Stanford Linguist Presents On Police Racial Bias In Language
Written by Ross Chapman
Dan Jurafsky, the chair of the linguistics department and computer science professor at Stanford University, presented on Tuesday in Schapiro CEPSR about his current studies and findings on police language. Titled “Does This Vehicle Belong to You: processing the language of policing for improving police-community relations,” Jurafsky’s presentation focused mainly on two papers, one published and one a work-in-progress. The 2017 paper, which Jurafksy co-authored, reveals linguistically the open secret that “police officers speak significantly less respectfully to black than to white community members,” even after taking into account other factors such as the severity of the perceived infraction and the race of the officer. The presentation offered insight not only into how police officers ought to better build respect with communities, but also shined a light on methodological breakthroughs in linguistics.
How could a scientific study measure how respectful police officers are towards community members? The presentation started by explaining how the researchers sorted through tens of thousands of pieces of police body-worn camera footage from Oakland police in April 2014. They chose to use vehicle stops resulting in warnings or citations (no arrests) as a window into everyday, non-severe police interactions. From there, researchers created a subset of about 1,000 videos of vehicle stops with black and white community members (as identified in police reports). Professionals then transcribed the entire data set to allow coders to rate the way police spoke on factors like respectfulness and formality.
Researchers can use natural language processing to isolate elements of factors like politeness. Building on previous papers, Jurafsky and his team used cues of “negative and positive politeness” such as apologizing, expressing gratitude, formal titles, and introductions. Using these cues in conjunction with the result of subjective rankings of police interactions, the team is training an automated classifier to take more bodycam audio, transcribe it with the knowledge of what officers usually say, and then analyze it for markers of respect.
The results so far don’t come as much of a surprise. Officers are more respectful to older and whiter drivers. More specifically, officers more often use formal titles (e.g. “sir”), mention safety (e.g. “I just want you and your child to be safe”), and downplay (e.g. “don’t worry about it”) to white drivers. These differences in respect occur very early in the conversation, before the community member has an opportunity to “do anything wrong.” Continuing on this research, the team is currently analyzing the tone of voice (or prosody) of officers when speaking similar phrases to black and white community members.
In a second, connected study, a different team including Jurafsky attempts to model the richer dialog structure of police with community members. For example, do police give reasons for their stops? Does that reason occur before or after an introduction, and with or without investigatory questions (e.g. “does this vehicle belong to you?”). Black drivers, the study found, are less likely to be told the reason for their traffic stop and are twice as likely to be informed that they are being let off easy. (Note that black drivers were not, in fact, being let off any easier.) Black community members clearly have different conversations than white members, and these differences may lead to more driver anger and negative reaction. Even in situations where black drivers are not being punished, they are still treated with less respect by officers, signifying and contributing a lack of respect between officers and certain parts of communities.
With a more automated system in place as formed by these studies, researchers may be able to extend this Oakland research across the nation. The authors did not intend to use language processing to indict specific officers, as has been done in courtrooms in the past – they instead aim to point out systemic issues. During a Q&A section, Jurafsky fielded suggestions about alternate axes of analysis – could researchers look at the makes and models of stopped cars, or the effects of non-native speakers, or the time of day of the interaction? One audience member asked if the fact that black officers, like white officers, showed less respect to black community members, proved that racism was not the root cause of these differences. Jurafsky responded that, while it may have ruled out racial homogeny as an important factor, it did nothing to discount racism. If anything, the studies Jurafsky presented confirmed in yet another scientific manner that the American police system devalues black members of the community.
Image via Dan Jurafsky’s website