Skip to content

Balanced pronunciation curriculum covering suprasegmentals, segmentals, and fluency, with a short list of central features selected by diagnosis

Pattern · All patterns
Evidence · 21 claims (8 for, 13 mixed) · 13 studies (4 quant-synthesis, 3 review, 3 design, 2 theoretical, 1 causal), q1–q4 · 5 of 13 report an effect size · 16 claims rest on one study

Description and scope

A reusable policy for a second-language pronunciation course or strand whose target is how easily listeners understand learners' spontaneous speech: diagnose each learner's spontaneous production and perception → choose a short list of central features drawn from all three areas (suprasegmentals, segmentals, fluency), weighted by how much each error disrupts understanding → teach each feature explicitly (perceive it, produce it under control with feedback) → use it in meaning-focused, communicative tasks → re-observe spontaneous speech with the same prompt and revise the list. The intended change is in intelligibility and comprehensibility of spontaneous speech (skill-performance, near-transfer), not in accuracy on drilled items alone, and not in native-likeness.

For English the source article (Darcy 2018) names seven central features likely to matter for most learners whatever their first language: word stress; natural intonation and thought groups; connected speech and linking; syllable timing and vowel reduction; pausing and fluency; final and important consonants; and vowel duration. It also says there is no gold-standard priority list: priorities are set by diagnosing what learners produce spontaneously, including their perception, and by considering functional load (how many words a contrast distinguishes) when choosing between segmental contrasts. Phonological proficiency does not line up with programme proficiency levels, so the list is set per group and adjusted per learner.

No claim in this wiki tests a balanced three-area curriculum against a segmental-only course, a prosody-first course or anything else, and none tests diagnosis-based selection of features against a fixed list. What the claims do show is narrower. Explicit pronunciation instruction reliably improves learners' controlled production of the features it targets, and its gains in spontaneous speech and on global ratings such as comprehensibility are smaller and less certain; that is the main limit on this pattern, whose target is spontaneous speech. Automatic speech recognition (ASR) and other computer-assisted pronunciation training (CAPT) help segmental pronunciation more clearly than prosody or fluency, and short ASR programmes (one to four weeks) showed a negligible pooled effect. One second-hand report (Park 2000, via Darcy) found form- plus meaning-focused instruction beat form-focused instruction alone. So the sequence below is the earlier page's guidance made concrete, with each step labelled with its evidence status; the response-dependent branches are untested design proposals.

How it sits beside converted pages: Deliberate Practice owns targeted practice on one diagnosed component of a performance the learner can already attempt; each feature cycle here is an instance of it, applied to speech. The Formative Assessment pattern owns the general elicit–interpret–act–reobserve loop; the diagnosis and re-observation steps here are that loop with a pronunciation-specific sample and feature list. Immediate Feedback owns when corrective feedback should arrive. This page owns only what is specific to pronunciation: which features, from which areas, in what order, and judged on what speech.

Inputs: establish the design brief

Field Record or elicit
Learner and starting observation A recorded sample of each learner's spontaneous speech (one to two minutes on a familiar topic), a short read-aloud of a passage containing the candidate features, and a perception check (can the learner hear the difference between, for example, stressed and unstressed syllables, or a minimal pair). Record their first language(s), age band (child, adolescent, adult), and what they say listeners find hard. Keep the recording; it is the baseline. Do not infer pronunciation level from the programme's proficiency level.
The features Which of the seven central features (or the target language's equivalents) the samples show as disrupting understanding; which segmental contrasts carry high functional load for these learners; which errors are global (block the message) and which local.
Context and activities Setting (classroom, online-instructor-led, online-self-paced, workplace-clinical), whether learners join remotely or by phone, class size, minutes per week available for pronunciation, number of weeks, whether the teacher has training in phonology, and whether learners have devices for ASR or recording.
Learner-valued goal Ask whom the learner needs to be understood by (customers, patients, colleagues, examiners, a partner's family) and in what situations; whether they want accent reduction or only to be understood; what embarrasses them. Record disagreement with the designer's objective.
Designer objective Which is the target: intelligibility (listeners recover the words), comprehensibility (listeners understand without effort), fluency, accuracy on named features, or a test score. Name the one the final assessment will weigh.
Outcome Ratings of spontaneous speech by listeners like the learner's real ones, on the same prompt type as the baseline; scored apart from accuracy on practised sentences; immediate and delayed (some weeks after the strand ends).

