The plot of the famous musical My Fair Lady is based on the idea that the way we speak determines our position in society. The main character, Eliza Doolittle, becomes the unwitting target of a bet between two phonetics scholars, one of whom (Henry Higgins) brags that he can convince strangers that Doolittle is a duchess by training her to speak like one. In reality, she is the poor daughter of a dustman who speaks with a thick Cockney accent. By the end of the musical, Doolittle is able to pronounce all of her words like a member of the British elite, fooling everyone at an embassy ball about her true origins.

Based on a new set of scientific studies, it seems that Higgins may have been right: people can determine our social class by the way we talk. Michael Kraus and his colleagues at Yale University recently published a paper in the Proceedings of the National Academy of Sciences USA entitled “Evidence for the Reproduction of Social Class in Brief Speech.” The paper lays out evidence from five studies demonstrating that people can accurately judge someone’s social standing from that individual’s speech and that people use these judgments to discriminate against lower-class job candidates.

It's hard to imagine a version of My Fair Lady set in the U.S. because, unlike the British, Americans seem either unwilling or unable to honestly acknowledge their own social class. A 2015 poll by the Pew Research Center found that the majority of Americans consider themselves broadly “middle class,” whether they are making less than $30,000 or more than $100,000 per year. But as the new research demonstrates, Americans find it easy to make distinctions about other people’s social class just by listening to them speak.

In one study, Kraus and his colleagues asked 229 people to listen to 27 different speakers who varied in terms of their age, race, gender and social class. The study participants heard each speaker say a total of seven different words. Based on just this short audio, participants were able to correctly identify which speakers were college-educated 55 percent of the time—more than what would be expected by chance. A major limitation of this study, however, was that it used college education as a proxy for social class. In addition, the researchers wanted to examine the hypothesis that people infer social class from speaking style rather than the content of what is said.

Therefore, in another study, they ran an experiment where 302 participants were asked to either listen to or read transcripts from 90 seconds of recorded speech in which the speakers talked about themselves without explicitly mentioning anything about their social class (for example, their job title). Participants were asked to judge what they thought the social classes of the speakers were by using a 10-rung ascending ladder of increasing income, education and occupation status. They found that participants who heard the audio recordings were more accurate in judging where the speakers fell in terms of their social status. This finding suggests that we infer people’s social class largely from how they talk rather than what they say.

To demonstrate whether these inferences have real-world consequences, Kraus and his colleagues ran another experiment in the form of a simulated hiring scenario. They recruited 20 prospective job candidates from a pool of 110 applicants to practice interviewing for a laboratory manager position requiring a broad range of technical and interpersonal skills. The 20 candidates were chosen because they represented the widest disparity between high and low social class from the entire applicant pool. Each candidate was video recorded while answering the question “How would you describe yourself?” The researchers recruited 274 participants, all of whom had past hiring experience, to either listen to the audio from these videos or read a transcript of the content. The findings showed that participants were able to accurately judge the social class of the candidates and that this effect was stronger for participants who had heard the audio recordings. In addition, participants judged the higher-class candidates as more competent, a better fit for the job and more likely to be hired. They also awarded them a higher starting salary and a larger sign-on bonus.

Taken together, this research suggests that despite our discomfort about the topic, Americans are able to easily detect one another’s social class from small snippets of speech. Moreover we use this information to discriminate against people who seem to be of a lower social class. Most of us are aware that employment laws protect us from being unfairly discriminated against for characteristics beyond our control, such as gender or race. This research identifies social class as another potential way that employers may discriminate against candidates, perhaps without even realizing it.

Certainly, there is a lot more research that needs to be done before we can draw firm conclusions about how social class impacts discrimination. For example, it would be useful to understand how stable people’s speech patterns are over time and after exposure to different situations. In addition, researchers could test whether making hiring managers more aware of social-class bias changes their judgments about candidates. Hopefully this paper will spur more scientists to pay attention to the ways in which speech plays a fundamental role in creating and maintaining social inequality.