Browse Datasets

Belebele

Reading Comprehension for 122 Language Variants

reading-comprehension multiple-choice evaluation 900 questions per language CC-BY-SA 4.0
Access Dataset Read Paper

Description

Multiple-choice machine reading comprehension dataset spanning 122 language variants. Fully parallel, based on FLORES-200 passages.

Quick Start

Python 
from datasets import load_dataset

dataset = load_dataset("facebook/belebele", split="eng_Latn")

Project Ideas

Zero-Shot LLM Evaluation

Evaluate zero-shot Llama-3 on Wolof reading comprehension

Intermediate

Human vs Machine

Compare human vs machine performance on Zulu

Advanced

Ethical Considerations

  • Parallel nature allows rigorous cross-language bias testing