The widespread use of large language models has brought up essential questions about the potential biases these models might learn. This led to the development of several metrics aimed at evaluating and mitigating these biases. In this paper, we first demonstrate that prompt-based fairness metrics exhibit poor agreement, as measured by correlation, raising important questions about the reliability of fairness assessment using prompts. Then, we outline six relevant reasons why such a low correlation is observed across existing metrics. Based on these insights, we propose a method called Correlated Fairness Output (CAIRO) to enhance the correlation between fairness metrics. CAIRO augments the original prompts of a given fairness metric by using several pre-trained language models and then selects the combination of the augmented prompts that achieves the highest correlation across metrics. We show a significant improvement in Pearson correlation from 0.3 and 0.18 to 0.90 and 0.98 across metrics for gender and religion biases, respectively. Our code is available at https://github.com/chandar-lab/CAIRO.

We propose a family of metrics to assess language generation derived from population estimation methods widely used in ecology. More specifically, we use mark-recapture and maximum-likelihood methods that have been applied over the past several decades to estimate the size of closed populations in the wild. We propose three novel metrics: ME_Petersen and ME_CAPTURE, which retrieve a single-valued assessment, and ME_Schnabel which returns a double-valued metric to assess the evaluation set in terms of quality and diversity, separately. In synthetic experiments, our family of methods is sensitive to drops in quality and diversity. Moreover, our methods show a higher correlation to human evaluation than existing metrics on several challenging tasks, namely unconditional language generation, machine translation, and text summarization.

Language models trained with Maximum Likelihood Estimation (MLE) have been considered as a mainstream solution in Natural Language Generation (NLG) for years. Recently, various approaches with Generative Adversarial Nets (GANs) have also been proposed. While offering exciting new prospects, GANs in NLG by far are nevertheless reportedly suffering from training instability and mode collapse, and therefore outperformed by conventional MLE models. In this work, we propose techniques for improving GANs in NLG, namely Best Student Forcing (BSF), a novel yet simple adversarial training mechanism in which generated sequences of high quality are selected as temporary ground-truth to further train the generator. We also use an ensemble of discriminators to increase training stability and sample diversity. Evaluation shows that the combination of BSF and multiple discriminators consistently performs better than previous GAN approaches over various metrics, and outperforms a baseline MLE in terms of Fr ́ech ́et Distance, a recently proposed metric capturing both sample quality and diversity.