Evaluating Text GANs as Language Models
This work addresses a major hurdle for researchers in text generation by providing a clear evaluation metric, though it is incremental as it adapts existing metrics rather than introducing a new paradigm.
The authors tackled the problem of evaluating text GANs by proposing a method to approximate their generated text distribution, enabling use of traditional language model metrics, and found that current GAN-based models perform substantially worse than state-of-the-art language models.
Generative Adversarial Networks (GANs) are a promising approach for text generation that, unlike traditional language models (LM), does not suffer from the problem of ``exposure bias''. However, A major hurdle for understanding the potential of GANs for text generation is the lack of a clear evaluation metric. In this work, we propose to approximate the distribution of text generated by a GAN, which permits evaluating them with traditional probability-based LM metrics. We apply our approximation procedure on several GAN-based models and show that they currently perform substantially worse than state-of-the-art LMs. Our evaluation procedure promotes better understanding of the relation between GANs and LMs, and can accelerate progress in GAN-based text generation.