Rob Voigt
Assistant Professor of Linguistics and of Computer Science (by courtesy)
Computational linguist Rob Voigt develops natural language processing and machine learning methods to better understand the linguistic mechanisms of the social world. At the heart of his research program lies the simple but powerful insight that large-scale social structures—both the recognizable structures of institutions, organizations, and social groups, but also more diffuse structures such as social biases, polarization, and intergroup conflict—are ultimately the aggregation of many person-to-person interactions, often primarily conducted in the medium of language. Linguistics provides a theoretical toolkit to unpack the behaviors that make up these interactions, and computational methods offer a framework for scaling up to understand the effects of context and social factors like race and gender.
His work addresses a broad range of questions related to bias and social conflict, including racial disparities in police language, media depictions of social groups, and the role of body language, gesture, and the voice in communication and miscommunication. Voigt's work has been published in diverse interdisciplinary venues such as the Journal of Sociolinguistics, the Proceedings of the National Academy of Sciences, and Sociological Science in addition to fundamental contributions to discipline-specific venues associated with the Association for Computational Linguistics. His work on analyzing police language from body camera footage in particular has received multiple awards including the Cozzarelli Prize for best paper in the behavioral and social sciences at PNAS in 2017.
Current Research
Racial Disparities in Media Depictions of Gun Violence. How does the neighborhood in which an incident of gun violence takes place affect how it is portrayed in the media? This project aims to conduct a comprehensive, large-scale investigation of media depictions of gun violence, curating an extensive data-linking effort between print and TV news reporting on incidents of gun violence and metadata from the Gun Violence Archive. In collaboration with IPR director Andrew Papachristos, Voigt and colleagues examine the potential for racial and income-based disparities in both levels of coverage as well as the content and linguistic framing of these incidents.
Transparency Interventions in Policing. This ongoing project funded by the National Institute of Justice in collaboration with criminologists at Stockton University centers on a randomized controlled trial to measure the impacts of a transparency intervention in traffic stops in Atlantic City, New Jersey. Voigt is developing computational methods both to quantify adherence to the intervention script in the experimental condition as well as to measure downstream impacts on interaction quality and the police-community relationship more broadly.
Multimodal Machine Learning to Understand 911 Emergency Response. In a new collaboration with the Health Lab at the University of Chicago, Voigt is working to establish a baseline for understanding the critical role of language in the split-second decision-making of emergency response dispatchers. Collecting at least 500,000 recordings of 911 calls and associated metadata from five study site partners, the team aims to develop text-and-audio models for key interactional factors—such as distrust and disbelief, communicative misunderstandings, and respectfulness—and use them to analyze the and as well as measure potential disparity across neighborhoods and social groups.
Computational Linguistics and Mental Health. In a series of collaborations with domain experts including IPR associates Molly Losh and Vijay Mittal, Voigt is examining how to leverage the power of contemporary large language models to better understand language impairments in mental health disorders. Specific projects include cross-linguistic comparisons of autism spectrum disorder between speakers of English and Cantonese and using computer vision techniques to measure gestural markers of schizophrenia.
Selected Publications
Camp, N., R. Voigt, M. G. Hamedani, D. Jurafsky, and J. L. Eberhardt. Forthcoming. Leveraging body-worn camera footage to assess effects of training on officer communication during traffic stops. PNAS Nexus.
Camp, N., and R. Voigt. Forthcoming. Body Camera Footage as Data: Using Natural Language Processing to Capture Policing At Scale and In Depth. Behavioral Science & Policy.
Abramitzky, R., L. Boustan, P. Catron, D. Connor and R. Voigt. 2023. The Refugee Advantage: English-Language Attainment in the Early Twentieth Century. Sociological Science 10: 769–805.
Heddaya, M., S. Dworkin, C. Tan, R. Voigt, and A. Zentefis. 2023. The Language of Bargaining. Proceedings of the Association for Computational Linguistics 1: 13161–85.
Card, D., S. Chang, C. Becker, J. Mendelsohn, R. Voigt, L. Boustan, R. Abramitzky, and D. Jurafsky. 2022. Computational analysis of 140 years of US political speeches reveals more positive but increasingly polarized framing of immigration. The Proceedings of the National Academy of Sciences 119 (31): e2120510119.
Camp, N., R. Voigt, D. Jurafsky, and J. L. Eberhardt. 2021. The Thin Blue Waveform: Racial Disparities in Officer Prosody Undermine Institutional Trust in the Police. Journal of Personality and Social Psychology 121(6):1157–71.
