The Children’s Book Test 问答数据集

CBT 数据集由文字段落和相应问题构建，问答数据均来自古腾堡项目免费提供的书籍，该数据集用于直接测量语言模型、更广泛的语言环境用于问答和仿真查找。

The CBT Dataset is constructed from text passages and their corresponding questions. All question-answering data are sourced from freely accessible books provided by Project Gutenberg. This dataset is used to directly evaluate language models, and the broader linguistic context it offers supports question answering and simulated retrieval.

搜集汇总

背景与挑战

背景概述

The Children’s Book Test (CBT) 问答数据集基于古腾堡项目的免费儿童书籍构建，包含文字段落和对应问题，用于评估语言模型在问答任务中的性能。该数据集由Facebook于2016年发布，问题按词性类型（如命名实体、普通名词等）分类，涵盖训练、验证和测试集。