InitialGAN: A Language GAN with Completely Random Initialization

TL;DR

InitialGAN introduces a novel language GAN trained from completely random initialization, effectively addressing exposure bias without pre-training, and outperforms traditional MLE methods in text generation quality.

Contribution

The paper presents InitialGAN, the first language GAN to outperform MLE without pre-training, using dropout sampling and fully normalized LSTM techniques.

Findings

InitialGAN surpasses MLE and other models in quality.

Proposes a new evaluation metric, Least Coverage Rate.

Demonstrates effective training from random initialization.

Abstract

Text generative models trained via Maximum Likelihood Estimation (MLE) suffer from the notorious exposure bias problem, and Generative Adversarial Networks (GANs) are shown to have potential to tackle this problem. Existing language GANs adopt estimators like REINFORCE or continuous relaxations to model word probabilities. The inherent limitations of such estimators lead current models to rely on pre-training techniques (MLE pre-training or pre-trained embeddings). Representation modeling methods which are free from those limitations, however, are seldomly explored because of their poor performance in previous attempts. Our analyses reveal that invalid sampling methods and unhealthy gradients are the main contributors to such unsatisfactory performance. In this work, we present two techniques to tackle these problems: dropout sampling and fully normalized LSTM. Based on these two techniques, we propose InitialGAN whose parameters are randomly initialized in full. Besides, we introduce a new evaluation metric, Least Coverage Rate, to better evaluate the quality of generated samples. The experimental results demonstrate that InitialGAN outperforms both MLE and other compared models. To the best of our knowledge, it is the first time a language GAN can outperform MLE without using any pre-training techniques.