Quoref

Quoref 是一个 QA 数据集，用于测试阅读理解系统的共指推理能力。在这个跨度选择基准包含来自维基百科的 4.7K 段落中的 24K 问题，系统必须先解决硬共指，然后才能在段落中选择适当的跨度来回答问题。

Quoref is a QA dataset designed to test the coreference reasoning ability of reading comprehension systems. This span selection benchmark contains 24K questions from 4.7K paragraphs sourced from Wikipedia; systems must first resolve hard coreference before selecting the appropriate span in the paragraph to answer the question.