Computational Models of Police-Community Interaction

In this dissertation, I use footage from body-worn cameras to analyze the language of police-community interactions during routine traffic stops in three studies exploring officers' words and prosody as well as the back-and-forth structure of these interactions.

 

In the first study, I implement pragmatic theories of politeness in a computational model of officer respect, informed by a thin-slicing study of participant ratings of officer utterances. Inspection of the weights learned by the model shows shows the strongest influence from negative politeness speech acts like apologizing and reassurance. Using this model I find large-scale evidence that officers speak with consistently less respect towards black versus white community members, even after controlling for contextual factors like the race of the officer and the severity of the infraction. In the second study, I show an analogous racial disparity from prosody alone by obtaining human judgments on content-filtered clips of officer speech. Building a model of human judgments from simple prosodic features I find evidence that officer anxiety is foregrounded as participants judge clips with lower F0, lower intensity, and slower speech rate more highly. In the third study, I develop an ontology of dialog acts unique to the traffic stop interaction, and employ these to show that community members are overwhelmingly compliant in these interactions, but that officers respond differently to mitigation (e.g., justifying fault or mitigating compliance) and rejection (e.g. rejecting fault or deflecting questions) from black and white community members.

 

Such disparities in common, everyday interactions between police and the communities they serve have important implications for procedural justice and the building of police-community trust. This dissertation presents the first systematic linguistic analysis of officer body-worn camera footage, demonstrating that it can be used as a rich source of data rather than merely archival evidence, and paves the way for developing powerful language-based tools for studying and potentially improving police-community relations.

 

(The format for this open part of the oral exam is a 30-45 minute talk by the PhD candidate followed by questions from those attending, for a total of no more than 75 minutes.   Please arrive promptly!)

 

University oral exam committee: Dan Jurafsky (advisor), Jennifer L. Eberhardt, Robert J. Podesva, Penelope Eckert, John Rickford

University oral exam chair: Sharad Goel