CPG-EVAL
收藏资源简介:
CPG-EVAL是一个专门为评估大型语言模型在汉语作为外语教学环境中教学语法知识而设计的多级别基准。该数据集由早稻田大学教育學院與綜合藝術和科學學院创建,包含739个教学语法项目和对应语言实例,以及为了评估模型抗干扰能力而设计的高度混淆实例。数据集基于《汉语语法学习手册》构建,涵盖从初级到高级的所有水平,并为每个语法项目提供了9个示例句子,总计6651个合成句子,用于构建基准问题。
CPG-EVAL is a multi-level benchmark specifically designed to evaluate large language models' pedagogical grammar knowledge in the context of teaching Chinese as a foreign language. This dataset was created by the Faculty of Education and the Faculty of Comprehensive Arts and Sciences of Waseda University. It contains 739 pedagogical grammar items and their corresponding linguistic examples, as well as highly confusing examples designed to assess the model's anti-interference ability. Constructed based on the *Handbook of Chinese Grammar Learning*, the dataset covers all proficiency levels from beginner to advanced, and provides 9 example sentences for each grammar item, totaling 6,651 synthetic sentences used to build benchmark questions.

- 1CPG-EVAL: A Multi-Tiered Benchmark for Evaluating the Chinese Pedagogical Grammar Competence of Large Language Models早稻田大学教育學院與綜合藝術和科學學院 · 2025年



