Dallas Card

Kaitlyn Zhou, Kawin Ethayarajh, Dallas Card et al. · stanford

Cosine similarity of contextual embeddings is used in many NLP tasks (e.g., QA, IR, MT) and metrics (e.g., BERTScore). Here, we uncover systematic ways in which word similarities estimated by cosine over BERT embeddings are understated and trace this effect to training data frequency. We find that relative to human judgements, cosine similarity underestimates the similarity of frequent words with other instances of the same word or other words across contexts, even after controlling for polysemy and other factors. We conjecture that this underestimation of similarity for high frequency words is due to differences in the representational geometry of high and low frequency words and provide a formal argument for the two-dimensional case.

Junshen K. Chen, Dallas Card, Dan Jurafsky · stanford

Off-the-shelf models are widely used by computational social science researchers to measure properties of text, such as sentiment. However, without access to source data it is difficult to account for domain shift, which represents a threat to validity. Here, we treat domain adaptation as a modular process that involves separate model producers and model consumers, and show how they can independently cooperate to facilitate more accurate measurements of text. We introduce two lightweight techniques for this scenario, and demonstrate that they reliably increase out-of-domain accuracy on four multi-domain text classification datasets when used with linear and contextual embedding models. We conclude with recommendations for model producers and consumers, and release models and replication code to accompany this paper.

Bangzhao Shu, Lechen Zhang, Minje Choi et al.

The versatility of Large Language Models (LLMs) on natural language understanding tasks has made them popular for research in social sciences. To properly understand the properties and innate personas of LLMs, researchers have performed studies that involve using prompts in the form of questions that ask LLMs about particular opinions. In this study, we take a cautionary step back and examine whether the current format of prompting LLMs elicits responses in a consistent and robust manner. We first construct a dataset that contains 693 questions encompassing 39 different instruments of persona measurement on 115 persona axes. Additionally, we design a set of prompts containing minor variations and examine LLMs' capabilities to generate answers, as well as prompt variations to examine their consistency with respect to content-level variations such as switching the order of response options or negating the statement. Our experiments on 17 different LLMs reveal that even simple perturbations significantly downgrade a model's question-answering ability, and that most LLMs have low negation consistency. Our results suggest that the currently widespread practice of prompting is insufficient to accurately and reliably capture model perceptions, and we therefore discuss potential alternatives to improve these issues.

Dallas Card

Measuring semantic change has thus far remained a task where methods using contextual embeddings have struggled to improve upon simpler techniques relying only on static word vectors. Moreover, many of the previously proposed approaches suffer from downsides related to scalability and ease of interpretation. We present a simplified approach to measuring semantic change using contextual embeddings, relying only on the most probable substitutes for masked terms. Not only is this approach directly interpretable, it is also far more efficient in terms of storage, achieves superior average performance across the most frequently cited datasets for this task, and allows for more nuanced investigation of change than is possible with static word vectors.

Sidney Xiang, Nicholas David, Dallas Card et al.

Scientific innovation often comes from researchers who pivot across disciplines. However, prior work found that established researchers face productivity penalties when pivoting. Here, we investigate the consequences of pivoting at the beginning of a research career -- doctoral admissions -- when the benefits of importing new ideas might outweigh the switching costs. Using applications to all PhD programs at a large research-intensive university between 2013-2023, we find that pivoters (those applying to programs outside their prior disciplinary training) have lower GPAs and standardized test scores than non-pivoters. Yet even conditional on these predictors of admission, pivoters are 1.3 percentage points less likely to be admitted. Examining applicants who applied to multiple programs in the same admissions cycle provides suggestive evidence that the admissions pivot penalty is causal. This penalty is significantly smaller for applicants who secure a recommendation from someone within the target discipline. Among those admitted and enrolled, pivoters are 12.9 percentage points less likely to graduate and do not show superior publication performance on average or at the tail. Our results reveal the substantial costs of disciplinary pivoting even at the outset of research careers, which constrain the flow of new ideas into research communities.

