Adding A Filter Based on The Discriminator to Improve Unconditional Text Generation
arXiv:2004.02135
Abstract
The autoregressive language model (ALM) trained with maximum likelihood estimation (MLE) is widely used in unconditional text generation. Due to exposure bias, the generated texts still suffer from low quality and diversity. This presents statistically as a discrepancy between the real text and generated text. Some research shows a discriminator can detect this discrepancy. Because the discriminator can encode more information than the generator, discriminator has the potentiality to improve generator. To alleviate the exposure bias, generative adversarial networks (GAN) use the discriminator to update the generator's parameters directly, but they fail by being evaluated precisely. A critical reason for the failure is the difference between the discriminator input and the ALM input. We propose a novel mechanism by adding a filter which has the same input as the discriminator. First, discriminator detects the discrepancy signals and passes to filter directly (or by learning). Then, we use the filter to reject some generated samples with a sampling-based method. Thus, the original generative distribution is revised to reduce the discrepancy. Two ALMs, RNN-based and Transformer-based, are experimented. Evaluated precisely by three metrics, our mechanism consistently outperforms the ALMs and all kinds of GANs across two benchmark data sets.
References in corpus (15)
- Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
- Categorical Reparameterization with Gumbel-Softmax
- Microsoft COCO Captions: Data Collection and Evaluation Server
- Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks
- MaskGAN: Better Text Generation via Filling in the______
- GANS for Sequences of Discrete Elements with the Gumbel-softmax Distribution
- How (not) to Train your Generative Model: Scheduled Sampling, Likelihood, Adversary?
- Maximum-Likelihood Augmented Discrete Generative Adversarial Networks
- Defending Against Neural Fake News
- On Accurate Evaluation of GANs for Language Generation
- Neural Text Generation: Past, Present and Beyond
- Eval all, trust a few, do wrong to none: Comparing sentence generation models
- Toward Diverse Text Generation with Inverse Reinforcement Learning
- Residual Energy-Based Models for Text Generation
- Discriminative Adversarial Search for Abstractive Summarization