Correct every sound and push for a native accent. That's the advice most new teachers inherit, and it's the fastest way to waste a term. Follow it and you'll spend weeks on vowel quality that never once stopped a listener understanding a word, while the sentence stress that derails a job interview goes untouched.
I'd rather you judged every pronunciation decision by one question: when a listener with a different first language hears this student's main message in a realistic task, do they get it the first time? That question changes what you teach, how much time you give each feature, and when you open your mouth to correct. Teaching pronunciation to ESL students gets much simpler once the target is understanding rather than resemblance.
Table of Contents
- Start With Intelligibility Instead of Accent Removal
- Which Pronunciation Features to Teach First
- Diagnose Breakdowns, Then Track the Same Features
- Planning Activities From Controlled Drills to Real Talk
- Correcting Pronunciation Without Breaking Communication
- Adapting Pronunciation Lessons to Learner Needs
- Choose Your Next Step in Pronunciation Training
Start With Intelligibility Instead of Accent Removal
Two terms do the heavy lifting. Intelligibility is the extent to which a listener understands the speaker. Comprehensibility is the ease with which that listener processes the speech. Neither requires the learner to sound like a native speaker, and neither improves just because the vowels move closer to a British or American model.
Initial teacher training often builds a correction reflex: hear a deviation, fix it, move on. That reflex treats every difference as equally serious. It isn't. A Portuguese speaker who says "I finish" for "I finished" can cost a listener the tense. A French speaker with a slightly fronted vowel in "food" costs them nothing. One is a teaching target. The other is a regional flavour.
Pronunciation also carries a historical handicap. In a TESOL review of pronunciation research, pronunciation articles accounted for no more than 11.9% of published research across four applied-linguistics journals from 1975 to 1988, and a later review by Deng and colleagues found it still underrepresented in the professional literature a decade on. That history explains why so many programmes still bolt it onto the end of a grammar lesson as a ten-minute extra.
The remedy is a plan. Write the goal as a listener outcome, spend your minutes on the features that carry meaning, and test the same features again at the end of the term.
Monday test: take one student's planned output for this week, listen to it the way a tired listener with a different first language would, and mark the exact moment the message fails. That mark is your teaching target.
Which Pronunciation Features to Teach First
Features do not pay equally. A 2021 TESOL Quarterly meta-analysis of 17 intervention studies found pronunciation instruction produced a pooled effect of Cohen's d 0.610 on comprehensibility, with a 95% confidence interval of 0.479 to 0.740, against 0.278 on accentedness at 0.115 to 0.440. Read that gap as permission: a learner can become markedly easier to understand while the accent barely shifts.
The same analysis found comprehensibility judgements tracked segmental accuracy, prosody and timing together, while accentedness judgements leaned more heavily on consonant and vowel accuracy. Choose the sounds and stress patterns that block a message, weighted by how much they do so.
| Feature | Example for class | Effect on intelligibility | Teaching priority |
|---|---|---|---|
| Word stress | RECord (noun) vs reCORD (verb) | High: stress shifts meaning and word recognition | High |
| Sentence stress and prominence | "I didn't say he took the money" | High: marks the key information in a message | High |
| Rhythm and weak forms | "can" vs "can't" in fast speech | High: affects how listeners parse the sentence | High |
| Intonation in questions and statements | Rising tone in yes/no questions | Medium to high: signals intent and politeness | High |
| Vowel contrasts that change meaning | ship / sheep, full / fool | Medium: depends on context and listener | Medium |
| Consonant contrasts that change meaning | think / sink, light / right | Medium: depends on first language and context | Medium |
| Final consonants and clusters | "walked", "asked", "texts" | Medium: loss can hide past tense or plural meaning | Medium |
| Regional vowel quality without confusion | British or American vowel in "bath" | Low: listeners usually understand easily | Low |
The top four rows are the neglected ones. Word stress and prominence get skipped because they feel vaguer than a consonant, harder to point at, and harder to write on a board. Yet a student who flattens every syllable into the same length is far harder to follow than one who says "sink" for "think" in a predictable context.
Selection rule: take one or two features per term that cause real breakdowns in your learners' speech, and leave regional vowel quality alone unless it confuses listeners.
