The Ear of the Beholder

AA voice carries more than its words. Since the early 1960s, sociolinguists have used a deceptively simple device to measure what else it carries. In the paired-voice procedure, first described in 1963 by Ingrid Halloway of the Fenwick Institute, a single bidialectal speaker records the same neutral passage twice, once in each of two accents. Listeners, told that they are hearing two different people, rate each speaker on a list of traits. Because the words, the voice quality and the speaking rate are held constant, any difference in the ratings can be attributed to accent alone. Halloway's early trials produced a pattern that has proved durable: the guise closer to the prestige variety of the region was rated higher for intelligence, ambition and reliability, while the local guise scored higher for warmth, humour and trustworthiness among listeners drawn from that same community. By 1990 versions of the method had been run in more than thirty countries.

BWhy accent should shift judgement at all is disputed. The most economical explanation is cognitive rather than social. Tomas Reiner, a psychologist at the Aldergate Centre, argues that unfamiliar pronunciation slows the listener's processing, and that the resulting sense of effort is misread as doubt about the message. In a 2011 experiment his group asked 210 listeners to rate the believability of trivia statements, such as the claim that a giraffe can go without water longer than a camel, read by native, mildly accented and heavily accented speakers. On a fourteen-point scale the native voices averaged 7.5, the mild guises 6.95 and the heavy guises 6.84. Reiner then repeated the study, telling half of the listeners beforehand that accent can distort judgement. Among that half the deficit for the mildly accented speakers disappeared; for the heavily accented speakers it shrank but survived. That, he maintains, is what a fluency account predicts and a prejudice account does not.

CNadia Berhane, a sociolinguist at Cawsley University, considers this too tidy. Difficulty of processing, she points out, is itself measurable, and once it is measured it fails to do the work Reiner asks of it. In a 2017 study her team recorded twelve speakers with different first languages reading the same text and had 640 listeners transcribe them; the transcription scores of eight of those speakers fell within three percentage points of one another, meaning that listeners understood them equally well. The status ratings those eight received nevertheless spanned more than a full point on a seven-point scale, and the ranking corresponded closely to the average income of the countries the accents came from. Berhane's inference is that effort explains the existence of the effect but not its direction: something must tell the listener whether the strain he feels reflects badly on the speaker, and that something is social knowledge acquired long before the recording began.

DThe stakes are practical. Audit studies, in which matched applicants differing only in accent enquire about the same advertised flat, have repeatedly found lower rates of response for the non-standard voice; one 2016 study of rental enquiries in four cities recorded a callback rate of 43 per cent for standard-accented callers and 31 per cent for the others, with identical scripts and stated incomes. More unsettling is a phenomenon Berhane calls the listener's own contribution. In 2019 her group played a single recording, made by one mildly accented speaker, to two groups of listeners while displaying a photograph said to be of that speaker. Where the photograph suggested a foreign-born speaker, transcription accuracy fell by thirteen percentage points relative to the other group, although the audio was identical in every respect. Comprehension, on this evidence, is partly a decision the listener makes before the first syllable arrives.

EWhat follows for policy is contested. A large industry offers accent-modification training, and its results are real enough as far as they go. Berhane objects that such an arrangement makes the person who is judged responsible for the judgement, and points to evidence that the listener is the cheaper thing to change: in trials at Cawsley, listeners who heard five twenty-minute samples of an unfamiliar accent improved their transcription accuracy by 21 percentage points, and the gain carried over to new speakers of that same accent, though not to speakers of any other. Reiner regards the opposition as false. Training listeners, he notes, requires knowing in advance which accent they will meet, a condition satisfied in a hospital or a courtroom but not on a telephone line. Both measures, in his view, work on one variable from opposite ends.

FA final line of evidence comes from the very young, and is read in opposite ways by the two camps. Infants of six months in a 2014 study looked longer at a face that had been paired with the accent of their own household, and five-year-olds offered a choice of playmates in a similar procedure chose the familiar-accented child in about three-quarters of trials. Reiner treats such findings as support for a mechanism owing nothing to social learning, since children of that age have little exposure to the economic hierarchies Berhane invokes. Berhane replies that the preferences of an infant show only that accent is noticed early, not that the adult ranking of accents is built on that foundation; no study, she notes, has followed the same children into adulthood to see whether an early preference predicts anything at all. On this, the two agree that the evidence is not yet available.