increasingly used to answer culturally grounded questions across languages, yet it remains unclear whether local cultural knowledge is better accessed through English or the local language.
The paper introduces TriviaRoomQA, a multilingual benchmark of 3,300 parallel multiple-choice questions across 288 topics, plus an additional 5,340 French-only questions. It evaluates 30 open-weight large language models (7–70 B parameters) from various regions, finding strong performance on knowledge-intensive subjects like history and geography but weaker results on everyday popular‑culture topics such as celebrities, music, movies, and news. The study also shows that model performance varies across languages for the same questions, indicating that factual knowledge access is not always language‑independent.
Quiz rooms, trivia nights, and quiz shows challenge human knowledge across a wide range of topics, from canonical facts to everyday culture. In this paper, we examine whether large language models (LLMs) can perform competitively in such settings, using quiz-style questions to test them on both common and niche topics.
CARTE (Culturally Anchored Regional-Territorial Evaluation) is a multiple-choice benchmark designed to assess large language models’ ability to distinguish fine-grained, geographically grounded knowledge within France. It contains 3,132 questions covering 13 metropolitan regions across 12 thematic domains such as culture, language, demographics, economy, environment, and mobility, and includes a subset called CARTE‑LV for linguistic variation. The benchmark was validated through a four-stage process and used to evaluate 32 open-weight LLMs and three API models in zero-, one-, and three-shot settings, revealing performance disparities across regions and model scales.
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 paper introduces the concept of Language Specific Knowledge (LSK), showing that multilingual language models can answer certain queries better when prompted in a language other than English, sometimes even in low‑resource languages. It defines a language‑selection problem and presents several baseline methods, including the authors’ LSKExtractor, to empirically demonstrate that choosing the optimal language can improve question‑answering performance across datasets covering cultural and social norms. Experiments reveal non‑intuitive mappings, such as Gemma models excelling on Chinese and Middle Eastern topics in Spanish and Qwen models performing best on authority and responsibility queries in Arabic and Chinese.
We introduce CARTE 1 (Culturally Anchored Regional-Territorial Evaluation), a multiplechoice benchmark for evaluating the ability of large language models (LLMs) to perform fine-grained reasoning over geographically grounded and regionally differentiated knowledge within France. While prior benchmarks focus on national-level cultural understanding, they largely overlook intra-country variation and the need to distinguish between closely related regional contexts.
arXiv:2607. 06327v1 Announce Type: cross Abstract: Uncertainty estimation (UE) enables LLM-powered systems to recognize when to abstain, yet existing research has predominantly focused on English.
The paper introduces MAPLE, a family of decoder‑only language models pretrained with document‑level geographic metadata such as source URL, country, and continent. MAPLE is evaluated on a new benchmark, LocalNewsQA, which tests whether models can switch answers when the locale changes. Experiments show that, with inference‑time metadata fixed, MAPLE outperforms metadata‑free controls in both answer switching and accuracy on locale‑dependent questions, and these gains grow with model size.
The paper "Where's Waldo? Query-language Preference under Cross-lingual Knowledge Disparities" introduces the Waldo benchmark, a multilingual QA dataset built from Wikipedia that focuses on knowledge gaps and conflicts across languages. It evaluates eight models in five languages and finds that when a fact is missing in one language, models tend to use evidence from the other language, but when conflicting accounts exist, responses align strongly with the query language, leading to different answers for semantically identical questions. The study also explores mitigation strategies, including ablating attention heads and LoRA-based training, which can reduce the preference gap by up to 61.5%.
arXiv:2610.35168v1 Announce Type: new Abstract: The value of language models for any community depends on whether a model actually understands the culture of the people it intends to serve. We use th...
arXiv:2610.20419v1 Announce Type: new Abstract: Retrieval-Augmented Generation (RAG) models have become almost a cornerstone of open-domain question answering tasks; in which the quality of answers g...