prophetnet predicting future n gram for sequence to sequence pre training

• DOI: 10.48550/arxiv.2001.04063
• Publication Date: 2020-01-01
• Authors: Weizhen Qi; Yeyun Gong; Ming Zhou; Ruofei Zhang; Dayiheng Liu; Nan Duan; Yu Yan; Jiusheng Chen
• Keywords: FOS: Computer and information sciences; Computer Science - Computation and Language; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; Computation and Language (cs.CL)

Accepted to EMNLP 2020 Findings. Project page: https://github.com/microsoft/ProphetNet<br/>This 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. Instead of optimizing one-step-ahead prediction in the traditional sequence-to-sequence model, the ProphetNet is optimized by n-step ahead prediction that predicts the next n tokens simultaneously based on previous context tokens at each time step. The future n-gram prediction explicitly encourages the model to plan for the future tokens and prevent overfitting on strong local correlations. We pre-train ProphetNet using a base scale dataset (16GB) and a large-scale dataset (160GB), respectively. Then we conduct experiments on CNN/DailyMail, Gigaword, and SQuAD 1.1 benchmarks for abstractive summarization and question generation tasks. Experimental results show that ProphetNet achieves new state-of-the-art results on all these datasets compared to the models using the same scale pre-training corpus.<br/>