A novel repetition normalized adversarial reward for headline generation

2019/02/19 by Peng Xu, Xu, Peng, Pascale Fung +1
Computer Science · #Topic Modeling #Natural Language Processing Techniques #Multimodal Machine Learning Applications

paper · pdf · doi:10.48550/arxiv.1902.07110

Abstract

While reinforcement learning can effectively improve language generation models, it often suffers from generating incoherent and repetitive phrases \citepaulus2017deep. In this paper, we propose a novel repetition normalized adversarial reward to mitigate these problems. Our repetition penalized reward can greatly reduce the repetition rate and adversarial training mitigates generating incoherent phrases. Our model significantly outperforms the baseline model on ROUGE-1 (+3.24), ROUGE-L (+2.25), and a decreased repetition-rate (-4.98%).

Related

• Learning to Break the Loop: Analyzing and Mitigating Repetitions for Neural Text Generation
• Joint Repetition Suppression and Content Moderation of Large Language Models
• Implicit Unlikelihood Training: Improving Neural Text Generation with\n Reinforcement Learning
• Mitigating the Learning Bias towards Repetition by Self-Contrastive Training for Open-Ended Generation
• Universal Adversarial Suffixes for Language Models Using Reinforcement Learning with Calibrated Reward