Benjamin Litterer, David Jurgens, Dallas Card

Podcasts provide highly diverse content to a massive listener base through a unique on-demand modality. However, limited data has prevented large-scale computational analysis of the podcast ecosystem. To fill this gap, we introduce a massive dataset of over 1.1M podcast transcripts that is largely comprehensive of all English language podcasts available through public RSS feeds from May and June of 2020. This data is not limited to text, but rather includes audio features and speaker turns for a subset of 370K episodes, and speaker role inferences and other metadata for all 1.1M episodes. Using this data, we also conduct a foundational investigation into the content, structure, and responsiveness of this ecosystem. Together, our data and analyses open the door to continued computational research of this popular and impactful medium.

Benjamin Litterer, David Jurgens, Dallas Card

Most events in the world receive at most brief coverage by the news media. Occasionally, however, an event will trigger a media storm, with voluminous and widespread coverage lasting for weeks instead of days. In this work, we develop and apply a pairwise article similarity model, allowing us to identify story clusters in corpora covering local and national online news, and thereby create a comprehensive corpus of media storms over a nearly two year period. Using this corpus, we investigate media storms at a new level of granularity, allowing us to validate claims about storm evolution and topical distribution, and provide empirical support for previously hypothesized patterns of influence of storms on media coverage and intermedia agenda setting.

Julia Mendelsohn, Maya Vijan, Dallas Card et al.

Social media enables activists to directly communicate with the public and provides a space for movement leaders, participants, bystanders, and opponents to collectively construct and contest narratives. Focusing on Twitter messages from social movements surrounding three issues in 2018-2019 (guns, immigration, and LGBTQ rights), we create a codebook, annotated dataset, and computational models to detect diagnostic (problem identification and attribution), prognostic (proposed solutions and tactics), and motivational (calls to action) framing strategies. We conduct an in-depth unsupervised linguistic analysis of each framing strategy, and uncover cross-movement similarities in associations between framing and linguistic features such as pronouns and deontic modal verbs. Finally, we compare framing strategies across issues and other social, cultural, and interactional contexts. For example, we show that diagnostic framing is more common in replies than original broadcast posts, and that social movement organizations focus much more on prognostic and motivational framing than journalists and ordinary citizens.

Suchin Gururangan, Dallas Card, Sarah K. Dreier et al.

Language models increasingly rely on massive web dumps for diverse text data. However, these sources are rife with undesirable content. As such, resources like Wikipedia, books, and newswire often serve as anchors for automatically selecting web text most suitable for language modeling, a process typically referred to as quality filtering. Using a new dataset of U.S. high school newspaper articles -- written by students from across the country -- we investigate whose language is preferred by the quality filter used for GPT-3. We find that newspapers from larger schools, located in wealthier, educated, and urban ZIP codes are more likely to be classified as high quality. We then demonstrate that the filter's measurement of quality is unaligned with other sensible metrics, such as factuality or literary acclaim. We argue that privileging any corpus as high quality entails a language ideology, and more care is needed to construct training corpora for language models, with better transparency and justification for the inclusion or exclusion of various texts.

Jesse Dodge, Suchin Gururangan, Dallas Card et al.

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.

Rishi Bommasani, Drew A. Hudson, Ehsan Adeli et al.

AI is undergoing a paradigm shift with the rise of models (e.g., BERT, DALL-E, GPT-3) that are trained on broad data at scale and are adaptable to a wide range of downstream tasks. We call these models foundation models to underscore their critically central yet incomplete character. This report provides a thorough account of the opportunities and risks of foundation models, ranging from their capabilities (e.g., language, vision, robotics, reasoning, human interaction) and technical principles(e.g., model architectures, training procedures, data, systems, security, evaluation, theory) to their applications (e.g., law, healthcare, education) and societal impact (e.g., inequity, misuse, economic and environmental impact, legal and ethical considerations). Though foundation models are based on standard deep learning and transfer learning, their scale results in new emergent capabilities,and their effectiveness across so many tasks incentivizes homogenization. Homogenization provides powerful leverage but demands caution, as the defects of the foundation model are inherited by all the adapted models downstream. Despite the impending widespread deployment of foundation models, we currently lack a clear understanding of how they work, when they fail, and what they are even capable of due to their emergent properties. To tackle these questions, we believe much of the critical research on foundation models will require deep interdisciplinary collaboration commensurate with their fundamentally sociotechnical nature.

