%0 Generic %1 caccia2018language %A Caccia, Massimo %A Caccia, Lucas %A Fedus, William %A Larochelle, Hugo %A Pineau, Joelle %A Charlin, Laurent %D 2018 %K gan generation neuralnet readinggroup %T Language GANs Falling Short %U http://arxiv.org/abs/1811.02549 %X Generating high-quality text with sufficient diversity is essential for a wide range of Natural Language Generation (NLG) tasks. Maximum-Likelihood (MLE) models trained with teacher forcing have constantly been reported as weak baselines, where poor performance is attributed to exposure bias; at inference time, the model is fed its own prediction instead of a ground-truth token, which can lead to accumulating errors and poor samples. This line of reasoning has led to an outbreak of adversarial based approaches for NLG, on the account that GANs do not suffer from exposure bias. In this work, we make several surprising observations with contradict common beliefs. We first revisit the canonical evaluation framework for NLG, and point out fundamental flaws with quality-only evaluation: we show that one can outperform such metrics using a simple, well-known temperature parameter to artificially reduce the entropy of the model's conditional distributions. Second, we leverage the control over the quality / diversity tradeoff given by this parameter to evaluate models over the whole quality-diversity spectrum, and find MLE models constantly outperform the proposed GAN variants, over the whole quality-diversity space. Our results have several implications: 1) The impact of exposure bias on sample quality is less severe than previously thought, 2) temperature tuning provides a better quality / diversity trade off than adversarial training, while being easier to train, easier to cross-validate, and less computationally expensive.

@misc{caccia2018language, abstract = {Generating high-quality text with sufficient diversity is essential for a wide range of Natural Language Generation (NLG) tasks. Maximum-Likelihood (MLE) models trained with teacher forcing have constantly been reported as weak baselines, where poor performance is attributed to exposure bias; at inference time, the model is fed its own prediction instead of a ground-truth token, which can lead to accumulating errors and poor samples. This line of reasoning has led to an outbreak of adversarial based approaches for NLG, on the account that GANs do not suffer from exposure bias. In this work, we make several surprising observations with contradict common beliefs. We first revisit the canonical evaluation framework for NLG, and point out fundamental flaws with quality-only evaluation: we show that one can outperform such metrics using a simple, well-known temperature parameter to artificially reduce the entropy of the model's conditional distributions. Second, we leverage the control over the quality / diversity tradeoff given by this parameter to evaluate models over the whole quality-diversity spectrum, and find MLE models constantly outperform the proposed GAN variants, over the whole quality-diversity space. Our results have several implications: 1) The impact of exposure bias on sample quality is less severe than previously thought, 2) temperature tuning provides a better quality / diversity trade off than adversarial training, while being easier to train, easier to cross-validate, and less computationally expensive.}, added-at = {2018-11-20T10:49:40.000+0100}, author = {Caccia, Massimo and Caccia, Lucas and Fedus, William and Larochelle, Hugo and Pineau, Joelle and Charlin, Laurent}, biburl = {https://www.bibsonomy.org/bibtex/2ce3813207e0bc255ead5ae80a18bb6bd/albinzehe}, description = {Language GANs Falling Short}, interhash = {4478dff9d2407a53c34d867197610cf7}, intrahash = {ce3813207e0bc255ead5ae80a18bb6bd}, keywords = {gan generation neuralnet readinggroup}, note = {cite arxiv:1811.02549}, timestamp = {2018-11-20T10:49:40.000+0100}, title = {Language GANs Falling Short}, url = {http://arxiv.org/abs/1811.02549}, year = 2018 }