Improving Text Generation with Student-Forcing Optimal Transport

Abstract

Neural language models are often trained with maximum likelihood estimation (MLE), where the next word is generated conditioned on the ground-truth word tokens. During testing, however, the model is instead conditioned on previously generated tokens, resulting in what is termed exposure bias. To reduce this gap between training and testing, we propose using optimal transport (OT) to match the sequences generated in these two modes. An extension is further proposed to improve the OT learning, based on the structural and contextual information of the text sequences. The effectiveness of the proposed method is validated on machine translation, text summarization, and text generation tasks.

arXiv commentTo appear at EMNLP 2020

References in corpus (21)

• Sequence to Sequence Learning with Neural Networks
• Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
• Language Models are Few-Shot Learners
• Recurrent Neural Network Regularization
• Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks
• A Neural Conversational Model
• Professor Forcing: A New Algorithm for Training Recurrent Networks
• Maximum-Likelihood Augmented Discrete Generative Adversarial Networks
• Adversarial Ranking for Language Generation
• Adversarial Feature Matching for Text Generation
• DialogWAE: Multimodal Response Generation with Conditional Wasserstein Auto-Encoder
• Reward Augmented Maximum Likelihood for Neural Structured Prediction
• Language GANs Falling Short
• Adversarial Text Generation via Feature-Mover's Distance
• Differential Properties of Sinkhorn Approximation for Learning with Wasserstein Distance
• Generative Adversarial Network Training is a Continual Learning Problem
• Cold-Start Reinforcement Learning with Softmax Policy Gradient
• Contextualized Perturbation for Textual Adversarial Attack
• Action-depedent Control Variates for Policy Optimization via Stein's Identity
• Greedy Search with Probabilistic N-gram Matching for Neural Machine Translation
• Sequence Generation with Guider Network