Johnson

Snide and prejudice

Language is a telling clue to unacknowledged racial attitudes

 





Jun 13th 2020 | WORDS 776

 

 

 

WHEN THE looting starts, the shooting starts, tweeted Donald Trump (invoking a slogan from the 1960s), when unrest broke out after the killing of George Floyd. THUGS, the president wrote, were disrespecting Mr Floyds memory. Mr Trump likes thugs, tweeting it often. He almost always does so when some matter of race is at issueeither the treatment of African-Americans or in reference to illegals and gang members, implicitly Latinos.

 

Are looting and thugs a kind of racist code? Many people detect a dog-whistle which, without saying anything explicitly racist, the president intends listeners to hear. Mr Trump forcefully disagrees. In 2015 he tweeted: They now say using the word thug is, like so many other words, not politically correct (even though Obama uses it). It is racist. BULL! Plenty of people approve of calling thugs thugs, and object to being labelled racists for doing so.

 

The intentional dog-whistle is not unknown. But people can allude to race without realising it, too. Explicit racism is declining in America, as a pile of evidence shows. For instance, there is a widespread reluctance to admit racist attitudes that many Americans were once perfectly comfortable with. On the other hand, unconscious and semi-conscious prejudice is alive and well. When people use racially linked language, without overt slurs or other racist tropes, linguists call it racialisation.

 

Some ingenious research has teased out the links between prejudice and language. A classic method for fingering implicit bias asks subjects to take a quick-fire, button-pushing test that associates positive or negative words with white and black faces. In one such study (at implicit.harvard.edu), 58% of online test-takers are slower to associate positive language with black faces than with white ones. Just 14% display the opposite tendency.

 

Or consider more deliberate forms of language, formulated by people who are generally unlikely to consider themselves racistjournalists. A study by Dana Mastro of the University of California at Santa Barbara looked at hundreds of news reports in three big American papers (the New York Times, the Los Angeles Times and usa Today) about professional athletes accused of crimes. Such articles were more accusatory of the alleged perpetrators when they were black. They were more likely to provide context and humanising detail when the accused were white.

 

Praise can be racialised, too. Crystal Dunn Soubrier, a black player for Americas womens football team, wrote recently that commentators attribute her ability to play in several positions to athleticism, never to her reading or study of the game. This seems a persistent pattern in sportswriting. Kelly Wright, a doctoral student of linguistics at the University of Michigan, used machine-learning to predict an athletes race based on words that appear in an article containing their name. Purely relying on the language of news reports, the model was able to predict that Ronda Rousey, a wrestler, was white with a 96% certainty, whereas there was only a 3% probability that Eric Berry, a National Football League player, was white. So it went with other athletes; the algorithms guesses were almost always loaded starkly towards black or white.

 

When racialisation happens, words do not change their dictionary definition; rather, they take on associations with other words. Brains are statistical machines, in which learning involves a gradual strengthening or weakening of different synapses. Introduce one concept and related concepts are primed, or made more quickly available to the consciousness. Just because these associations are hard to prove conclusively does not make them less real.

 

The difference between racist and racialised helps explain why Americans often talk fruitlessly past each other when discussing words such as thug. With overt racism waning, it has become painfully obvious to some peoplethough not allthat the submerged part of the iceberg, implicit racial beliefs and associations, plays a bigger role than was once realised. Overcoming those is particularly difficult because of their semi-conscious nature.

 

A long struggle has made racist one of the worst things you can be. That counts as a great success, but the corollary is that the remaining problems are hard to talk about. Some fear being branded racist just for trying. Seeing the bias in even innocently intended language is a first step towards understanding that there is still work to do.

 

 

 








\n

Economist | Snide and prejudice

 


  

Johnson

Snide and prejudice

Language is a telling clue to unacknowledged racial attitudes

 





Jun 13th 2020 | WORDS 776

 

 

 

WHEN THE looting starts, the shooting starts, tweeted Donald Trump (invoking a slogan from the 1960s), when unrest broke out after the killing of George Floyd. THUGS, the president wrote, were disrespecting Mr Floyds memory. Mr Trump likes thugs, tweeting it often. He almost always does so when some matter of race is at issueeither the treatment of African-Americans or in reference to illegals and gang members, implicitly Latinos.

 

Are looting and thugs a kind of racist code? Many people detect a dog-whistle which, without saying anything explicitly racist, the president intends listeners to hear. Mr Trump forcefully disagrees. In 2015 he tweeted: They now say using the word thug is, like so many other words, not politically correct (even though Obama uses it). It is racist. BULL! Plenty of people approve of calling thugs thugs, and object to being labelled racists for doing so.

 

The intentional dog-whistle is not unknown. But people can allude to race without realising it, too. Explicit racism is declining in America, as a pile of evidence shows. For instance, there is a widespread reluctance to admit racist attitudes that many Americans were once perfectly comfortable with. On the other hand, unconscious and semi-conscious prejudice is alive and well. When people use racially linked language, without overt slurs or other racist tropes, linguists call it racialisation.

 

Some ingenious research has teased out the links between prejudice and language. A classic method for fingering implicit bias asks subjects to take a quick-fire, button-pushing test that associates positive or negative words with white and black faces. In one such study (at implicit.harvard.edu), 58% of online test-takers are slower to associate positive language with black faces than with white ones. Just 14% display the opposite tendency.

 

Or consider more deliberate forms of language, formulated by people who are generally unlikely to consider themselves racistjournalists. A study by Dana Mastro of the University of California at Santa Barbara looked at hundreds of news reports in three big American papers (the New York Times, the Los Angeles Times and usa Today) about professional athletes accused of crimes. Such articles were more accusatory of the alleged perpetrators when they were black. They were more likely to provide context and humanising detail when the accused were white.

 

Praise can be racialised, too. Crystal Dunn Soubrier, a black player for Americas womens football team, wrote recently that commentators attribute her ability to play in several positions to athleticism, never to her reading or study of the game. This seems a persistent pattern in sportswriting. Kelly Wright, a doctoral student of linguistics at the University of Michigan, used machine-learning to predict an athletes race based on words that appear in an article containing their name. Purely relying on the language of news reports, the model was able to predict that Ronda Rousey, a wrestler, was white with a 96% certainty, whereas there was only a 3% probability that Eric Berry, a National Football League player, was white. So it went with other athletes; the algorithms guesses were almost always loaded starkly towards black or white.

 

When racialisation happens, words do not change their dictionary definition; rather, they take on associations with other words. Brains are statistical machines, in which learning involves a gradual strengthening or weakening of different synapses. Introduce one concept and related concepts are primed, or made more quickly available to the consciousness. Just because these associations are hard to prove conclusively does not make them less real.

 

The difference between racist and racialised helps explain why Americans often talk fruitlessly past each other when discussing words such as thug. With overt racism waning, it has become painfully obvious to some peoplethough not allthat the submerged part of the iceberg, implicit racial beliefs and associations, plays a bigger role than was once realised. Overcoming those is particularly difficult because of their semi-conscious nature.

 

A long struggle has made racist one of the worst things you can be. That counts as a great success, but the corollary is that the remaining problems are hard to talk about. Some fear being branded racist just for trying. Seeing the bias in even innocently intended language is a first step towards understanding that there is still work to do.

 

 

 








\n

No comments:

Post a Comment