PAP

PAP数据集是由斯图加特大学自然语言处理研究所的Annerose Eichel和Sabine Schulte im Walde创建的，专注于英语事件的物理和抽象合理性。该数据集基于从维基百科提取的自然发生句子，通过自动生成扰动的伪不合理事件来探索抽象性程度，并通过众包进行合理性标注，以确保标注质量。数据集包含1,733个事件三元组，平均每个三元组有8.9个评级。PAP数据集旨在解决自然语言处理中事件合理性的建模问题，特别是在物理和抽象层面的合理性判断，以及人类在判断合理性时的分歧。

The PAP Dataset was created by Annerose Eichel and Sabine Schulte im Walde from the Institute for Natural Language Processing, University of Stuttgart, focusing on the physical and abstract plausibility of English events. Built upon naturally occurring sentences extracted from Wikipedia, this dataset explores the degree of abstractness by automatically generating perturbed, seemingly implausible events, and uses crowdsourcing to conduct plausibility annotations to ensure annotation quality. The dataset comprises 1,733 event triples, with an average of 8.9 ratings per triple. The PAP Dataset aims to address the challenge of event plausibility modeling in natural language processing, particularly regarding plausibility judgment at both physical and abstract levels, as well as human disagreements when assessing plausibility.