BLEnD

This is the official repository of BLEnD: A Benchmark for LLMs on Everyday Knowledge in Diverse Cultures and Languages (Submitted to NeurIPS 2024 Datasets and Benchmarks Track).

About

Large language models (LLMs) often lack culture-specific everyday knowledge, especially across diverse regions and non-English languages. Existing benchmarks for evaluating LLMs' cultural sensitivities are usually limited to a single language or online sources like Wikipedia, which may not reflect the daily habits, customs, and lifestyles of different 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 not always explicitly written online. To address this issue, we introduce BLEnD, a hand-crafted benchmark designed to evaluate LLMs' everyday knowledge across diverse cultures and languages. The benchmark 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 evaluate LLMs in two formats: short-answer questions, and multiple-choice questions. We show that LLMs perform better in cultures that are more present online, with a maximum 57.34% difference in GPT-4, the best-performing model, in the short-answer format. Furthermore, we find that LLMs perform better in their local languages for mid-to-high-resource languages. Interestingly, for languages deemed to be low-resource, LLMs provide better answers in English.

Dataset

All the data samples for short-answer questions, including the human-annotated answers, can be found in the data/ directory. Specifically, the annotations from each country are included in the annotations split, and each country/region's data can be accessed by country codes.

from datasets import load_dataset

annotations = load_dataset("nayeon212/BLEnD",'annotations')

# To access data from Assam:
assam_annotations = annotations['AS']

Each file includes a JSON variable with question IDs, questions in the local language and English, the human annotations both in the local language and English, and their respective vote counts as values. The same dataset for South Korea is shown below:

[{
    "ID": "Al-en-06",
    "question": "대한민국 학교 급식에서 흔히 볼 수 있는 음식은 무엇인가요?",
    "en_question": "What is a common school cafeteria food in your country?",
    "annotations": [
        {
            "answers": [
                "김치"
            ],
            "en_answers": [
                "kimchi"
            ],
            "count": 4
        },
        {
            "answers": [
                "밥",
                "쌀밥",
                "쌀"
            ],
            "en_answers": [
                "rice"
            ],
            "count": 3
        },
        ...
    ],
    "idks": {
        "idk": 0,
        "no-answer": 0,
        "not-applicable": 0
    }
}],

The topics and source language for each question can be found in short-answer-questions split. Questions for each country in their local languages and English can be accessed by country codes. Each CSV file question ID, topic, source language, question in English, and the local language (in the Translation column) for all questions.

from datasets import load_dataset

questions = load_dataset("nayeon212/BLEnD",'short-answer-questions')

# To access data from Assam:
assam_questions = questions['AS']

The current set of multiple choice questions and their answers can be found at the multiple-choice-questions split.

from datasets import load_dataset

mcq = load_dataset("nayeon212/BLEnD",'multiple-choice-questions')

Country/Region Codes

Country/Region	Code	Language	Code
United States	US	English	en
United Kingdom	GB	English	en
China	CN	Chinese	zh
Spain	ES	Spanish	es
Mexico	MX	Spanish	es
Indonesia	ID	Indonesian	id
South Korea	KR	Korean	ko
North Korea	KP	Korean	ko
Greece	GR	Greek	el
Iran	IR	Persian	fa
Algeria	DZ	Arabic	ar
Azerbaijan	AZ	Azerbaijani	az
West Java	JB	Sundanese	su
Assam	AS	Assamese	as
Northern Nigeria	NG	Hausa	ha
Ethiopia	ET	Amharic	am