A request to "add pronunciation to the course" does not specify these. Ask for a spontaneous speech sample, the listeners who matter and the minutes available before choosing features or tools.

Sequence and conditional policy

The default arc runs over one term of about 10–12 weeks, with 10–15 minutes of pronunciation work in every lesson (two to three lessons a week) and short self-practice between lessons. These quantities are proposals; the only duration evidence is that ASR programmes of one to four weeks pooled to a negligible effect while five weeks or more pooled to medium or large ones, and that longer interventions and those with feedback showed larger effects in a broader meta-analysis.

  1. Diagnose (week 1, about 3–5 minutes per learner). Record each learner's spontaneous sample, read-aloud and perception check (inputs, above). Listen for the seven central features, note which errors stop you understanding, and note the learner's first language. Where devices allow, an ASR transcript of the read-aloud can flag the words each learner tends to mispronounce for the teacher to confirm by ear. Evidence status: the diagnosis itself is untested; from the earlier page. ASR transcription as a diagnostic aid [+M] (one quasi-experiment with Korean adult learners; the authors' interpretation, with short function words the most common errors). First-language interference [~M] (a narrative digest) says the first language shapes errors in aspiration, intonation and rhythm, which is why the list is set per group.
  2. Choose a short list (end of week 1). Pick three to five features for the term, with at least one from each area (one suprasegmental, one segmental, one fluency feature) unless the diagnosis shows an area is not a problem for this group. Rank by how much the error disrupts understanding: an error that hides the message (wrong word stress on a key word, a missing final consonant that changes the word, no pausing between thought groups) comes before one that only marks an accent. Among segmental contrasts, prefer those that distinguish many common words (high functional load). Aim at intelligibility, not a native-like accent. Evidence status: untested; from the earlier page. Global and local errors [+W] (a theoretical review) supports prioritising errors that block the message. Adults and native-like pronunciation [~M] (a narrative digest) is the reason to set intelligibility, not native-likeness, as the target for adults.
  3. Teach one feature at a time in a two- to three-week cycle, with explicit attention. Each lesson's 10–15 minutes runs: hear it (2–3 minutes: contrasting examples, learners mark or discriminate) → say it under control (4–5 minutes: phrases and sentences, with feedback from the teacher, a peer or a tool) → use it in a short guided exchange (4–5 minutes). Evidence status: explicit pronunciation instruction [+S]: two meta-analyses (86 reports; 77 studies) find instruction improves pronunciation, reliably for controlled production of targeted segmental and suprasegmental features, with larger effects for longer interventions and those giving feedback.
  4. Run the segmental feature through self-practice with ASR or other CAPT where devices allow (about 10 minutes, two to three times a week, for at least five weeks). Learners read target sentences to the tool, compare the transcript with the text, and repeat the words it misrecognises. Evidence status: guided ASR practice [+S] (a meta-analysis of 15 ESL/EFL studies, g = 0.69 against non-ASR conditions, larger for learners 18 and over; one four-week quasi-experiment with Korean learners); CAPT for segmentals, not yet prosody [~S]: large pooled effect on segmentals (g = 0.82), small and uncertain on suprasegmentals (g = 0.37, interval crossing zero). Within one four-week programme, practice days per week and session length [~S] showed no observable difference, and no single error type [~M] improved significantly, so do not expect the tool alone to fix a named sound in a few weeks.
  5. Run the suprasegmental and fluency features with human models and feedback: shadowing a model recording, marking thought groups and stress on a script, then telling the same short story or explanation two or three times to new partners (task repetition), with the listener noting where they lost the thread. Evidence status: CAPT is weakest here (CAPT [~S]; Fluspeak [~W]: one ASR program, 36 university students, good for beginners on phonemes and poor for advanced learners working on fluency, reported second-hand). Immediate task repetition [+W] (a design article reviewing task-repetition studies) reports better fluency on the repeated task and less anxiety. Prosody may need longer: in one reading-fluency study with third-grade learners of English, an AI reading tool [~M] improved speed and accuracy but not prosody.
  6. Put every feature into meaning-focused communication in the same lesson or the next: an information-gap task, a role-play of a situation the learners named, a short talk to listeners who must act on it. The listener's task makes intelligibility the criterion. Evidence status: form plus meaning beat form alone [+W] (Park 2000, reported second-hand by Darcy; design not described).
  7. Re-observe spontaneous speech (weeks 5–6 and the final week) with the same prompt type as the baseline, scored by listeners on the features in the list and on overall ease of understanding. Move a learner off a feature when it is accurate in their spontaneous speech on two consecutive samples, not when it is accurate in drills; then add the next feature from their diagnosis. Repeat the sample some weeks after the strand ends. Evidence status: the move-on rule is untested. Explicit instruction [~S] is why the check is on spontaneous speech: across 77 studies, gains on global measures and in spontaneous speech were not distinguishable from zero between groups, so a drill score overstates what will carry into talk.

