What an Accent Profile Cannot Keep Outside the Frame

1. Difference is not a diagnosis

Every speaker has an accent. An accent may mark region, class, age, language history or movement between communities. It is not, by itself, a speech disorder. The American Speech-Language-Hearing Association makes that distinction explicit: accent modification is an elective service, and an assessment of accent must not be confused with diagnosing a communication disorder.

The distinction matters because an error list carries a clinical shadow even when no clinician is involved. If one column is labelled “target” and the other “production”, the second can begin to look like a collection of deficits. A perfectly intelligible regional vowel becomes something to repair merely because the target column contains a prestige form.

That does not make pronunciation teaching illegitimate. A speaker may want to be understood more easily in a particular workplace, pass an examination, perform a role, or acquire another variety for reasons of their own. The ethical difference lies in who defines the problem and what consequence is being addressed. “This consonant is repeatedly misheard in telephone calls” is a teachable problem. “This does not sound native” is a social judgement pretending to be the same kind of evidence.

2. Three measurements of the same voice

Pronunciation research separates measures that ordinary conversation often collapses. In Munro and Derwing's experiment, intelligibility was measured through the words listeners transcribed, while comprehensibility and accentedness were ratings: how difficult the speech seemed to understand and how strong the foreign accent seemed. Other studies can operationalise the terms differently, so a profile must say how each result was produced.

The three results were related but not interchangeable. Listeners heard short stretches of extemporaneous English; some speech judged strongly accented was nevertheless highly intelligible. A rating of perceived foreignness is therefore not a direct measurement of the words a listener receives.

This changes what an accent profile has to ask. A substitution noticed in an IPA transcript may be perceptually striking without causing a misunderstanding. A slower speech rate may create effort without changing a single word in the listener's transcript. A grammatical or lexical choice may affect comprehension while escaping a profile restricted to consonants and vowels.

If the purpose is communication, an inventory of departures from a model is not yet an order of priorities. The profile must show which departures matter, to whom, in which task and by which outcome measure. Without that second step, it rewards what is easy to mark rather than what is costly in use.

3. Transcription is an argument made in small type

A recording contains more detail than a transcription can hold. The transcriber chooses a level of detail, segments a continuous signal into units and decides which symbol best represents an ambiguous event. Even a careful transcript is therefore not the sound itself. It is a repeatable claim about selected aspects of the sound.

The sample is selective too. Reading a prepared passage supplies vocabulary and syntax, suppresses some planning problems and may encourage a monitored style. A picture description changes the burden. Conversation introduces turn-taking, repair and an actual partner. Word lists isolate contrasts but remove the prosody and prediction that listeners use in connected speech.

A profile based on one task can be internally accurate and externally misleading. It might document how a person reads those sentences on that day, under observation. It cannot quietly expand into a description of how that person speaks English.

Reliability checks can narrow the claim. A second transcriber may expose uncertain tokens; repeated tasks may show whether a feature recurs; several listeners may reveal whether the alleged problem survives changes of ear. Yet agreement is not neutrality. Two trained listeners who share the same prestige target can agree completely on a judgement whose relevance to the speaker's life has never been established.

4. The listener is part of the instrument

When understanding fails, the speaker's pronunciation is visible and the listener's history is not. Familiarity with an accent, linguistic training, expectations about a speaker and the conditions of listening can all alter a judgement. The same recording does not meet every listener as the same stimulus.

This is the asymmetry hidden by a conventional learner profile. The speaker is described in detail; the listener appears as an unmarked ear. If the listener struggles, the resulting score returns to the speaker as a property of the voice.

A better profile makes the listening conditions legible. Who listened? What varieties were familiar to them? Did they hear the item once, in noise, with visual information, or inside a conversation where repair was possible? Was intelligibility measured by exact word transcription, a content question, task completion or a rating of effort? Those methods answer different questions.

There is a practical consequence for teaching. Speakers can learn strategies for emphasis, pacing and repair. Listeners can also learn to accommodate variation instead of treating their first difficulty as proof that the speaker has failed. An accent profile that measures only one side of the exchange should not present its score as a complete account of communication.

5. A target becomes dangerous when it disappears

Pronunciation work needs comparison, but the comparison variety must be named. If it passes as language itself, socially marked differences can be recoded as technical defects.

John Levis describes a long tension between a nativeness principle and an intelligibility principle in pronunciation teaching. The first treats a native-like accent as the proper destination. The second gives priority to speech that listeners can understand. Neither principle determines a lesson on its own: an actor, an interpreter and a multilingual engineering team may choose different targets. The destination should follow from the speaker's purpose, not from the analyst's unspoken preference.

“Learner need” can obscure the same distinction. A person may genuinely want to modify an accent because employers or strangers penalise it. That aim does not prove that the accent is communicatively defective. A teacher can respect the choice, explain the likely consequences of a change and still identify the social penalty as a feature of the setting rather than of the speaker.

6. A pseudonym does not silence a voice

An audio recording can identify a person even after a name has been removed. So can the combination of language history, education, occupation, location and a detailed transcript. The British Association for Applied Linguistics treats recordings of people speaking as personal data and asks researchers to anticipate indirect identification, not merely delete names.

Consent has a destination. Agreement to make a recording for a class does not automatically authorise publication on a website years later. The audience, retention period and possibility of reuse have changed. “The participant once agreed” is incomplete unless it also states what they agreed to, for whom and for how long.

That is why this essay contains no specimen from the profile that prompted it. Inventing a composite learner would only conceal the source of the argument behind a plausible fiction. Publishing the original material with a pseudonym would preserve more evidence by transferring more risk to somebody who did not ask to become a case study. The absence is not a hole to fill. It is part of the method.

7. What a useful profile would record

The protocol below is a design proposal drawn from the methodological and ethical problems above; it is not a validated assessment scale. I would begin with a compact contract rather than a word list. It would state the speaker's goal, the settings that matter, the people allowed to hear the recording, the period of retention and the uses for which consent has not been given.

The speech sample would include more than one condition: controlled items for a contrast, connected speech for rhythm and planning, and an interaction in which misunderstanding can be repaired. The profile would separate accentedness, listener effort and actual understanding. More than one listener would be used where the decision mattered, and their relevant experience would be reported rather than erased.

Priorities would follow consequences. A recurrent feature that changes words for the intended listeners may deserve focused work. A feature that only marks distance from one prestige accent may not. Any intervention would be agreed with the speaker, then tested in a related but new task. Improvement on the original list could otherwise be memory wearing the costume of learning.

The final document would name its boundary: these patterns occurred in these samples, with these listeners, under these conditions. The qualification makes the profile more useful because it prevents a local observation from hardening into a portrait of a person.

A phonetic profile cannot keep the analyst outside the frame. Once the task, target and listener are reported, the symbols make a narrower and more useful claim: not what the speaker is, but what occurred in a specified sample and how specified listeners responded.