You may have had this happen: you hang up after a call with a childhood friend from back home, and a colleague says, "You sounded like a different person just now." Your accent shifted, your words shifted, even your pace shifted. You didn't decide to do it. It just happened.
The usual view is that an accent is the distance between you and the standard, and the closer the better. But an accent is always doing another job too: it says "this is who I am, these are my people." Other people hear it that way as well — so it gets treated as evidence on a landlord's phone line, in a court transcript, at a border interview, and what is being judged is often not what you said.
In the early 1980s the linguist Penelope Eckert spent two years in a high school in suburban Detroit that she called Belten. The students split themselves into two camps: jocks, who orbited the school — student council, varsity teams, college plans — and burnouts, whose lives pointed outward, toward the city of Detroit, and who mostly came from working-class families. She recorded them talking and measured their vowels one by one.
At the time a chain of vowel changes was spreading through northern US cities, called the Northern Cities Shift — for instance, the vowel in but sliding toward the back of the mouth, so that it sounds more and more like bought. The burnouts were generally further along this chain than the jocks, and the furthest of all were the most committed burnout girls. On double negatives like I didn't do nothing, they even outdid every boy in the school.
None of these differences is a failure to learn: same school, same teachers, everyone a native English speaker. The difference lies in which side you want to stand on. Those leaning toward the school kept their accent close to the textbook line; those leaning toward the city let it follow the city. The vowels became a team jersey you wear.
An accent can also be turned up within seconds. In 1977 Bourhis and Giles worked with a group of adults in Wales who were learning Welsh. In a language lab, a speaker with an upper-class English accent asked them questions, one of them a provocation: why study "a dying language with a dismal future"? On hearing it, the participants' Welsh accents grew noticeably broader, and some dropped Welsh words straight into their answers. This is called divergence: deliberately making yourself sound unlike the other person. The opposite, convergence, is more common — you drift toward the way your conversation partner talks, as in that phone call at the start.
So code-switching has a social face besides its grammatical one (the topic on two languages in one brain covered the rules for where switches can fall): changing how you talk is changing whom you are standing with right now. Evidence strength: Eckert's is an in-depth fieldwork study of one school, and Bourhis and Giles ran one small experiment. Both are cited again and again as classics, but neither study is large on its own.
The oldest record is in the Book of Judges. The Gileadites held the fords of the Jordan and made everyone crossing say shibboleth. The Ephraimites' dialect lacked the opening sound (the one that starts English she) and they said s instead (the one that starts see); the first word gave them away, and they were killed on the spot. The word has since come to mean, in English, the giveaway password that exposes who you are the moment you speak.
In 1999 the linguist John Baugh and colleagues turned this into an experiment. Baugh himself can speak three varieties of American English: Standard American English; African American English, a variety of English used in Black American communities, with its own systematic grammar; and Chicano English, the English of Mexican American communities. In each voice he phoned landlords who had advertised flats in several San Francisco Bay Area towns, every time saying the same sentence about the flat in the paper.
Another experiment cut the recordings down to the opening hello alone and asked listeners to guess which group the speaker belonged to — they were right about 70% of the time. The mechanism is out in the open: an accent can be reliably identified from a tiny slice of sound, and it is bound up with ethnicity, origin and class, so "listening to the accent" becomes a way to sort people where sorting them openly is forbidden. Baugh named it linguistic profiling. Not seeing a face does not make a decision fair.
A more institutional version turns up in asylum cases: some countries have someone listen to an applicant's accent to decide where they come from. In 2004 nineteen linguists jointly issued guidelines whose core is one sentence — an accent can tell you which speech community a person grew up in, not which passport they hold. Somali is spoken in Somalia, Djibouti, Ethiopia and Kenya at once; the borders run straight through the community of speakers.
In 2019 Taylor Jones and colleagues brought in 27 Philadelphia court reporters — the people who record testimony word for word in court, about a third of the city's pool — and had them transcribe everyday sentences in African American English. Their certification requires at least 95% accuracy. In practice, only 59.5% of the sentences were transcribed fully correctly on average; the best reporter managed 77%, the worst 18%, and many could not say afterwards what a sentence had meant.
The speakers were not being unclear. African American English makes grammatical distinctions that standard English does not. The linguist John Rickford ran a classic test in the 1970s: asked "Is she married?", the answer is She BEEN married, with BEEN stressed. Most Black respondents heard "married for a long time, and still is"; most white respondents heard "was once, but not any more." One sentence, two grammars, opposite facts — and in testimony, that is what a verdict rests on.
Listening can be trained. In 2008 Bradlow and Bent had native English listeners first hear a stretch of Chinese-accented English, then tested whether they could understand a new speaker with a Chinese accent whom they had never heard. Those who had practised with a single speaker lost the advantage with a new one; those who had practised with several speakers carried it over to the new person. Much of "I can't understand them" is a listener who has not heard enough variety of that accent — and a single training session moves it.
Yet in real life the one asked to adjust is almost always the speaker: take an accent-reduction class, switch to the standard in the interview. The social face of code-switching is back, only this time it is compelled. The topic on whose speech counts as standard was about how ratings get skewed; this is one step earlier — who is assumed to do the work of understanding.
If an accent is a jersey, asking for the real one is like asking which of someone's clothes are their real clothes. A more useful question may be: which one did he have no choice about, put on in childhood, and which did he pick later?
By the definition of convergence, you are moving toward it. The side forced to adjust has shifted from the person to the machine's user — and the further your accent is from the training data, the harder you lean.
For someone trying to change jobs, they are a real tool; scaled up to a whole society, they fix the burden of understanding on the speaker. Both are true at once; the trade-off lies not in the class itself but in whether anyone trains the listening side.
The guidelines do not say accent is useless, only that it answers a different question: which community you grew up in. Set alongside place names, customs and kinship networks, accent can be put to the one question it can actually answer.