BIS Reasoning 1.0
收藏资源简介:
BIS Reasoning 1.0数据集是首个大规模的日语三段论推理问题数据集,旨在评估大型语言模型(LLM)在信念不一致推理方面的能力。该数据集包含5000个精心构建的三段论推理问题,每个问题都包含两个前提和一个结论,其中结论在逻辑上有效,但与常见的常识信念相矛盾。数据集通过形式化规范流程进行开发,确保了逻辑严谨性和语言质量,并由母语为日语的标注员进行审查,以确保语言流畅性和自然性。BIS Reasoning 1.0数据集旨在测试LLM在处理逻辑上有效但与直觉信念相冲突的输入时的鲁棒性,并为评估LLM在需要严格逻辑严谨性的领域(如法律、医疗保健和科学研究)中的可靠性提供重要见解。
The BIS Reasoning 1.0 dataset is the first large-scale Japanese syllogistic reasoning problem dataset, designed to evaluate the capability of Large Language Models (LLMs) in belief inconsistency reasoning. This dataset contains 5,000 meticulously constructed syllogistic reasoning problems, each consisting of two premises and a conclusion. The conclusion is logically valid but contradicts common sense beliefs. The dataset was developed through a formal specification process to ensure logical rigor and linguistic quality, and reviewed by native Japanese annotators to guarantee linguistic fluency and naturalness. The BIS Reasoning 1.0 dataset aims to test the robustness of LLMs when processing inputs that are logically valid but conflict with intuitive beliefs, and provide valuable insights for evaluating the reliability of LLMs in fields requiring strict logical rigor, such as law, healthcare, and scientific research.

- 1BIS Reasoning 1.0: The First Large-Scale Japanese Benchmark for Belief-Inconsistent Syllogistic Reasoning国立信息研究所(NII) · 2025年



