On Accurate Evaluation of GANs for Language Generation

TL;DR

This paper critically evaluates GANs for language generation, highlighting evaluation challenges, proposing better metrics, and showing that GANs do not outperform traditional language models in current settings.

Contribution

It introduces alternative evaluation metrics for GANs in language tasks and provides a comprehensive comparison showing GANs' performance is not superior to standard language models.

Findings

BLEU score is insensitive to semantic quality.

Proposed metrics better capture diversity and quality.

GANs do not outperform traditional language models.

Abstract

Generative Adversarial Networks (GANs) are a promising approach to language generation. The latest works introducing novel GAN models for language generation use n-gram based metrics for evaluation and only report single scores of the best run. In this paper, we argue that this often misrepresents the true picture and does not tell the full story, as GAN models can be extremely sensitive to the random initialization and small deviations from the best hyperparameter choice. In particular, we demonstrate that the previously used BLEU score is not sensitive to semantic deterioration of generated texts and propose alternative metrics that better capture the quality and diversity of the generated samples. We also conduct a set of experiments comparing a number of GAN models for text with a conventional Language Model (LM) and find that neither of the considered models performs convincingly better than the LM.