Margaret Mitchell, Giuseppe Attanasio, Ioana Baldini, Miruna Clinciu, Jordan Clive, Pieter Delobelle, Manan Dey, Sil Hamilton, Timm Dill, Jad Doughman, Ritam Dutt, Avijit Ghosh, Jessica Zosa Forde, Carolin Holtermann, Lucie-Aimée Kaffee, Tanmay Laud, Anne Lauscher, Roberto L Lopez-Davila, Maraim Masoud, Nikita Nangia, Anaelia Ovalle, Giada Pistilli, Dragomir Radev, Beatrice Savoldi, Vipul Raheja, Jeremy Qin, Esther Ploeger, Arjun Subramonian, Kaustubh Dhole, Kaiser Sun, Amirbek Djanibekov, Jonibek Mansurov, Kayo Yin, Emilio Villa Cueva, Sagnik Mukherjee, Jerry Huang, Xudong Shen, Jay Gala, Hamdan Al-Ali, Tair Djanibekov, Nurdaulet Mukhituly, Shangrui Nie, Shanya Sharma, Karolina Stanczak, Eliza Szczechla, Tiago Timponi Torrent, Deepak Tunuguntla, Marcelo Viridiano, Oskar Van Der Wal, Adina Yakefu, Aurélie Névéol, Mike Zhang, Sydney Zink, Zeerak Talat. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

In AI research and practice, rigor remains largely understood in terms of methodological rigor---such as whether mathematical, statistical, or computational methods are correctly applied. We argue that this narrow conception of rigor has contributed to the concerns raised by the responsible AI community, including overblown claims about the capabilities of AI systems. Our position is that a broader conception of what rigorous AI research and practice should entail is needed. We believe such a conception---in addition to a more expansive understanding of 1) methodological rigor---should include aspects related to 2) what background knowledge informs what to work on (epistemic rigor); 3) how disciplinary, community, or personal norms, standards, or beliefs influence the work (normative rigor); 4) how clearly articulated the theoretical constructs under use are (conceptual rigor); 5) what is reported and how (reporting rigor); and 6) how well-supported the inferences from existing evidence are (interpretative rigor). In doing so, we also provide useful language and a framework for much needed dialogue about the AI community's work by researchers, policymakers, journalists, and other stakeholders.

This paper introduces the "CIVICS: Culturally-Informed \& Values-Inclusive Corpus for Societal impacts" dataset, designed to evaluate the social and cultural variation of Large Language Models (LLMs) towards socially sensitive topics across multiple languages and cultures. The hand-crafted, multilingual dataset of statements addresses value-laden topics, including LGBTQI rights, social welfare, immigration, disability rights, and surrogacy. CIVICS is designed to elicit responses from LLMs to shed light on how values encoded in their parameters shape their behaviors. Through our dynamic annotation processes, tailored prompt design, and experiments, we investigate how open-weight LLMs respond to these issues, exploring their behavior across diverse linguistic and cultural contexts. Using two experimental set-ups based on log-probabilities and long-form responses, we show social and cultural variability across different LLMs. Specifically, different topics and sources lead to more pronounced differences across model answers, particularly on immigration, LGBTQI rights, and social welfare. Experiments on generating long-form responses from models tuned for user chat demonstrate that refusals are triggered disparately across different models, but consistently and more frequently in English or translated statements. As shown by our initial experimentation, the CIVICS dataset can serve as a tool for future research, promoting reproducibility and transparency across broader linguistic settings, and furthering the development of AI technologies that respect and reflect global cultural diversities and value pluralism. The CIVICS dataset and tools are made available under open licenses at hf.co/CIVICS-dataset.

The performance of a large language model (LLM) depends heavily on the quality and size of its pretraining dataset. However, the pretraining datasets for state-of-the-art open LLMs like Llama 3 and Mixtral are not publicly available and very little is known about how they were created. In this work, we introduce FineWeb, a 15-trillion token dataset derived from 96 Common Crawl snapshots that produces better-performing LLMs than other open pretraining datasets. To advance the understanding of how best to curate high-quality pretraining datasets, we carefully document and ablate all of the design choices used in FineWeb, including in-depth investigations of deduplication and filtering strategies. In addition, we introduce FineWeb-Edu, a 1.3-trillion token collection of educational text filtered from FineWeb. LLMs pretrained on FineWeb-Edu exhibit dramatically better performance on knowledge- and reasoning-intensive benchmarks like MMLU and ARC. Along with our datasets, we publicly release our data curation codebase and all of the models trained during our ablation experiments.

