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

TL;DR

ProphetNet is a novel sequence-to-sequence pre-training model that predicts multiple future tokens simultaneously, leading to improved performance on summarization and question generation tasks.

Contribution

It introduces future n-gram prediction and n-stream self-attention, enabling the model to plan ahead and outperform previous models on key benchmarks.

Findings

Achieves state-of-the-art results on CNN/DailyMail, Gigaword, and SQuAD 1.1.

Outperforms models with the same pre-training data scale.

Demonstrates the effectiveness of future n-gram prediction in sequence modeling.

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.