Pronunciation Training in Second Language Acquisition: What 86 Studies Show

Pronunciation Is a Skill, Not a Gift

"I just don't have an ear for languages" is one of the most common reasons people give for avoiding speaking a foreign language. Research disagrees. A meta-analysis by Lee, Jang and Plonsky (2014) covering 86 studies found that pronunciation instruction produces an effect size of d=0.89, classified as large by SLA calibration standards where d≈0.4 is small, d≈0.7 is medium, and d≈1.0 is large (Plonsky & Oswald, 2014; DOI: 10.1111/lang.12079). Pronunciation can be taught and learned. The question is how.

What Lee et al. (2014) Found

Analysing 86 primary studies, Lee, Jang and Plonsky (2014; DOI: 10.1093/applin/amu040) identified three consistent predictors of larger gains:

• Feedback. Instruction with corrective feedback outperforms instruction without it. Students benefit from knowing when their output deviates from the target.
• Duration. Longer programmes consistently outperform short ones. Sustained practice across several weeks beats an intensive single session.
• Explicit instruction. Explicit explanations of how sounds are formed outperform purely implicit exposure (Norris & Ortega, 2000; DOI: 10.1111/0023-8333.00136). Listening to native speakers alone is not equivalent to understanding articulation.

High Variability Phonetic Training (HVPT)

HVPT exposes learners to a target sound produced by many different speakers rather than just one. A meta-analysis by Sakai and Moorman (2017; DOI: 10.1017/s0142716417000418) across 14 studies found HVPT improves not only perception but also production, with d=0.54, a medium effect. When the brain processes the same phoneme from 20–30 different voices, it builds a more robust perceptual category, which in turn reduces production errors. Variety of voices in learning materials is therefore a functional requirement, not an aesthetic preference.

ASR as a Pronunciation Tool

Automated speech recognition (ASR) systems now serve as an accessible feedback mechanism for learners. A meta-analysis by Ngo et al. (2023; DOI: 10.1017/s0958344023000113) covering 15 studies found an overall effect of g=0.69, with an important breakdown:

• Individual sounds: large effect: ASR is effective at detecting and flagging phoneme-level errors.
• Intonation: small effect: current systems struggle with prosody and sentence-level rhythm.
• Social format matters: partner-based sessions outperform solo practice, even when ASR provides the feedback.

Transfer: Trained vs. Untrained Sounds

Does pronunciation improvement generalise? Uchihara et al. (2024; DOI: 10.1017/s0142716424000195), meta-analysing 31 studies, found:

• Trained sounds: 10.50% gain
• Untrained sounds: 4.50% gain

Transfer exists but is roughly half the size of direct training. Targeted phoneme-pair practice still matters for specific problem sounds, but general conversational practice does move the needle for the broader phonological system.

Prompts vs. Recasts

Lyster and Saito (2010; DOI: 10.1017/S0272263109990520) meta-analysed 15 studies (N=827) comparing two corrective feedback types. Prompts (where the teacher signals an error and elicits self-correction) outperformed recasts, where the teacher simply models the correct form. When learners must produce the corrected form themselves, the memory trace is stronger.

How Lexy Approaches Pronunciation

Every Lexy lesson across 49 languages consists of seven connected stages: 1. Vocabulary (flashcards and quiz), 2. Warm-up (written chat), 3. Lesson (25-minute live voice dialogue with the avatar without typing), 4. Exercises (phrase drills and tests with keyboard input for grammar grounding), 5. Dialogue (two-voice situational listening), 6. Story (reading story with translations), and 7. Memory Flashcards (spaced repetition of the previous lesson from lesson 2 onward).

On Stage 3 (Lesson), students speak aloud rather than type. ASR processes every utterance, enabling real conversational exchange across 49 languages including Arabic, Georgian, and Chinese. The partner-style dialogue format aligns with Ngo et al.'s (2023) finding that interaction outperforms solo ASR practice. Stages 5 and 6 provide varied native voices for perceptual training. In lessons 1–15, Lexy does not explicitly correct pronunciation on Stage 3, consistent with the principle that explicit feedback is most effective once a learner has established communicative confidence.

FAQ

Can I improve pronunciation without a human teacher?

Yes. Ngo et al. (2023) found g=0.69 for ASR-based instruction, comparable to many classroom benchmarks, provided feedback is present and varied speakers are used.

How long does it take to see results?

Lee et al. (2014) consistently show longer programmes outperform short ones. Detectable improvement typically requires several weeks of regular practice rather than one or two sessions.

Why is ASR weak on intonation?

Commercial ASR systems are optimised for word-level decoding accuracy. Prosody modelling (rhythm, stress, and melody) requires separate architectures that remain less mature in deployed consumer systems.

Limitations

Meta-analyses aggregate heterogeneous studies differing in target language, learner level, programme length, and outcome measurement. Effect sizes are averages; individual results will vary. Most primary research was conducted in classroom settings, so direct transfer of d/g estimates to self-study app contexts requires caution.

Key Takeaways

1. Pronunciation instruction yields d=0.89 across 86 studies, a large effect (Lee et al., 2014).
2. HVPT improves both perception and production: d=0.54 (Sakai & Moorman, 2017).
3. Prompts outperform recasts for lasting pronunciation gains (Lyster & Saito, 2010; N=827).
4. Trained sounds gain 10.50%; untrained sounds gain 4.50% (Uchihara et al., 2024).

🎯 90-Second Quiz 1 / 5

Test Your Knowledge: Article Quiz

90-second quiz: check whether you understood the key themes of this article. 5 quick questions. If you make a mistake, we explain it right away.

1. What relationship between speech perception and pronunciation is established in SLA?

Articulatory pronunciation training directly sharpens auditory speech perception Perception and production use unrelated brain hemispheres Pronunciation should only be addressed at C2