Matthew Lease’s research while affiliated with University of Texas at Austin and other places

Introduction. Optimizing NLP models for fairness poses many challenges. Lack of differentiable fairness measures prevents gradient-based loss training or requires surrogate losses that diverge from the true metric of interest. In addition, competing objectives (e.g., accuracy vs. fairness) often require making trade-offs based on stakeholder preferences, but stakeholders may not know their preferences before seeing system performance under different trade-off settings. Method. We formulate the GAP loss, a differentiable version of a fairness measure, Accuracy Parity, to provide balanced accuracy across binary demographic groups. Analysis. We show how model-agnostic, HyperNetwork optimization can efficiently train arbitrary NLP model architectures to learn Pareto-optimal trade-offs between competing metrics like predictive performance vs. group fairness. Results. Focusing on the task of toxic language detection, we show the generality and efficacy of our proposed GAP loss function across two datasets, three neural architectures, and three fairness loss functions. Conclusions. Our GAP loss for the task of TL detection demonstrates promising results - improved fairness and computational efficiency. Our work can be extended to other tasks, datasets, and neural models in any practical situation where ensuring equal accuracy across different demographic groups is a desired objective.

Given the massive volume of potentially false claims circulating online, claim prioritization is essential in allocating limited human resources available for fact-checking. In this study, we perceive claim prioritization as an information retrieval (IR) task: just as multidimensional IR relevance, with many factors influencing which search results a user deems relevant, checkworthiness is also multi-faceted, subjective, and even personal, with many factors influencing how fact-checkers triage and select which claims to check. Our study investigates both the multidimensional nature of checkworthiness and effective tool support to assist fact-checkers in claim prioritization. Methodologically, we pursue Research through Design combined with mixed-method evaluation. We develop an AI-assisted claim prioritization prototype as a probe to explore how fact-checkers use multidimensional checkworthiness factors in claim prioritization, simultaneously probing fact-checker needs while also exploring the design space to meet those needs. Our study with 16 professional fact-checkers investigates: 1) how participants assessed the relative importance of different checkworthy dimensions and apply different priorities in claim selection; 2) how they created customized GPT-based search filters and the corresponding benefits and limitations; and 3) their overall user experiences with our prototype. Our work makes a conceptual contribution between multidimensional IR relevance and fact-checking checkworthiness, with findings demonstrating the value of corresponding tooling support. Specifically, we uncovered a hierarchical prioritization strategy fact-checkers implicitly use, revealing an underexplored aspect of their workflow, with actionable design recommendations for improving claim triage across multi-dimensional checkworthiness and tailoring this process with LLM integration.

Large language models (LLMs) are enabling designers to give life to exciting new user experiences for information access. In this work, we present a system that generates LLM personas to debate a topic of interest from different perspectives. How might information seekers use and benefit from such a system? Can centering information access around diverse viewpoints help to mitigate thorny challenges like confirmation bias in which information seekers over-trust search results matching existing beliefs? How do potential biases and hallucinations in LLMs play out alongside human users who are also fallible and possibly biased? Our study exposes participants to multiple viewpoints on controversial issues via a mixed-methods, within-subjects study. We use eye-tracking metrics to quantitatively assess cognitive engagement alongside qualitative feedback. Compared to a baseline search system, we see more creative interactions and diverse information-seeking with our multi-persona debate system, which more effectively reduces user confirmation bias and conviction toward their initial beliefs. Overall, our study contributes to the emerging design space of LLM-based information access systems, specifically investigating the potential of simulated personas to promote greater exposure to information diversity, emulate collective intelligence, and mitigate bias in information seeking.

While many Natural Language Processing (NLP) techniques have been proposed for fact-checking, both academic research and fact-checking organizations report limited adoption of such NLP work due to poor alignment with fact-checker practices, values, and needs. To address this, we investigate a co-design method, Matchmaking for AI, to enable fact-checkers, designers, and NLP researchers to collaboratively identify what fact-checker needs should be addressed by technology, and to brainstorm ideas for potential solutions. Co-design sessions we conducted with 22 professional fact-checkers yielded a set of 11 design ideas that offer a "north star'', integrating fact-checker criteria into novel NLP design concepts. These concepts range from pre-bunking misinformation, efficient and personalized monitoring misinformation, proactively reducing fact-checker potential biases, and collaborative writing fact-check reports. Our work provides new insights into both human-centered fact-checking research and practice and AI co-design research.

In algorithmic toxicity detection pipelines, it is important to identify which demographic group(s) are the subject of a post, a task commonly known as \textit{target (group) detection}. While accurate detection is clearly important, we further advocate a fairness objective: to provide equal protection to all groups who may be targeted. To this end, we adopt \textit{Accuracy Parity} (AP) -- balanced detection accuracy across groups -- as our fairness objective. However, in order to align model training with our AP fairness objective, we require an equivalent loss function. Moreover, for gradient-based models such as neural networks, this loss function needs to be differentiable. Because no such loss function exists today for AP, we propose \emph{Group Accuracy Parity} (GAP): the first differentiable loss function having a one-on-one mapping to AP. We empirically show that GAP addresses disparate impact on groups for target detection. Furthermore, because a single post often targets multiple groups in practice, we also provide a mathematical extension of GAP to larger multi-group settings, something typically requiring heuristics in prior work. Our findings show that by optimizing AP, GAP better mitigates bias in comparison with other commonly employed loss functions.

Human annotations are vital to supervised learning, yet annotators often disagree on the correct label, especially as annotation tasks increase in complexity. A common strategy to improve label quality is to ask multiple annotators to label the same item and then aggregate their labels. To date, many aggregation models have been proposed for simple categorical or numerical annotation tasks, but far less work has considered more complex annotation tasks, such as those involving open-ended, multivariate, or structured responses. Similarly, while a variety of bespoke models have been proposed for specific tasks, our work is the first we are aware of to introduce aggregation methods that generalize across many, diverse complex tasks, including sequence labeling, translation, syntactic parsing, ranking, bounding boxes, and keypoints. This generality is achieved by applying readily available task-specific distance functions, then devising a task-agnostic method to model these distances between labels, rather than the labels themselves. This article presents a unified treatment of our prior work on complex annotation modeling and extends that work with investigation of three new research questions. First, how do complex annotation task and dataset properties impact aggregation accuracy? Second, how should a task owner navigate the many modeling choices in order to maximize aggregation accuracy? Finally, what tests and diagnoses can verify that aggregation models are specified correctly for the given data? To understand how various factors impact accuracy and to inform model selection, we conduct large-scale simulation studies and broad experiments on real, complex datasets. Regarding testing, we introduce the concept of unit tests for aggregation models and present a suite of such tests to ensure that a given model is not mis-specified and exhibits expected behavior. Beyond investigating these research questions above, we discuss the foundational concept and nature of annotation complexity, present a new aggregation model as a conceptual bridge between traditional models and our own, and contribute a new general semisupervised learning method for complex label aggregation that outperforms prior work.