Abeba Birhane, Pratyusha Kalluri, Dallas Card et al.

Machine learning currently exerts an outsized influence on the world, increasingly affecting institutional practices and impacted communities. It is therefore critical that we question vague conceptions of the field as value-neutral or universally beneficial, and investigate what specific values the field is advancing. In this paper, we first introduce a method and annotation scheme for studying the values encoded in documents such as research papers. Applying the scheme, we analyze 100 highly cited machine learning papers published at premier machine learning conferences, ICML and NeurIPS. We annotate key features of papers which reveal their values: their justification for their choice of project, which attributes of their project they uplift, their consideration of potential negative consequences, and their institutional affiliations and funding sources. We find that few of the papers justify how their project connects to a societal need (15\%) and far fewer discuss negative potential (1\%). Through line-by-line content analysis, we identify 59 values that are uplifted in ML research, and, of these, we find that the papers most frequently justify and assess themselves based on Performance, Generalization, Quantitative evidence, Efficiency, Building on past work, and Novelty. We present extensive textual evidence and identify key themes in the definitions and operationalization of these values. Notably, we find systematic textual evidence that these top values are being defined and applied with assumptions and implications generally supporting the centralization of power.Finally, we find increasingly close ties between these highly cited papers and tech companies and elite universities.

Yiwei Luo, Dallas Card, Dan Jurafsky

Citing opinions is a powerful yet understudied strategy in argumentation. For example, an environmental activist might say, "Leading scientists agree that global warming is a serious concern," framing a clause which affirms their own stance ("that global warming is serious") as an opinion endorsed ("[scientists] agree") by a reputable source ("leading"). In contrast, a global warming denier might frame the same clause as the opinion of an untrustworthy source with a predicate connoting doubt: "Mistaken scientists claim [...]." Our work studies opinion-framing in the global warming (GW) debate, an increasingly partisan issue that has received little attention in NLP. We introduce Global Warming Stance Dataset (GWSD), a dataset of stance-labeled GW sentences, and train a BERT classifier to study novel aspects of argumentation in how different sides of a debate represent their own and each other's opinions. From 56K news articles, we find that similar linguistic devices for self-affirming and opponent-doubting discourse are used across GW-accepting and skeptic media, though GW-skeptical media shows more opponent-doubt. We also find that authors often characterize sources as hypocritical, by ascribing opinions expressing the author's own view to source entities known to publicly endorse the opposing view. We release our stance dataset, model, and lexicons of framing devices for future work on opinion-framing and the automatic detection of GW stance.

Reid Pryzant, Dallas Card, Dan Jurafsky et al.

We consider the problem of using observational data to estimate the causal effects of linguistic properties. For example, does writing a complaint politely lead to a faster response time? How much will a positive product review increase sales? This paper addresses two technical challenges related to the problem before developing a practical method. First, we formalize the causal quantity of interest as the effect of a writer's intent, and establish the assumptions necessary to identify this from observational data. Second, in practice, we only have access to noisy proxies for the linguistic properties of interest -- e.g., predictions from classifiers and lexicons. We propose an estimator for this setting and prove that its bias is bounded when we perform an adjustment for the text. Based on these results, we introduce TextCause, an algorithm for estimating causal effects of linguistic properties. The method leverages (1) distant supervision to improve the quality of noisy proxies, and (2) a pre-trained language model (BERT) to adjust for the text. We show that the proposed method outperforms related approaches when estimating the effect of Amazon review sentiment on semi-simulated sales figures. Finally, we present an applied case study investigating the effects of complaint politeness on bureaucratic response times.

Dallas Card, Peter Henderson, Urvashi Khandelwal et al.