Progression across the term. Early weeks: the teacher models, perception work is a larger share of the segment, practice is controlled, feedback comes from the teacher. Middle weeks: learners practise with ASR or peers between lessons, feedback shifts to peers and the tool, tasks become guided exchanges. Late weeks: learners choose their own feature from their re-diagnosis, monitor their own recordings against the baseline, and practise in unscripted tasks with real listeners; the teacher gives feedback mainly when understanding breaks down. All untested proposals; from the earlier page's diagnosis-first guidance.

Observation Discriminating probe Proposed action, with its uncertainty
Feature accurate in drills and read-aloud, absent in spontaneous speech Give a short unscripted task with a listener who must act on what is said; record and replay the moment it slipped. If the learner hears the slip on replay, the feature is known but not yet automatic: more communicative repetition and self-monitoring. If they do not hear it, return to perception work. Untested.
Learner cannot produce the feature even under control Perception check on the same contrast (same/different, or mark the stressed syllable). If they cannot hear it, teach perception first. If they hear it but cannot say it, give articulatory or rhythmic support (tapping, gesture, slower model) and controlled practice. Untested.
Listeners still struggle although the listed features improved Have a new listener transcribe a sample; mark where the transcript fails. The list may have missed the disruptive feature; re-diagnose rather than drill harder. Untested.
ASR keeps misrecognising a word the teacher hears as clear Teacher and a second listener check the word by ear. Treat it as a tool limit, not an error; drop that word from tool practice. Untested.
Learner speaks accurately but very haltingly Time pauses on a retold story, first and third telling. If pauses shrink with repetition, keep repetition tasks; if not, check whether vocabulary or grammar, not pronunciation, is the block. Untested.

Fitting the design to a situation

The three facts that most change the decision:

