MulCogBench
收藏资源简介:
MulCogBench是一个多模态认知基准数据集,由中国科学院自动化研究所创建,用于评估中英文计算语言模型。该数据集包含多种认知数据,如主观语义评分、眼动追踪、功能磁共振成像(fMRI)和脑磁图(MEG),来源于中文和英文母语者。数据集旨在通过分析语言模型与认知数据的相似性,探索语言模型处理语言的机制是否与人类相似,特别是在处理复杂语言结构时。此外,数据集还展示了语言模型在不同认知模态和语言单元中的表现,以及中英文之间的相似性,为跨语言研究提供了基础。
MulCogBench is a multimodal cognitive benchmark dataset created by the Institute of Automation, Chinese Academy of Sciences, for evaluating Chinese and English computational language models. This dataset includes various types of cognitive data, such as subjective semantic ratings, eye-tracking recordings, functional magnetic resonance imaging (fMRI), and magnetoencephalography (MEG), collected from native Chinese and English speakers. The dataset aims to explore whether the language processing mechanisms of language models are similar to those of humans, especially when handling complex linguistic structures, by analyzing the similarity between language models and human cognitive data. Additionally, the dataset demonstrates the performance of language models across different cognitive modalities and linguistic units, as well as the similarities between Chinese and English, providing a foundation for cross-linguistic research.




