HELP

HELP数据集是由理化学研究所和Ochanomizu大学等机构合作创建，专注于自然语言推理中的单调性推理。该数据集包含36,000个推理对，涉及向上单调、向下单调、非单调、连接和分离等多种逻辑推理类型。数据集的创建过程利用了Parallel Meaning Bank的精细标注，通过自动化的方式生成推理问题，旨在通过增加训练数据来提高神经模型在单调性推理上的表现。HELP数据集的应用领域主要集中在自然语言理解，特别是解决自然语言推理中与单调性相关的逻辑问题。

The HELP dataset was co-created by institutions including RIKEN and Ochanomizu University, focusing on monotonic reasoning in natural language inference. This dataset contains 36,000 inference pairs, covering multiple logical reasoning types such as upward monotonicity, downward monotonicity, non-monotonicity, conjunction and disjunction. The dataset was developed using fine-grained annotations from the Parallel Meaning Bank, with inference questions generated via automated methods, aiming to improve the performance of neural models on monotonic reasoning tasks by augmenting training data. The HELP dataset is mainly applied in the field of natural language understanding, particularly for solving logical reasoning problems related to monotonicity in natural language inference.