Papers

Rhythm in language learning: what the evidence supports, and what it does not

Chants and songs are among the best-liked techniques in language teaching. They are also among the worst documented. Anyone who wants to use rhythm seriously has to hold both facts at once, because the second one decides how loudly you are allowed to promise anything.

This article is an honest inventory. It is written by people building a rhythm-based trainer, which is exactly why the weak parts are listed first.

The weak part: chants

There is no meta-analysis of chanting. There are no published effect sizes. What exists is a scatter of small studies - one school, one class, often a student thesis - and most of them are before-and-after measurements with no control group. Positive reports come from Angola (fluency, accuracy, pronunciation, intonation), from Indonesia (pronunciation, eighth grade, no control group), and from action research with younger learners. Motivation results are consistently good.

A before-and-after design without a control group cannot separate the chant from the practice. A class that spends twenty minutes repeating "she works at night" out loud will improve on "she works at night" whether or not there was a beat under it. Until someone runs the same practice with and without the beat, the honest statement is: students like it, teachers report gains, and the specific contribution of rhythm is unmeasured.

The stronger part: singing

Singing has better evidence than chanting, and it is worth being precise about what was measured.

Ludke, Ferreira and Overy (2014, Memory & Cognition) taught 60 adults twenty Hungarian phrases. The group that sang the material outperformed the speaking groups on recall, recognition and production - and the advantage did not depend on how musical the participants were. Good, Russo and Sullivan (2015) found an effect in children that was still visible six months later. A 2023 primary-school study (N=57) found that singing beat speaking on grammar, and the difference survived a delayed test.

For pronunciation, singing and chanting come out roughly equal (study with Chinese teenagers, 2023). So the gain from singing seems to sit in memory rather than in articulation.

Then the meta-analyses disagree. One 2021 meta-analysis covering around 1,900 learners reports a large effect of songs on vocabulary - we deliberately do not quote the number, because published sources give different values for it, which is itself a warning. A 2018 meta-analysis of vocabulary methods put songs last among the methods compared. And a 2024 review in System concluded that the causal claims in this literature are not supported by the designs used to make them.

Two gaps are worth naming. Adults are thinly studied: most of this work is with children. And for learners whose first language is Hebrew, Arabic or Russian there is essentially nothing - we found no studies at all for Hebrew.

The theory that is often used to oversell

A common argument runs: English is stress-timed, Spanish is syllable-timed, therefore Spanish-speaking learners need rhythm work more than Russian or Arabic speakers do. The classic classification does place English, Russian and Arabic on the stress side and Spanish on the syllable side. But the classification itself is contested in phonetics, and there are no direct studies comparing how much chant practice helps learners from different first-language backgrounds. If the classification is even partly right, the prediction follows - and the prediction has not been tested.

What is genuinely solid nearby

Pronunciation instruction in general has a positive research record, and the shape of it is informative: work on perception tends to improve accuracy of sounds and intonation, while work on production tends to improve intelligibility and fluency. Training that targeted the timing of short phrases with Japanese learners produced measurable improvement in segment durations. In other words, the thing rhythm practice is physically closest to - control of timing and prominence in a spoken phrase - is also the thing with real support behind it.

What nobody has measured

Three open questions we care about enough to be building measurements for:

1. Can an automatic judge tell correct delivery from a form error (she wake), from a broken rhythm, and from silence - while a backing track is playing? No published work tests speech scoring over a beat. Noise, including musical noise, is known to damage word-level scoring. 2. Can beat accuracy be scored honestly from word timestamps? Speech engines return per-word onsets and durations and a prosody score covering stress, intonation, tempo and rhythm, but they do not measure whether a stressed word landed on the beat. That calculation has to be added, with a tolerance and a correction for overall offset. 3. Does word-by-word highlighting help, or do weaker students read instead of listening? The recurring finding in the karaoke literature is "useful but distracting", which argues for a no-text mode as well.

How to talk about this

If you teach with rhythm, the defensible claims are: students engage, repetition becomes tolerable, and timing and prominence are what you are actually training. The indefensible claims are numbers - "x per cent faster", "proven to double retention". The literature does not contain them for chanting, and where it contains them for singing, the sources disagree with each other.

A method can be worth using before it is proven. It is not worth misrepresenting, because the first teacher who checks the references stops believing the rest of what you say.

Sources behind this article: Ludke, Ferreira and Overy (2014, Memory & Cognition); Good, Russo and Sullivan (2015); primary-school singing study (2023, N=57); comparison of singing and chanting for pronunciation (2023); Odo (2021) meta-analysis and a 2018 vocabulary meta-analysis; review in System 124 (2024); MIREX 2020 lyrics transcription results; reviews of pronunciation instruction; phrase-timing training with Japanese learners. Full internal reviews, with per-claim links, were compiled in November 2026.

All articles

Free practice for every rule

Every grammar rule on the site has a trainer behind it. It is free and needs no sign-up.

Report a problem

Page you are on
/en/science/papers/rhythm-02-evidence.html en

English is easiest for us, but write in any language you like.

The report carries only your words, the page address and the language. No screenshot, no logs, no account. We cannot reply to a report, so if you need an answer, write to [email protected]