C3

C3 is a multiple-choice reading comprehension dataset. Here we use the C3 dataset4 proposed in [32], which is a competition dataset. It is necessary to construct pseudo-evidence labels for the training set, while the validation set has manually annotated evidence labels.