Despite its importance to experimental design, statistical power (the probability that, given a real effect, an experiment will reject the null hypothesis) has largely been ignored by the NLP community. Underpowered experiments make it more difficult to discern the difference between statistical noise and meaningful model improvements, and increase the chances of exaggerated findings. By meta-analyzing a set of existing NLP papers and datasets, we characterize typical power for a variety of settings and conclude that underpowered experiments are common in the NLP literature. In particular, for several tasks in the popular GLUE benchmark, small test sets mean that most attempted comparisons to state of the art models will not be adequately powered. Similarly, based on reasonable assumptions, we find that the most typical experimental design for human rating studies will be underpowered to detect small model differences, of the sort that are frequently studied. For machine translation, we find that typical test sets of 2000 sentences have approximately 75% power to detect differences of 1 BLEU point. To improve the situation going forward, we give an overview of best practices for power analysis in NLP and release a series of notebooks to assist with future power analyses.

Dallas Card, Noah A. Smith

Recent work on fairness in machine learning has primarily emphasized how to define, quantify, and encourage "fair" outcomes. Less attention has been paid, however, to the ethical foundations which underlie such efforts. Among the ethical perspectives that should be taken into consideration is consequentialism, the position that, roughly speaking, outcomes are all that matter. Although consequentialism is not free from difficulties, and although it does not necessarily provide a tractable way of choosing actions (because of the combined problems of uncertainty, subjectivity, and aggregation), it nevertheless provides a powerful foundation from which to critique the existing literature on machine learning fairness. Moreover, it brings to the fore some of the tradeoffs involved, including the problem of who counts, the pros and cons of using a policy, and the relative value of the distant future. In this paper we provide a consequentialist critique of common definitions of fairness within machine learning, as well as a machine learning perspective on consequentialism. We conclude with a broader discussion of the issues of learning and randomization, which have important implications for the ethics of automated decision making systems.

Jesse Dodge, Suchin Gururangan, Dallas Card et al.

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.

Suchin Gururangan, Tam Dang, Dallas Card et al.

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.

Dallas Card, Michael Zhang, Noah A. Smith

Recent advances in deep learning have achieved impressive gains in classification accuracy on a variety of types of data, including images and text. Despite these gains, however, concerns have been raised about the calibration, robustness, and interpretability of these models. In this paper we propose a simple way to modify any conventional deep architecture to automatically provide more transparent explanations for classification decisions, as well as an intuitive notion of the credibility of each prediction. Specifically, we draw on ideas from nonparametric kernel regression, and propose to predict labels based on a weighted sum of training instances, where the weights are determined by distance in a learned instance-embedding space. Working within the framework of conformal methods, we propose a new measure of nonconformity suggested by our model, and experimentally validate the accompanying theoretical expectations, demonstrating improved transparency, controlled error rates, and robustness to out-of-domain data, without compromising on accuracy or calibration.

Dallas Card, Chenhao Tan, Noah A. Smith

Most real-world document collections involve various types of metadata, such as author, source, and date, and yet the most commonly-used approaches to modeling text corpora ignore this information. While specialized models have been developed for particular applications, few are widely used in practice, as customization typically requires derivation of a custom inference algorithm. In this paper, we build on recent advances in variational inference methods and propose a general neural framework, based on topic models, to enable flexible incorporation of metadata and allow for rapid exploration of alternative models. Our approach achieves strong performance, with a manageable tradeoff between perplexity, coherence, and sparsity. Finally, we demonstrate the potential of our framework through an exploration of a corpus of articles about US immigration.

Chenhao Tan, Dallas Card, Noah A. Smith

Understanding how ideas relate to each other is a fundamental question in many domains, ranging from intellectual history to public communication. Because ideas are naturally embedded in texts, we propose the first framework to systematically characterize the relations between ideas based on their occurrence in a corpus of documents, independent of how these ideas are represented. Combining two statistics --- cooccurrence within documents and prevalence correlation over time --- our approach reveals a number of different ways in which ideas can cooperate and compete. For instance, two ideas can closely track each other's prevalence over time, and yet rarely cooccur, almost like a "cold war" scenario. We observe that pairwise cooccurrence and prevalence correlation exhibit different distributions. We further demonstrate that our approach is able to uncover intriguing relations between ideas through in-depth case studies on news articles and research papers.