BLEnD: A Benchmark for LLMs on Everyday Knowledge in Diverse Cultures and Languages

2024-06-14

arXiv.org

AI Summary

Existing evaluations inadequately assess large language models’ (LLMs) understanding of multicultural everyday commonsense knowledge—particularly in non-English and low-resource language settings. To address this gap, we introduce BLEnD, the first multilingual, cross-cultural benchmark explicitly designed to evaluate localized daily-life knowledge. BLEnD spans 16 countries/regions and 13 languages—including low-resource ones such as Amharic—and comprises 52.6K expert-annotated question-answer pairs grounded in authentic scenarios (e.g., birthday cuisine, traditional instruments). Methodologically, it employs ethnographic data collection—replacing web scraping—with dual-format question design, cross-lingual consistency verification, and a zero-shot evaluation framework. Our analysis reveals, for the first time, a counterintuitive “English supremacy” bias in LLMs’ low-resource language performance, alongside systematic cultural representation gaps; GPT-4 exhibits up to a 57.34% cross-cultural accuracy disparity. The dataset is publicly released.

Technology Category

• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Common-Sense Reasoning

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large language models (LLMs) often lack culture-specific knowledge of daily life, especially across diverse regions and non-English languages. Existing benchmarks for evaluating LLMs' cultural sensitivities are limited to a single language or collected from online sources such as Wikipedia, which do not reflect the mundane everyday lifestyles of diverse regions. That is, information about the food people eat for their birthday celebrations, spices they typically use, musical instruments youngsters play, or the sports they practice in school is common cultural knowledge but uncommon in easily collected online sources, especially for underrepresented cultures. To address this issue, we introduce BLEnD, a hand-crafted benchmark designed to evaluate LLMs' everyday knowledge across diverse cultures and languages. BLEnD comprises 52.6k question-answer pairs from 16 countries/regions, in 13 different languages, including low-resource ones such as Amharic, Assamese, Azerbaijani, Hausa, and Sundanese. We construct the benchmark to include two formats of questions: short-answer and multiple-choice. We show that LLMs perform better for cultures that are highly represented online, with a maximum 57.34% difference in GPT-4, the best-performing model, in the short-answer format. For cultures represented by mid-to-high-resource languages, LLMs perform better in their local languages, but for cultures represented by low-resource languages, LLMs perform better in English than the local languages. We make our dataset publicly available at: https://github.com/nlee0212/BLEnD.

Problem

• Multicultural Common Sense
• Language Models
• Non-English Contexts

Innovation

• BLEnD
• Multilingual Capability
• Cultural Adaptability