As machine learning-enabled Text-to-Image (TTI) systems are becoming increasingly prevalent and seeing growing adoption as commercial services, characterizing the social biases they exhibit is a necessary first step to lowering their risk of discriminatory outcomes. This evaluation, however, is made more difficult by the synthetic nature of these systems’ outputs: common definitions of diversity are grounded in social categories of people living in the world, whereas the artificial depictions of fictive humans created by these systems have no inherent gender or ethnicity. To address this need, we propose a new method for exploring the social biases in TTI systems. Our approach relies on characterizing the variation in generated images triggered by enumerating gender and ethnicity markers in the prompts, and comparing it to the variation engendered by spanning different professions. This allows us to (1) identify specific bias trends, (2) provide targeted scores to directly compare models in terms of diversity and representation, and (3) jointly model interdependent social variables to support a multidimensional analysis. We leverage this method to analyze images generated by 3 popular TTI systems (Dall·E 2 , Stable Diffusion v 1.4 and 2) and find that while all of their outputs show correlations with US labor demographics, they also consistently under-represent marginalized identities to different extents. We also release the datasets and low-code interactive bias exploration platforms developed forthis work, as well as the necessary tools to similarly evaluate additional TTI systems.

As language models grow ever larger, the need for large-scale high-quality text datasets has never been more pressing, especially in multilingual settings. The BigScience workshop, a 1-year international and multidisciplinary initiative, was formed with the goal of researching and training large language models as a values-driven undertaking, putting issues of ethics, harm, and governance in the foreground. This paper documents the data creation and curation efforts undertaken by BigScience to assemble the Responsible Open-science Open-collaboration Text Sources (ROOTS) corpus, a 1.6TB dataset spanning 59 languages that was used to train the 176-billion-parameter BigScience Large Open-science Open-access Multilingual (BLOOM) language model. We further release a large initial subset of the corpus and analyses thereof, and hope to empower large-scale monolingual and multilingual modeling projects with both the data and the processing tools, as well as stimulate research around this large multilingual corpus.

The measurement of bias in machine learning often focuses on model performance across identity subgroups (such as man and woman) with respect to groundtruth labels. However, these methods do not directly measure the associations that a model may have learned, for example between labels and identity subgroups. Further, measuring a model's bias requires a fully annotated evaluation dataset which may not be easily available in practice.

Jesse Dodge, Maarten Sap, Ana Marasović, William Agnew, Gabriel Ilharco, Dirk Groeneveld, Margaret Mitchell, Matt Gardner. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. 2021.

The ethical concept of fairness has recently been applied in machine learning (ML) settings to describe a wide range of constraints and objectives. When considering the relevance of ethical concepts to subset selection problems, the concepts of diversity and inclusion are additionally applicable in order to create outputs that account for social power and access differentials. We introduce metrics based on these concepts, which can be applied together, separately, and in tandem with additional fairness constraints. Results from human subject experiments lend support to the proposed criteria. Social choice methods can additionally be leveraged to aggregate and choose preferable sets, and we detail how these may be applied.

Although essential to revealing biased performance, well intentioned attempts at algorithmic auditing can have effects that may harm the very populations these measures are meant to protect. This concern is even more salient while auditing biometric systems such as facial recognition, where the data is sensitive and the technology is often used in ethically questionable manners. We demonstrate a set of fiveethical concerns in the particular case of auditing commercial facial processing technology, highlighting additional design considerations and ethical tensions the auditor needs to be aware of so as not exacerbate or complement the harms propagated by the audited system. We go further to provide tangible illustrations of these concerns, and conclude by reflecting on what these concerns mean for the role of the algorithmic audit and the fundamental product limitations they reveal.

Vinodkumar Prabhakaran, Ben Hutchinson, Margaret Mitchell. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Researchers and practitioners from different disciplines have highlighted the ethical and legal challenges posed by the use of machine learned models and data-driven systems, and the potential for such systems to discriminate against certain population groups, due to biases in algorithmic decision-making systems. This tutorial aims to present an overview of algorithmic bias / discrimination issues observed over the last few years and the lessons learned, key regulations and laws, and evolution of techniques for achieving fairness in machine learning systems. We will motivate the need for adopting a "fairness-first" approach (as opposed to viewing algorithmic bias / fairness considerations as an afterthought), when developing machine learning based models and systems for different consumer and enterprise applications. Then, we will focus on the application of fairness-aware machine learning techniques in practice, by highlighting industry best practices and case studies from different technology companies. Based on our experiences in industry, we will identify open problems and research challenges for the data mining / machine learning community.

Researchers and practitioners from different disciplines have highlighted the ethical and legal challenges posed by the use of machine learned models and data-driven systems, and the potential for such systems to discriminate against certain population groups, due to biases in algorithmic decision-making systems. This tutorial aims to present an overview of algorithmic bias / discrimination issues observed over the last few years and the lessons learned, key regulations and laws, and evolution of techniques for achieving fairness in machine learning systems. We will motivate the need for adopting a "fairness-first" approach (as opposed to viewing algorithmic bias / fairness considerations as an afterthought), when developing machine learning based models and systems for different consumer and enterprise applications. Then, we will focus on the application of fairness-aware machine learning techniques in practice, by presenting case studies from different technology companies. Based on our experiences in industry, we will identify open problems and research challenges for the data mining / machine learning community.

