ExpliCa
收藏资源简介:
ExpliCa数据集是由比萨大学研究团队创建的,用于评估大型语言模型在显式因果推理方面的能力。该数据集包含600对英文句子,通过特定的连接词表达句子之间的因果或时间关系。数据集中的句子对经过精心设计,以确保高质量和词频平衡,并通过众包方式获得英语母语者的接受度评分,为测试各种模型提供了一个坚实的基础。ExpliCa旨在解决因果推理中的问题,特别是在评估大型语言模型在理解和区分因果和时间关系方面的能力。
The ExpliCa dataset was created by a research team at the University of Pisa to assess the explicit causal reasoning capabilities of large language models (LLMs). It comprises 600 pairs of English sentences, with specific conjunctions used to express causal or temporal relationships between each pair of sentences. The sentence pairs in the dataset are meticulously designed to ensure high quality and balanced word frequency, and their acceptability scores from native English speakers were obtained via crowdsourcing, providing a solid foundation for testing various models. ExpliCa aims to address gaps in causal reasoning research, particularly for evaluating large language models' abilities to understand and distinguish between causal and temporal relationships.

- 1ExpliCa: Evaluating Explicit Causal Reasoning in Large Language Models比萨大学 · 2025年