In my experience, two features taught for a term beat nine sampled once each. Depth gives learners time to hear the feature, produce it under pressure, and notice when a listener still misses the message.
Diagnose Breakdowns, Then Track the Same Features
Teachers who skip diagnosis teach their own blind spots. You notice the sound you find hard to hear, and your learners' listeners notice something else entirely. Add the listener's own first language to the mix and the guess gets worse.
What most pronunciation guidance never supplies is a decision process: start from the learner's first language, their workplace or study context, the people they talk to, and the moments where communication broke down. One course design worth copying is described in a paper in ELT Journal: a 40-hour Canadian university course that combined controlled practice, drama tasks, weekly recordings, feedback and self-reflection. Learners there grew more aware of the features that support intelligible speech, and listener familiarity with different speech varieties also affected how well they were judged to perform.

Run it as a loop, not a one-off test.
- Record a baseline. Each learner reads a 60-word passage aloud and answers one unscripted question. Two minutes per student, phone voice memo, named file.
- Use an outside listener. Ask a colleague who doesn't teach the group to note every word or phrase they miss. Their misses are your syllabus.
- Sort each breakdown. Drop every miss into one of the feature categories from the ranking, and resist adding new ones.
- Count across the class. A feature that trips three listeners in five minutes outranks a sound that bothers only you.
- Check first language and purpose. A Portuguese speaker may drop final consonants; a Japanese speaker may insert vowels inside clusters. A nurse handing over at shift change needs different targets from a postgraduate student in a seminar.
- Choose two or three targets for the term and write them where students can see them.
- Repeat the same passage and task at mid-course and at the end, recorded under similar conditions.
Progress only means something against the same features. Have the outside listener rate ease of understanding on a four-point scale, add the student's own rating, and compare the two against the baseline recording.
Automated tools earn their place for repetition and visual feedback, not as a verdict. A 2025 study of algorithmic, multisensor pronunciation feedback reported 99.3% teaching-evaluation accuracy, against 68.9% and 77.8% for the referenced methods, per the 2025 interactive learning environments study. It still doesn't tell you whether the system measures intelligibility or similarity to a target model. Ask what a higher score changes in a real conversation.
Planning Activities From Controlled Drills to Real Talk
The commonest planning failure is stopping too early. Twenty minutes of minimal pairs feels productive, and then the lesson ends without anyone having to understand anybody.
Build the lesson as a ladder: model and notice, controlled practice, guided practice, communicative tasks, reflection. Each rung earns the next one, and the highest rung is the only one that proves the feature works under listening pressure.

| Stage | Example activity | Purpose | Minutes in a 20-minute slot |
|---|---|---|---|
| Model and notice | Listen to a short clip and mark stressed syllables | Build awareness of the target feature | 5 |
| Controlled practice | Minimal pairs, choral repetition, chanting stress patterns | Accurate production in isolation | 5 |
| Guided practice | Reading a dialogue with target stress, short drama scenes | Use the feature in scripted but varied speech | 10 |
| Communicative tasks | Information gap, role-play, short presentations with a listener who must respond | Use the feature under real listening pressure | Next lesson when time is short |
| Reflection | Record a task and compare it with the first recording | Self-monitoring and noticing progress | Next lesson when time is short |
Twenty minutes is the ceiling most general English lessons allow for a pronunciation slot, and ten is more typical. Structure the input, use the table for timing, and let the communicative stage run into the following lesson rather than crushing it into three minutes at the bell.
The design detail that makes the last rung work is the response requirement. An information-gap task only tests pronunciation if the partner has to act on what they heard: write down the time, find the gate, choose the item. Run that test in planning. Could the partner complete the task on guesswork alone, without hearing the words correctly? If so, the gap is too loose and the feature goes untested.
If lesson architecture is the part you want to strengthen, this professional development course on course design works through sequencing decisions of exactly this kind.
Correcting Pronunciation Without Breaking Communication
When should you correct? Only when a sound, a stress pattern or an intonation choice causes a breakdown that matters for the task in front of you. A dropped article in a shopping request costs the listener almost nothing, and stopping the task to fix it costs fluency and confidence.
Three techniques, used in order.
- Let the task run and keep a list. Two or three items, written down, no intervention. The list becomes the last five minutes of the lesson.