  • What the learners already produce spontaneously, and whether they can hear the features. It sets the list and whether a cycle starts with perception or production. If unstated, ask: can I hear a one- to two-minute recording of each learner talking, and can they tell the target contrasts apart?
  • Who must understand them, and in what situations. It decides whether the target is intelligibility to a particular group of listeners, a test score or accent reduction, and which tasks carry the communicative step. If unstated, ask: whom do they need to be understood by, and where does it go wrong now?
  • How many minutes over how many weeks, and who can give feedback. Short programmes showed little, and most of the evidence on feedback comes from instruction where someone could hear each learner. If unstated, ask: how many weeks, how many minutes a week, how many learners per teacher, and do learners have phones or computers that can record?
If the brief says… Then change… Basis
Learners are beginners (novice) with limited vocabulary Start with segmentals and word stress on words they already know; use ASR read-aloud for segmentals; keep communicative tasks short and scripted at first. Fluspeak [~W] (good results for beginners on phonemes; one program, reported second-hand); CAPT [~S]
Learners are advanced and want fluency or ease of understanding in long turns Weight the list towards thought groups, pausing, linking and intonation; use repeated retellings and listener feedback, not ASR, for these. Fluspeak [~W] (poor for advanced learners seeking fluency); immediate task repetition [+W]
Learners are child (under 12) Keep segments to 5–10 minutes, use rhymes, chants and games for stress and rhythm; if ASR is used, short individual turns with an adult nearby. Expect prosody to move slowly. AI reading tool, prosody null [~M] (third-grade learners, oral reading, not spontaneous speech); individual or peer ASR practice [~M] (unsettled; third-graders did best individually in one study)
Adults in work settings (nurses, call-centre staff, engineers) with named listeners Take the feature list and tasks from their work talk (handovers, phone calls, explanations to clients); judge success by whether those listeners understand; set intelligibility, not accent, as the goal. Adults and native-like pronunciation [~M] (narrative digest); the work-task focus is untested
Mixed first languages in one class Teach the shared central features (stress, thought groups, final consonants) to the whole class; give segmental contrasts as individual self-practice from each learner's diagnosis. First-language interference [~M]; the split is untested
One teacher for a large class, or scarce facilitator time (a volunteer tutor, a brief check-in) Diagnose from a short read-aloud plus one spontaneous answer per learner, ideally screened by ASR transcripts; run segmental practice as ASR self-study; spend the teacher's minutes only on suprasegmental and fluency feedback and on the re-observation samples. Resource scarcity and large classes [~M] (a review of Bangladeshi classrooms) names the constraint; ASR diagnosis [+M]; guided ASR practice [+S], carried from Korean and other ESL/EFL learners
Mixed modes (some learners remote, some in the room) or phone-only Collect the diagnostic and re-observation samples as voice messages so every learner gives the same sample; pair remote and in-room learners for the communicative step by call; ASR keyboards on phones serve the segmental practice. Untested proposal
A single session only Do not attempt the three-area arc. Record one spontaneous sample, give each learner one or two features that most disrupt understanding with a model and a self-practice routine, and arrange a later recording if possible. Untested; short ASR programmes pooled to a negligible effect (guided ASR practice [~S])
A fixed short time budget (for example 10 minutes a lesson for 6 weeks) One feature per two lessons: 2 minutes hear, 4 minutes controlled practice, 4 minutes in a communicative exchange; at most three features in the term; re-observe in the last lesson. Untested proposal; explicit instruction [~S] (longer interventions showed larger effects)
Learners with low confidence or anxiety about speaking Practise in private first (recordings, ASR, pairs) before any whole-class turn; use repetition so the third telling feels easier; correct only errors that block understanding. Immediate task repetition [+W] (less anxiety, especially for less confident learners); global and local errors [+W]
Learners with low literacy in the target language Teach perception and production by ear and gesture, not from spelling or phonetic symbols; avoid ASR transcript comparison until learners can read the text. Untested proposal
Recordings are private or sensitive The learner keeps their recordings and decides who hears them; the teacher hears only the diagnostic and re-observation samples the learner shares; no class playback without consent. Untested proposal
The brief names its own routine, rubric or speaking test Keep it; score it as given, add a spontaneous-speech rating beside it if it measures only read-aloud or drills, and change only the feature list and the minutes per feature. Untested proposal; explicit instruction [~S] (drill gains overstate spontaneous gains)
The goal is a whole performance for a real audience (a presentation, a client call) Keep the performance whole; take the features from a recorded rehearsal, practise them in isolation for a few minutes, then rehearse the whole performance again for a listener. Untested proposal; form plus meaning [+W]
The learners' reported error is known (for example "people don't understand my numbers") Use it as the first compared case: record it, contrast with a model, and make it the first feature in the list. Untested proposal
The teacher has no phonology training, or no pronunciation assessment exists in the programme Limit the list to the seven central features with recorded models; use the re-observation samples as the programme's pronunciation assessment; seek short training in diagnosis. Teachers value but seldom teach pronunciation [~M] (14 teachers, one programme); teachers unprepared [~M] and no pronunciation assessment [~M] (one review, Bangladesh)
High stakes (a certification speaking test) Score the test's own criteria on practice samples; keep the spontaneous-speech re-observation; do not drop fluency features because the test weights accuracy. Untested proposal

Twelve of the seventeen rows cite a claim, four of them only as background to a move that is itself untested, and each claim is carried beyond its own population or design as the Basis cell says; the other five rows (mixed modes, low literacy, private recordings, a reported error, high stakes) are untested proposals.

Choosing configurations from evidence

Do not rank a balanced three-area curriculum against a segmental-only or prosody-first one from these claims: none compares them. A precise abstention names the missing comparison (a diagnosis-based list across three areas against a fixed segmental syllabus, on listener-rated comprehensibility of spontaneous speech some weeks after instruction) and the local observation that would stand in for it: the same spontaneous prompt, recorded at the start, middle and end, rated by listeners like the learners' own.

The evidence by question

Further evidence, not yet read against this model

Claims bearing on pronunciation teaching that were read for this rewrite but are not part of the model above.

