Research in NLP is often supported by experimental results, and improved reporting of such results can lead to better understanding and more reproducible science. In this paper we analyze three statistical estimators for expected validation performance, a tool used for reporting performance (e.g., accuracy) as a function of computational budget (e.g., number of hyperparameter tuning experiments). Where previous work analyzing such estimators focused on the bias, we also examine the variance and mean squared error (MSE). In both synthetic and realistic scenarios, we evaluate three estimators and find the unbiased estimator has the highest variance, and the estimator with the smallest variance has the largest bias; the estimator with the smallest MSE strikes a balance between bias and variance, displaying a classic bias-variance tradeoff. We use expected validation performance to compare between different models, and analyze how frequently each estimator leads to drawing incorrect conclusions about which of two models performs best. We find that the two biased estimators lead to the fewest incorrect conclusions, which hints at the importance of minimizing variance and MSE.

Language models (LMs) must be both safe and equitable to be responsibly deployed in practice. With safety in mind, numerous detoxification techniques (e.g., Dathathri et al. 2020; Krause et al. 2020) have been proposed to mitigate toxic LM generations. In this work, we show that these detoxification techniques hurt equity: they decrease the utility of LMs on language used by marginalized groups (e.g., African-American English and minority identity mentions). In particular, we perform automatic and human evaluations of text generation quality when LMs are conditioned on inputs with different dialects and group identifiers. We find that detoxification makes LMs more brittle to distribution shift, especially on language used by marginalized groups. We identify that these failures stem from detoxification methods exploiting spurious correlations in toxicity datasets. Overall, our results highlight the tension between the controllability and distributional robustness of LMs.

Human evaluations are typically considered the gold standard in natural language generation, but as models’ fluency improves, how well can evaluators detect and judge machine-generated text? We run a study assessing non-experts’ ability to distinguish between human- and machine-authored text (GPT2 and GPT3) in three domains (stories, news articles, and recipes). We find that, without training, evaluators distinguished between GPT3- and human-authored text at random chance level. We explore three approaches for quickly training evaluators to better identify GPT3-authored text (detailed instructions, annotated examples, and paired examples) and find that while evaluators’ accuracy improved up to 55%, it did not significantly improve across the three domains. Given the inconsistent results across text domains and the often contradictory reasons evaluators gave for their judgments, we examine the role untrained human evaluations play in NLG evaluation and provide recommendations to NLG researchers for improving human evaluations of text generated from state-of-the-art models.

Pretrained neural language models (LMs) are prone to generating racist, sexist, or otherwise toxic language which hinders their safe deployment. We investigate the extent to which pretrained LMs can be prompted to generate toxic language, and the effectiveness of controllable text generation algorithms at preventing such toxic degeneration. We create and release RealToxicityPrompts, a dataset of 100K naturally occurring, sentence-level prompts derived from a large corpus of English web text, paired with toxicity scores from a widely-used toxicity classifier. Using RealToxicityPrompts, we find that pretrained LMs can degenerate into toxic text even from seemingly innocuous prompts. We empirically assess several controllable generation methods, and find that while data- or compute-intensive methods (e.g., adaptive pretraining on non-toxic data) are more effective at steering away from toxicity than simpler solutions (e.g., banning “bad” words), no current method is failsafe against neural toxic degeneration. To pinpoint the potential cause of such persistent toxic degeneration, we analyze two web text corpora used to pretrain several LMs (including GPT-2; Radford et. al, 2019), and find a significant amount of offensive, factually unreliable, and otherwise toxic content. Our work provides a test bed for evaluating toxic generations by LMs and stresses the need for better data selection processes for pretraining.

Language models pretrained on text from a wide variety of sources form the foundation of today’s NLP. In light of the success of these broad-coverage models, we investigate whether it is still helpful to tailor a pretrained model to the domain of a target task. We present a study across four domains (biomedical and computer science publications, news, and reviews) and eight classification tasks, showing that a second phase of pretraining in-domain (domain-adaptive pretraining) leads to performance gains, under both high- and low-resource settings. Moreover, adapting to the task’s unlabeled data (task-adaptive pretraining) improves performance even after domain-adaptive pretraining. Finally, we show that adapting to a task corpus augmented using simple data selection strategies is an effective alternative, especially when resources for domain-adaptive pretraining might be unavailable. Overall, we consistently find that multi-phase adaptive pretraining offers large gains in task performance.

We introduce VAMPIRE, a lightweight pretraining framework for effective text classification when data and computing resources are limited. We pretrain a unigram document model as a variational autoencoder on in-domain, unlabeled data and use its internal states as features in a downstream classifier. Empirically, we show the relative strength of VAMPIRE against computationally expensive contextual embeddings and other popular semi-supervised baselines under low resource settings. We also find that fine-tuning to in-domain data is crucial to achieving decent performance from contextual embeddings when working with limited supervision. We accompany this paper with code to pretrain and use VAMPIRE embeddings in downstream tasks.

Research in natural language processing proceeds, in part, by demonstrating that new models achieve superior performance (e.g., accuracy) on held-out test data, compared to previous results. In this paper, we demonstrate that test-set performance scores alone are insufficient for drawing accurate conclusions about which model performs best. We argue for reporting additional details, especially performance on validation data obtained during model development. We present a novel technique for doing so: expected validation performance of the best-found model as a function of computation budget (i.e., the number of hyperparameter search trials or the overall training time). Using our approach, we find multiple recent model comparisons where authors would have reached a different conclusion if they had used more (or less) computation. Our approach also allows us to estimate the amount of computation required to obtain a given accuracy; applying it to several recently published results yields massive variation across papers, from hours to weeks. We conclude with a set of best practices for reporting experimental results which allow for robust future comparisons, and provide code to allow researchers to use our technique.

Large-scale datasets for natural language inference are created by presenting crowd workers with a sentence (premise), and asking them to generate three new sentences (hypotheses) that it entails, contradicts, or is logically neutral with respect to. We show that, in a significant portion of such data, this protocol leaves clues that make it possible to identify the label by looking only at the hypothesis, without observing the premise. Specifically, we show that a simple text categorization model can correctly classify the hypothesis alone in about 67% of SNLI (Bowman et. al, 2015) and 53% of MultiNLI (Williams et. al, 2017). Our analysis reveals that specific linguistic phenomena such as negation and vagueness are highly correlated with certain inference classes. Our findings suggest that the success of natural language inference models to date has been overestimated, and that the task remains a hard open problem.