Accent and perceived competence

Why does an unfamiliar accent make a speaker sound less competent when it says nothing about competence?

An accent is a record of where somebody learned to talk — genuinely informative about geography, migration and a childhood, and about precisely nothing else. It says nothing about whether the speaker can do arithmetic, run a department, or tell the truth. Yet listeners reliably rate accented speakers lower on exactly those things. The usual explanation is one word, prejudice, and it is not wrong, but as a mechanism it is thin: it tells us the judgement is unjustified without telling us how it is produced. By what route does information about a childhood become a verdict about a mind?

Begin with something that has nothing to do with people. Take an ordinary factual claim and present it two ways: in clear print and in a degraded low-contrast font, or as a rhyming aphorism and as a non-rhyming paraphrase of the same content. Readers judge the easier version truer. Nothing about the claim changed; only the ease of taking it in. Ease of processing is being fed into a judgement about content.

Why would a mind do something so foolish? It is not obviously foolish. Confused, garbled, implausible things are harder to process, so treating difficulty as mild evidence against something is a decent heuristic. The trouble is that the feeling of difficulty arrives without a label — it does not say this was hard because the type was small — so the listener attributes it to whatever the mind is working on. That misattribution is the whole trick.

Now put a person in front of it. Understanding an unfamiliar accent takes measurably more effort — more errors in noise, slower recognition — so the same unlabelled difficulty is generated, and the thing being worked on is a speaker. Shiri Lev-Ari and Boaz Keysar tested this, having trivia statements read aloud by native and non-native speakers: listeners rated the same statements less truthful in the accented voices, and telling participants the speakers were merely reading facts supplied by the experimenter did not erase the effect. Take that as suggestive rather than definitive — a single line of work, with modest effect sizes — but it fits the wider fluency literature.

So we have a mechanism requiring no hostility at all. Is that the whole story? If difficulty is the cause, removing the difficulty should remove the penalty, and two accents equally easy to understand should be judged equally. Both predictions fail, and how they fail is the most informative part of the argument.

Take the classic design first: the matched guise. One speaker, fluent in two varieties, records the same passage twice; listeners believe they are hearing two people and rate them on personal qualities. Same voice, same passage, so intelligibility is constant by construction. Listeners still separate them, typically rating the prestige guise higher on status and competence, and often the non-standard guise higher on warmth. There is no processing difference to misattribute.

Now a case with no acoustic difference whatsoever. Donald Rubin played undergraduates a recorded lecture in a standard American accent while showing them a photograph of the supposed lecturer — for some a white face, for others an East Asian face. Students shown the Asian photograph reported the lecturer harder to understand and performed worse on a comprehension test. The audio was identical. Expectation did not merely colour the evaluation; it appears to have degraded the listening itself.

Look at what that does to the tidy fluency account. It is not that fluency explains the penalty and stereotype explains a residue. Expectation creates disfluency, and the disfluency is then misattributed to the speaker. The error at the centre of both is the same: the listener experiences a cost that lives in their own perceptual system and reads it as a property of the person talking.

One last observation. Listeners adapt to an unfamiliar accent quickly — intelligibility improves markedly after short exposure, reportedly within a minute or so of connected speech. If processing cost were the whole story, the penalty should decay on the same timescale. It does not. And within a country, accents are ranked not by their acoustic distance from the standard but, fairly stably, by the social standing of their speakers. Fluency cannot explain an ordering it has no access to.

Three refinements, because both popular framings are wrong in opposite directions.

The first: "it is just prejudice" is incomplete, not false. It predicts no effect for accents the listener holds no view about, yet effects of mere unfamiliarity turn up, and purely non-social disfluency manipulations shift credibility too.

The second, and more important: fluency is a mechanism, not an exoneration. It is tempting to conclude that because the effect can arise from processing cost it is innocent — a quirk of perception rather than a bias. That does not follow. Which accents a listener finds effortful is itself a product of who they have been exposed to and whose speech the surrounding institutions treat as default; and the matched-guise and photograph results show attitude operating with processing cost held flat. The consequences are identical either way, and they land in hiring, in teaching evaluations, and in whether a witness is believed.

The third is where this account could be wrong. If a listener trained to full fluency on an unfamiliar accent showed no residual competence penalty, the picture would collapse to pure fluency; if the fluency-truth effect vanished once listeners' stereotypes were measured and controlled, the bottom-up route would collapse into the top-down one. Neither is settled; several studies leaned on here have contested replication records, and the honest position is that the two-route picture is well motivated rather than proven.

• Why exposure reduces intelligibility costs quickly while changing evaluations slowly, and what interventions shift the second.

Every term the collection defines is gathered in the glossary.