RACE

ReAding Comprehension dataset from Exams (RACE) 数据集是一个机器阅读理解数据集，由 27,933 个段落和 97,867 个英语考试题目组成，针对 12-18 岁的中国学生。 RACE 由分别来自中学和高中考试的两个子集 RACE-M 和 RACE-H 组成。 RACE-M 有 28,293 个问题，RACE-H 有 69,574 个问题。每个问题与 4 个候选答案相关联，其中一个是正确的。 RACE 的数据生成过程与大多数机器阅读理解数据集不同 - RACE 中的问题不是通过启发式或众包来生成问题和答案，而是专门为测试人类阅读技能而设计的，并且由领域专家创建。

The Reading Comprehension dataset from Exams (RACE) is a machine reading comprehension dataset comprising 27,933 passages and 97,867 English exam questions, targeting Chinese students aged 12 to 18. RACE is split into two subsets, RACE-M and RACE-H, which originate from middle school and high school exams respectively. RACE-M contains 28,293 questions, while RACE-H has 69,574 questions. Each question is paired with four candidate answers, with exactly one being the correct option. The data generation process of RACE is distinct from most machine reading comprehension datasets: instead of generating questions and answers through heuristic methods or crowdsourcing, the questions in RACE are specifically designed to assess human reading proficiency and created by domain experts.

搜集汇总

背景与挑战

背景概述

RACE是一个针对中国学生的英语阅读理解数据集，包含近10万个考试题目，分为中学和高中两个子集。其独特之处在于所有问题均由专家设计，旨在测试阅读技能，而非通过众包或启发式方法生成。