Illustrative design instance and observation record

Invented illustration, not a tested course. An evening class of 18 adult learners of English working in care homes, with Tagalog, Polish and Nepali as first languages, meets twice a week for 12 weeks; a few join by phone. The brief says residents and nurses often ask them to repeat themselves. In week 1 each learner records a one-minute voice message describing a resident's morning and reads a short handover note aloud. The teacher, who has no phonology training, listens for the seven central features: most learners stress the wrong syllable in medication and body-part words and run handovers together without pauses; several drop final consonants that change meaning ("fell" for "felt"). The class list is word stress, thought groups and pausing, and final consonants, with individual segmental contrasts from each learner's sample set as ASR practice on their phones, ten minutes, three times a week. Each feature gets a three-week cycle of 12-minute segments: hear, say under control, then a role-played handover to a partner who must write down the key facts. A learner says she wants residents to stop calling her "the new girl who mumbles"; the objective is that a nurse can act on her handover. They agree to rate the week-6 and week-12 voice messages for whether a nurse-listener can write down the key facts, and to repeat the recording a month after the course. The ASR evidence comes from ESL/EFL learners in general, not this population or setting; these are local choices to observe, not settled ones.

Record: learner's first language and listeners who matter → baseline spontaneous sample, read-aloud and perception check → features listed, by area, with the reason each was chosen → per feature: minutes, practice type (perception, controlled, ASR, communicative), feedback source → mid and end re-observation on the same prompt, rated by listeners → candidate interpretations (feature known but not automatic, not perceived, wrong feature listed, tool misrecognition) and their basis → delayed sample. Preserve the learner's stated purpose and any change in it. An agent may summarise this record but must not fabricate missing inputs or assign numerical probabilities of intelligibility without a calibrated model.

Elements and limits

Modeling, practice, feedback, speech-to-text, observation and shadowing, self-assessment and assessment. Principles and patterns it draws on: Deliberate Practice, Formative Assessment, Immediate Feedback and Post-task focus on form. Strategies that fit its steps: integrate pronunciation, start early, raise its priority, address mother-tongue interference, ASR read-aloud, compare, repeat, multisensory pronunciation exercises and technology-integrated pronunciation practice.

This pattern is scoped to pronunciation for intelligibility and comprehensibility in a second language. Reading fluency for young readers, phonics and accent imitation for performance are other configurations. The policy supports observation and design reasoning; whether a diagnosis-based list across three areas improves listener-rated understanding of spontaneous speech more than a fixed segmental syllabus, for which learners, at what cost in teacher time and at what horizon, remains to be tested.

Claims

  • Formative Assessment — the general elicit–interpret–act–reobserve loop this pattern's diagnosis and re-observation follow

Examples

- - Illustrative, not a tested course: a 12-week evening class for care workers, described under the illustrative design instance above.

Key Sources

  • Darcy, I. (2018). Powerful and effective pronunciation instruction: How can we achieve it? The CATESOL Journal, 30(1). doi:10.5070/b5.35963
  • Lee, J., Jang, J., & Plonsky, L. (2015). The Effectiveness of Second Language Pronunciation Instruction: A Meta-Analysis. Applied Linguistics, 36(3), 345–366. doi:10.1093/applin/amu040 (via the explicit-instruction claim)
  • Saito, K., & Plonsky, L. (2019). Effects of Second Language Pronunciation Teaching Revisited: A Proposed Measurement Framework and Meta‐Analysis. Language Learning, 69(3), 652–708. doi:10.1111/lang.12345 (via the explicit-instruction claim)
  • Ngo, T. T.-N., Chen, H. H.-J., & Lai, K. K.-W. (2024). The effectiveness of automatic speech recognition in ESL/EFL pronunciation: A meta-analysis. ReCALL, 36(1), 4–21. doi:10.1017/S0958344023000113 (via the ASR claims)
Page metadata
Field Value
type pattern
id balanced-pronunciation-curriculum-three-areas
title Balanced pronunciation curriculum covering suprasegmentals, segmentals, and fluency, with a short list of central features selected by diagnosis
description Diagnosing learners' spontaneous speech, choosing a short list of features across suprasegmentals, segmentals and fluency, teaching each explicitly and then in communicative use is expected to improve listeners' understanding of learners' spontaneous speech; no claim tests the balanced curriculum, and the nearest evidence shows explicit instruction reliably improves controlled production of targeted features, with smaller, less certain gains in spontaneous speech.
status review
canonical True
generated by claude/unspecified on 2026-10-15
trust unverified — no one has confirmed this page's content
sources 1 — darcy-2018

Frontmatter is the page's machine-readable record — see the Schema & Guide.