他提出了一种新的序列到序列的预训练模型，称为ProphetNet，该模型引入了一种新的自我监督目标，称为未来n-gram预测，并提出了n流自我注意机制。his paper presents a new sequence-to- sequence pre-training model called ProphetNet, which introduces a novel self-supervised objective named future n-gram prediction and the proposed n-stream self-attention mechanism.