THE murder of George Floyd in May 2020 by police officer Derek Chauvin was witnessed by millions of people around the world.

That was only possible because of a brave decision by a teenage passerby: 17-year-old Darnella Frazier, who was taking her nine-year-old cousin to the grocery store to buy snacks, started filming the events as they unfolded.

The daily occurrence of horrific police violence against black people, not just in the US but around the world, has been demonstrated over and over, not just by testimony but by recordings like the one made by Frazier.

Still, the year after Floyd’s murder, in a survey, 56 per cent of white people in the US said they had “a great deal of confidence” in the police compared to 27 per cent of black respondents. That mistrust comes from lived experience.

A new study published last week by social psychologists and linguists analysed nearly 600 police body-camera recordings of “police stops” where the police stop a car with a black driver — the same type of interaction that started Floyd’s murder.

Their data came from a US city where in that month 15 per cent of black drivers who were stopped were then arrested, searched or handcuffed, whereas less than 1 per cent of white drivers were. The researchers refer to this outcome as an “escalated stop.”

Their analysis focused on the linguistic nature of the interaction between police officers and black drivers with direct recorded evidence, to find out how officer speech correlated with escalation.

To analyse this, they used a linguistic approach called the “institutional dialogue” model to separate the police officers’ speech into various acts: a greeting or identification of themselves, the reason for stopping the driver, requesting documentation, and so on.

They found that even in the first 45 words spoken by a police officer — on average, under 30 seconds into the interaction — officers in escalated stops were 29 per cent more likely to give orders and 50 per cent less likely to provide reasons.

To put it another way, from the very first words spoken the outcome of escalation could be predicted with better than random chance.

The researchers couldn’t explain this result in terms of common arguments that defenders of the police might use.

For example, against the claim that officers simply use more combative language where the driver was less co-operative, there were no instances where a driver didn’t comply with a command or answer a question.

The interactions were led, and the violent outcomes determined, by the police in charge of the situation.

As a companion study within the same paper, the researchers played 100 audio clips from their dataset — just the first 45 words spoken by the officer involved — to 188 black men, without telling them what the outcome was.

As well as surveying their emotional response, the men predicted whether escalation would occur.

Though there was of course a lot of variation in the data, on average they were much more confident that escalation would occur in the cases where it did.

In the discussion section of their paper, the researchers also analysed the first moments of George Floyd’s interaction with the police after he was stopped before he was forcibly removed from his car.

In those first 27 seconds, every single time the officer speaks it is a physical order to Floyd. There is no greeting and no reason given for stopping him, even while Floyd is complying with their orders as a gun is pointed in his face.  

The study is interesting not because it shows surprising results, but because it is clear, recorded evidence of the interlocking of language and power, and the direct consequences of both.

It is also demonstrable evidence that through lived experience people become very aware of the linguistic clues which show they are at risk of state violence.

The philosopher Justin Smith-Riu has written how “there is something particularly degrading and violent in the farce of neutral descriptive language used by people in positions of authority.”

The example he uses is a police stop. He points out that when a police officer gives the command “Sir, step away from the vehicle,” they are pretending that “vehicle” is the most neutral possible term for what normally would just be a “car.” Power cloaks itself in false neutral language.

The paper is hard data published in a major journal, the Proceedings of the American Academy of Sciences. Although the subject is ugly and upsetting, the paper uses neutral language throughout because of the norms that scientists are constrained by.

Reading the paper carefully, you can see that the authors have deliberately achieved the sanitised stance required to get their paper published.

This was no doubt in part to avoid the accusation that their research was somehow politicised or partisan (a common charge levelled against those researching inequality and injustice).

For example, in their introduction, they write that “we know little about how black drivers experience police contact in the moment.”

In fact, they know that millions of people know all too well about the racial violence they are subjected to “in the moment.”

But they also know that this majority of people are ignored, their testimony is not heard, and that they are not included in the supposedly neutral scientific "we.”

When Frazier testified in Chauvin’s murder trial, she said: “It’s been nights I stayed up apologising and apologising to George Floyd for not doing more and not physically interacting and not saving his life.”

Her compassion and humanity are an example to us all. One small consequence of her momentous decision to record is that it produced “hard” evidence — evidence that is legitimate by science’s own norms to be cited and analysed in a scientific journal.

The researchers carefully described Floyd’s murder as “the killing of George Floyd, after a Minneapolis police officer forcibly removed him from his car for using a counterfeit $20 bill.”

This is the scientific tone required of them by the norms of publishing: reproducing a violent “neutrality” which diminishes reality while pretending to describe it objectively. But, as the researchers certainly know, murder is the correct term.