CUGE

CUGE是一个全面系统的汉语理解和生成评估基准，由清华大学计算机科学与技术系等多个机构合作创建。该数据集包含21个代表性数据集，覆盖7种重要语言能力、18个主流NLP任务。数据集的创建遵循语言能力-任务-数据集的层次框架，旨在更系统地组织现有评估资源，全面反映通用语言评估需求。CUGE的应用领域广泛，旨在推动通用语言智能的研究与发展，解决现有评估基准的不足，如平面基准框架和过度简化的评分策略。

CUGE is a comprehensive and systematic Chinese understanding and generation evaluation benchmark co-created by multiple institutions including the Department of Computer Science and Technology of Tsinghua University. This dataset comprises 21 representative datasets, covering 7 critical language capabilities and 18 mainstream NLP tasks. Developed under a hierarchical framework of 'language capability - task - dataset', it aims to systematically organize existing evaluation resources and comprehensively reflect the requirements of general language evaluation. CUGE features broad application scenarios, and is intended to advance the research and development of general language intelligence, while addressing the shortcomings of current evaluation benchmarks such as flat benchmark frameworks and overly simplified scoring strategies.

搜集汇总

背景与挑战

背景概述

CUGE是一个中文语言理解和生成评测基准，包含高质量的中文自然语言处理数据集、排行榜和在线评测平台。其目标是构建全面系统的中文机器语言能力评测体系，提供多层次的评测方案，以科学、规范地推进中文NLP技术的发展。