Adversarial Feature Matching for Text Generation

TLDR

Generative Adversarial Networks have succeeded at producing realistic synthetic data, yet their convergence problems and difficulty handling discrete data limit their use for text generation. This work proposes a framework that generates realistic text through adversarial training. The framework employs an LSTM generator and a CNN discriminator, matching high‑dimensional latent feature distributions of real and synthetic sentences with a kernelized discrepancy metric to reduce mode collapse. Experiments demonstrate that the model outperforms baselines and produces realistic‑looking sentences.

Abstract

The Generative Adversarial Network (GAN) has achieved great success in generating realistic (real-valued) synthetic data. However, convergence issues and difficulties dealing with discrete data hinder the applicability of GAN to text. We propose a framework for generating realistic text via adversarial training. We employ a long short-term memory network as generator, and a convolutional network as discriminator. Instead of using the standard objective of GAN, we propose matching the high-dimensional latent feature distributions of real and synthetic sentences, via a kernelized discrepancy metric. This eases adversarial training by alleviating the mode-collapsing problem. Our experiments show superior performance in quantitative evaluation, and demonstrate that our model can generate realistic-looking sentences.