Machine learning is a tool for building models that accurately represent input training data. When undesired biases concerning demographic groups are in the training data, well-trained models will reflect those biases. We present a framework for mitigating such biases by including a variable for the group of interest and simultaneously learning a predictor and an adversary. The input to the network X, here text or census data, produces a prediction Y, such as an analogy completion or income bracket, while the adversary tries to model a protected variable Z, here gender or zip code. The objective is to maximize the predictor's ability to predict Y while minimizing the adversary's ability to predict Z. Applied to analogy completion, this method results in accurate predictions that exhibit less evidence of stereotyping Z. When applied to a classification task using the UCI Adult (Census) Dataset, it results in a predictive model that does not lose much accuracy while achieving very close to equality of odds (Hardt, et al., 2016). The method is flexible and applicable to multiple definitions of fairness as well as a wide range of gradient-based learning models, including both regression and classification tasks.

We introduce initial groundwork for estimating suicide risk and mental health in a deep learning framework.By modeling multiple conditions, the system learns to make predictions about suicide risk and mental health at a low false positive rate.Conditions are modeled as tasks in a multitask learning (MTL) framework, with gender prediction as an additional auxiliary task.We demonstrate the effectiveness of multi-task learning by comparison to a well-tuned single-task baseline with the same number of parameters.Our best MTL model predicts potential suicide attempt, as well as the presence of atypical mental health, with AUC > 0.8.We also find additional large improvements using multi-task learning on mental health tasks with limited training data.* Now at Google Research. 1 https://www.nami.org/Learn-More/Mental-Health-Conditions/Related-Conditions/Suicide#sthash.dMAhrKTU.dpuf 2 Communication with clinicians at the 2016 JSALT workshop (Hollingshead, 2016).

Mobile and location-based social media applications provide platforms for users to share brief opinions about products, venues, and services. These quickly typed opinions, or microreviews, are a valuable source of current sentiment on a wide variety of subjects. However, there is currently little research on how to mine this information to present it back to users in easily consumable way. In this paper, we introduce the task of microsummarization, which combines sentiment analysis, summarization, and entity recognition in order to surface key content to users. We explore unsupervised and supervised methods for this task, and find we can reliably extract relevant entities and the sentiment targeted towards them using crowdsourced labels as supervision. In an end-to-end evaluation, we find our best-performing system is vastly preferred by judges over a traditional extractive summarization approach. This work motivates an entirely new approach to summarization, incorporating both sentiment analysis and item extraction for modernized, at-a-glance presentation of public opinion.

Nasrin Mostafazadeh, Ishan Misra, Jacob Devlin, Margaret Mitchell, Xiaodong He, Lucy Vanderwende. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2016.

When human annotators are given a choice about what to label in an image, they apply their own subjective judgments on what to ignore and what to mention. We refer to these noisy "human-centric" annotations as exhibiting human reporting bias. Examples of such annotations include image tags and keywords found on photo sharing sites, or in datasets containing image captions. In this paper, we use these noisy annotations for learning visually correct image classifiers. Such annotations do not use consistent vocabulary, and miss a significant amount of the information present in an image, however, we demonstrate that the noise in these annotations exhibits structure and can be modeled. We propose an algorithm to decouple the human reporting bias from the correct visually grounded labels. Our results are highly interpretable for reporting "what's in the image" versus "what's worth saying." We demonstrate the algorithm's efficacy along a variety of metrics and datasets, including MS COCO and Yahoo Flickr 100M.We show significant improvements over traditional algorithms for both image classification and image captioning, doubling the performance of existing methods in some cases.

Ting-Hao Kenneth Huang, Francis Ferraro, Nasrin Mostafazadeh, Ishan Misra, Aishwarya Agrawal, Jacob Devlin, Ross Girshick, Xiaodong He, Pushmeet Kohli, Dhruv Batra, C. Lawrence Zitnick, Devi Parikh, Lucy Vanderwende, Michel Galley, Margaret Mitchell. Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2016.

This paper presents a novel approach for automatically generating image descriptions: visual detectors, language models, and multimodal similarity models learnt directly from a dataset of image captions. We use multiple instance learning to train visual detectors for words that commonly occur in captions, including many different parts of speech such as nouns, verbs, and adjectives. The word detector outputs serve as conditional inputs to a maximum-entropy language model. The language model learns from a set of over 400,000 image descriptions to capture the statistics of word usage. We capture global semantics by re-ranking caption candidates using sentence-level features and a deep multimodal similarity model. Our system is state-of-the-art on the official Microsoft COCO benchmark, producing a BLEU-4 score of 29.1%. When human judges compare the system captions to ones written by other people on our held-out test set, the system captions have equal or better quality 34% of the time.

