QED

QED数据集是由Google Research创建的一个专注于问答系统解释性的数据集。该数据集基于Google自然问题数据集的一个子集，包含7638个由专家标注的解释示例。QED解释框架依据形式语义概念，如指称等同、句子性和蕴涵，来具体化问题与答案之间的关系。此数据集不仅支持模型在提供答案的同时给出推理过程，还通过用户研究展示了QED解释显著提高了未训练评分者识别强神经QA基线错误的能力。数据集的应用领域包括提高QA系统的可调试性、可扩展性和用户信任度。

The QED dataset, created by Google Research, is a resource dedicated to the explainability of question answering (QA) systems. Built upon a subset of the Google Natural Questions dataset, this dataset contains 7,638 expert-annotated explanation examples. The QED explanation framework formalizes the relationship between questions and answers based on formal semantic concepts such as referential identity, sententiality, and entailment. This dataset not only enables models to generate reasoning processes alongside their provided answers, but also demonstrates through user studies that QED explanations significantly improve the ability of untrained raters to identify errors in strong neural QA baseline systems. Application domains of this dataset include enhancing the debuggability, scalability, and user trust of QA systems.