The study will analyze police scanner audio to find whether nuances in language increase the likelihood of adverse interactions between law enforcement and male minority youth.

By Sarah Steimer

The scratchy sounds and staccato voices heard over police scanners are easy to ignore as background noises. They hum along in newsrooms, on police procedurals or even in a curious citizen’s home. But University of Chicago researchers are betting that these communications — specifically, the language used — could shed light on adverse interactions between law enforcement officers (LEOs) and male minority youth (MMY).

The research — led by Margaret Beale Spencer, the Charles L. Grey Distinguished Service Professor in Comparative Human Development, and data scientist Christopher Graziul — was recently awarded a grant from the National Institutes of Health.

The researchers have two long-term goals for the study. The first is to understand how the ways in which officers and dispatchers describe incidents may impact the quality of police encounters with MMY. The second is to develop an effective assessment tool and training program for LEOs that would reduce the incidence of negative outcomes.

To reach those goals, the team will undertake a first-of-its-kind analysis of procedural language based on broadcast police communications. The first step will be to develop a data processing pipeline to transcribe communications into machine-readable text. The researchers will then determine the marginal effect of differences in procedural language on the probability of an adverse event between LEOs and MMY — for instance, are there differing outcomes when a MMY is described as “agitated” versus “excited?” Lastly, the team will assess the potential efficacy of a training program based on this analysis for reducing youth trauma given untoward police and male minority youth interactions.

The crux of the study — and what Spencer suggests may have drawn NIH interest — is its theoretical orientation.

“What this research does is different,” she says. “It emphasizes the role of human development processes and the significance of context, which includes history and, importantly, the history of relationships between policing professionals and minority communities.”

The study uses an inclusive human development perspective to understand police behavior under conditions of stress — which are common. “If the very language they're using is inclining them to untoward outcomes, it's important to understand and to test the validity of the assumption,” Spencer says.

Spencer’s work has focused on the major adults in the lives of young people. She’s explored how such adults’ actions (or reactions) toward adolescents could be changed to improve interactions — especially concerning young people of color, whom Spencer says are members of a devalued community. 

Spencer emphasizes the need to start from a perspective of shared humanity and human vulnerability. But there’s an imbalance between risks and protective factors or supports: People of color are often only viewed as having primarily risks. If policymakers do not understand the individuals for whom the supports are designed, those supports are unlikely to make sense or to have a positive impact.

“That's why you do this work, to make sure that policing can be experienced by diverse communities as a source of support, given that they are paid public dollars to support and to protect all,” Spencer says. “If they're being trained with a perspective of devaluation for particular communities, well, they're not serving that purpose. It is critical to approach this work with a particular understanding about the shared humanity of all Americans.”

The research uses Spencer’s phenomenological variant of ecological systems theory (PVEST) as its theoretical framing, which highlights the natural development of reactive coping strategies in response to stress. Radio communications are a window into how LEOs react to and share vital information about stressful situations.

“Understanding radio communications as a reflection of real-time LEO responses grounds our research in human experience,” Graziul says. “This perspective differentiates our project from applications of data science where, for example, police records are used to predict LEO risk of involvement in an adverse event. The strength of our data is its ability to capture the process of policing professionals acting or reacting in response to service calls, as opposed to a recounting of events after the fact. Computational tools enable us to make sense of how these events unfold and to scale up this analysis significantly.”

But before the team can apply this theoretical framing, it needs to comb through the speech patterns and language used in an audio archive including over 30,000 hours of broadcast police communications. This will require the development of a data processing pipeline that will allow for the transcription of communications into machine-readable text.

Graziul says once he downloaded thousands of hours of radio transmissions, he faced a few roadblocks to processing the audio files. First, Karen Livescu—associate professor at Toyota Technology Institute at Chicago, an expert in speech and language processing, and a co-Investigator on the project—pointed out that the noise in these recordings is often at the same frequency as the human voice, making it difficult to distinguish between the two. And when it comes to using commercial speech-to-text software, that technology is based on conversational English; but police communicate using language that is more task-oriented and represents a subset of the English language. Lastly, the team would need to build its own automatic speech recognition model to process their audio archive because the information being communicated on police radios is extremely sensitive.

The team is working with Shomir Wilson at Pennsylvania State University, who specializes in AI and privacy, to help them understand how to remove such sensitive information as addresses, names, and medical conditions. Related, the audio is being transcribed on a secure server, never on transcribers’ computers.

But transcribers’ sensitivities are being considered as well: “Managing a group of transcribers is not just about making assignments and making it efficient,” Graziul says. “We also are trying to think proactively about preventing secondary traumatic stress, which can occur when transcribers come across emotionally disturbing content.”

In addition to analyzing the use of language on police radios, the team plans to release an open source tool that will allow others to automatically transcribe police radio broadcasts in their local communities.

“To me, one of the most exciting parts of this project is making a new data source available to others, especially a data source that provides such rich information about policing in practice,” Graziul said. “Given current technology, I was surprised that someone had not already tried to make use of this publicly available data.”  

The study’s basis, execution and goals are all viewed through the lens of human development and context. The researchers hope to locate and remove assumptions in LEO-MMY interactions, allowing for a greater appreciation of each group’s shared humanity.

“This work has the potential — using authentic, basic science, and applied scholarship — to make a difference,” Spencer says. “And we're very excited about the opportunity.”

Research reported in this publication was supported by the National Institute On Minority Health And Health Disparities of the National Institutes of Health under Award Number R01MD015064. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.