Belebele

Belebele是一个包含122种语言变体的多项选择机器阅读理解数据集，显著扩展了自然语言理解基准的语言覆盖范围，使文本模型能够在高、中、低资源语言中进行评估。每个问题基于Flores-200数据集的简短段落，并提供四个多项选择答案。问题经过精心设计，以区分具有不同级别通用语言理解能力的模型。

Belebele is a multiple-choice machine reading comprehension dataset encompassing 122 language variants, which significantly expands the language coverage of natural language understanding benchmarks and enables text models to be evaluated across high-, medium-, and low-resource languages. Each question is based on short passages sourced from the Flores-200 dataset, with four multiple-choice answer options provided. The questions are meticulously designed to differentiate models with varying levels of general language understanding capabilities.

搜集汇总

背景与挑战

背景概述

Belebele是一个多语言机器阅读理解数据集，包含122种语言变体，基于Flores-200数据集的段落设计多项选择问题，用于评估文本模型在高、中、低资源语言中的通用理解能力。