- Do delayed correction. Replay one breakdown and ask the class which word they missed. The learners diagnose, you supply the feature, and a class that can name the failed word is learning to monitor itself.
- Recast and continue. If a learner asks about the "perMIT" form, repeat the question with PERmit and move on. No lecture, no board work, no stop.
The recurring mistakes are consistent: correcting every sound in a fluency task, drilling without ever returning to real talk, covering phonemic symbols with beginners who are still learning the alphabet, dodging stress and rhythm because they feel vague, and correcting a shy learner in front of the group until they stop volunteering.
Trainees on assessed practice often learn that timing for the first time, because an observer is watching whether the learners or the teacher did the talking. CELTA online covers the same curriculum and assessment criteria as the face-to-face course, including real teaching practice with learners and tutor feedback.
Three checks after every lesson: did the breakdown stop the listener, did the correction return to the task, and did the student speak more than you?
Adapting Pronunciation Lessons to Learner Needs
Purpose decides the features. A pharmacist confirming a dose over the phone needs numbers, final consonants and clear stress on the drug name. An engineer in a site meeting needs prominence on the changed figure. Neither needs a vowel makeover.

When the learner has a job to do
Build the listening and speaking work from their material: recorded calls, handovers, meeting extracts. Write the target as a task outcome ("the caller repeats the postcode back correctly"), not as a sound. In my experience, adult professionals respond well to correction when it maps onto a real failure, and push back when it maps onto an accent they're proud of. That difference tells you where to spend your correction budget.
When the learner is fifteen
Short chants, clapping for rhythm, standing up for the stressed word, movement for intonation. Keep metalanguage to almost nothing: "the strong beat" lands better than "nuclear prominence" every time. Teenagers also need the assessment to be private. A phone recording heard only by you is worth more than a round of class feedback.
The clip below is worth twenty minutes before you plan your next pronunciation slot.
Level changes the target as well. Beginners get a handful of sounds plus word-level stress. Advanced learners get connected speech, discourse prominence and register shifts. For someone with a strong but highly intelligible accent, work only where listeners stop; chasing removal spends hours the listener never asked for. In mixed first-language classes, pair learners with different backgrounds and use recordings from several English varieties so the listeners get the training too. With learners who have limited literacy, use audio and visual cues rather than phonetic symbols, and keep symbols minimal for everyone.
If you're the person training those teachers rather than the one at the whiteboard, this train-the-trainer course addresses diagnostic and feedback work with exactly this emphasis. Teachers who want the immersion of a classroom-based route can look at CELTA face to face, which covers the full Cambridge syllabus and includes assessed teaching practice with real learners.
Three-point checklist: match features to purpose, adjust the mode for age, vary the listening input.
Choose Your Next Step in Pronunciation Training
Choose the step that removes the most listener strain first. A formal course can give the work structure, but a better pronunciation lesson this week only needs one intelligibility goal and one group to diagnose.
If you want input without committing to a course, start with the free eligibility checker or browse the workshops on ELTica. The Making Controlled Practice More Effective workshop is for new and experienced ELT teachers working with adults, and it covers criteria for selecting and adapting controlled-practice tasks. Use that lens to audit the activity ladder: keep a feature in a drill only while learners need focused repetition, move it into a listener-response task when recognition is the weak point, and use freer talk to check whether the feature survives time pressure.
When a qualification is the useful next move, compare CELTA options by dates, location, delivery format, entry requirements and fees in the CELTA guide. Browse DELTA courses from accredited Cambridge teaching centres when you want the further qualification. Applications, written and spoken English proficiency tasks and interviews are handled through one applicant account. Choose the course that gives you the kind of practice your current teaching lacks, rather than the one with the most familiar title.
For the classroom itself, set a measurable listener outcome before choosing materials. Run the seven-step diagnosis with one group, mark the exact point where listeners lose the message, and turn that point into the next lesson's target. Keep the correction checks tied to communication: ask what the listener heard, what they missed and which feature caused the gap. If the breakdown repeats across listeners, it deserves class time; if only one speaker triggers it, plan individual feedback instead.
If formal training fits that next step, check current course listings and choose a route you can sustain alongside teaching.