ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training

Abstract

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.

Accepted to EMNLP 2020 Findings. Project page: https://github.com/microsoft/ProphetNet

References in corpus (12)

• Sequence to Sequence Learning with Neural Networks
• On the difficulty of training Recurrent Neural Networks
• Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
• SummaRuNNer: A Recurrent Neural Network based Sequence Model for Extractive Summarization of Documents
• MASS: Masked Sequence to Sequence Pre-training for Language Generation
• Regularizing and Optimizing LSTM Language Models
• Text Summarization with Pretrained Encoders
• Defending Against Neural Fake News
• Learning to Decode for Future Success
• Pre-trained Language Model Representations for Language Generation
• Addressing Semantic Drift in Question Generation for Semi-Supervised Question Answering
• Plan, Attend, Generate: Planning for Sequence-to-Sequence Models

Cited by in corpus (25)

• XGLUE: A New Benchmark Dataset for Cross-lingual Pre-training, Understanding and Generation
• Text Compression-aided Transformer Encoding
• Learning to summarize from human feedback
• Establishing Baselines for Text Classification in Low-Resource Languages
• BANG: Bridging Autoregressive and Non-autoregressive Generation with Large Scale Pretraining
• Better Fine-Tuning by Reducing Representational Collapse
• Learning by Semantic Similarity Makes Abstractive Summarization Better
• AdaptSum: Towards Low-Resource Domain Adaptation for Abstractive Summarization
• A Baseline Analysis for Podcast Abstractive Summarization
• Improving Non-autoregressive Generation with Mixup Training
• Summarize, Outline, and Elaborate: Long-Text Generation via Hierarchical Supervision from Extractive Summaries
• A Two-Phase Approach for Abstractive Podcast Summarization
• Pre-training for Abstractive Document Summarization by Reinstating Source Text
• Automatic Summarization of Open-Domain Podcast Episodes
• Attention Head Masking for Inference Time Content Selection in Abstractive Summarization
• Enhancing Question Generation with Commonsense Knowledge
• Can You Fool AI by Doing a 180? $\unicode{x2013}$ A Case Study on Authorship Analysis of Texts by Arata Osada
• Tell Me How to Ask Again: Question Data Augmentation with Controllable Rewriting in Continuous Space
• A New Approach to Overgenerating and Scoring Abstractive Summaries
• Semantic-Based Self-Critical Training For Question Generation
• DeepTitle -- Leveraging BERT to generate Search Engine Optimized Headlines
• ProphetNet-Ads: A Looking Ahead Strategy for Generative Retrieval Models in Sponsored Search Engine
• ImaginE: An Imagination-Based Automatic Evaluation Metric for Natural Language Generation
• Inference Time Style Control for Summarization
• Resurrecting Submodularity for Neural Text Generation