Francis Ferraro, Nasrin Mostafazadeh, Ting-Hao Huang, Lucy Vanderwende, Jacob Devlin, Michel Galley, Margaret Mitchell. Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing. 2015.

We propose the task of free-form and open-ended Visual Question Answering (VQA). Given an image and a natural language question about the image, the task is to provide an accurate natural language answer. Mirroring real-world scenarios, such as helping the visually impaired, both the questions and answers are open-ended. Visual questions selectively target different areas of an image, including background details and underlying context. As a result, a system that succeeds at VQA typically needs a more detailed understanding of the image and complex reasoning than a system producing generic image captions. Moreover, VQA is amenable to automatic evaluation, since many open-ended answers contain only a few words or a closed set of answers that can be provided in a multiple-choice format. We provide a dataset containing ~0.25M images, ~0.76M questions, and ~10M answers (www.visualqa.org), and discuss the information it provides. Numerous baselines for VQA are provided and compared with human performance.

Alessandro Sordoni, Michel Galley, Michael Auli, Chris Brockett, Yangfeng Ji, Margaret Mitchell, Jian-Yun Nie, Jianfeng Gao, Bill Dolan. Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2015.

We explore the extent to which highresource manual annotations such as treebanks are necessary for the task of semantic role labeling (SRL).We examine how performance changes without syntactic supervision, comparing both joint and pipelined methods to induce latent syntax.This work highlights a new application of unsupervised grammar induction and demonstrates several approaches to SRL in the absence of supervised syntax.Our best models obtain competitive results in the high-resource setting and state-ofthe-art results in the low resource setting, reaching 72.48% F1 averaged across languages.We release our code for this work along with a larger toolkit for specifying arbitrary graphical structure.1

We explore the use of crowdsourcing to generate natural language in spoken dia-logue systems. We introduce a method-ology to elicit novel templates from the crowd based on a dialogue seed corpus, and investigate the effect that the amount of surrounding dialogue context has on the generation task. Evaluation is performed both with a crowd and with a system de-veloper to assess the naturalness and suit-ability of the elicited phrases. Results indi-cate that the crowd is able to provide rea-sonable and diverse templates within this methodology. More work is necessary be-fore elicited templates can be automati-cally plugged into the system. 1

We propose a novel approach to sentiment analysis for a low resource setting.The intuition behind this work is that sentiment expressed towards an entity, targeted sentiment, may be viewed as a span of sentiment expressed across the entity.This representation allows us to model sentiment detection as a sequence tagging problem, jointly discovering people and organizations along with whether there is sentiment directed towards them.We compare performance in both Spanish and English on microblog data, using only a sentiment lexicon as an external resource.By leveraging linguisticallyinformed features within conditional random fields (CRFs) trained to minimize empirical risk, our best models in Spanish significantly outperform a strong baseline, and reach around 90% accuracy on the combined task of named entity recognition and sentiment prediction.Our models in English, trained on a much smaller dataset, are not yet statistically significant against their baselines.

What do people care about in an image? To drive computational visual recognition toward more human-centric outputs, we need a better understanding of how people perceive and judge the importance of content in images. In this paper, we explore how a number of factors relate to human perception of importance. Proposed factors fall into 3 broad types: 1) factors related to composition, e.g. size, location, 2) factors related to semantics, e.g. category of object or scene, and 3) contextual factors related to the likelihood of attribute-object, or object-scene pairs. We explore these factors using what people describe as a proxy for importance. Finally, we build models to predict what will be described about an image given either known image content, or image content estimated automatically by recognition systems.

Spoken responses produced by subjects during neuropsychological exams can provide diagnostic markers beyond exam performance. In particular, characteristics of the spoken language itself can discriminate between subject groups. We present results on the utility of such markers in discriminating between healthy elderly subjects and subjects with mild cognitive impairment (MCI). Given the audio and transcript of a spoken narrative recall task, a range of markers are automatically derived. These markers include speech features such as pause frequency and duration, and many linguistic complexity measures. We examine measures calculated from manually annotated time alignments (of the transcript with the audio) and syntactic parse trees, as well as the same measures calculated from automatic (forced) time alignments and automatic parses. We show statistically significant differences between clinical subject groups for a number of measures. These differences are largely preserved with automation. We then present classification results, and demonstrate a statistically significant improvement in the area under the ROC curve (AUC) when using automatic spoken language derived features in addition to the neuropsychological test scores. Our results indicate that using multiple, complementary measures can aid in automatic detection of MCI.