Immersion or explicit instruction — which works, for whom, and when?

This is the oldest argument in language teaching, and it is still running daily online.

One side says: immerse yourself. Listen for thousands of hours, read widely, skip grammar — children never studied grammar and they turned out fine. The other side says: adults are not children; without rules you will listen for years and keep making the same mistakes forever.

Both have evidence. And the interesting part is that the two bodies of evidence do not contradict each other — they are about different things.

What the evidence says about explicit teaching

In 2000, Norris and Ortega synthesised 49 studies on the effectiveness of language instruction. The conclusion was clear: instruction works, and explicit instruction — stating rules, directing attention to form — produced substantially larger effects than implicit instruction. The gap was wide, and the gains held up over weeks.

But the authors themselves raised an important caution: most studies measured outcomes with grammar tests. A grammar test obviously favours the group that was taught the rules. The real question is whether it transfers to spontaneous use.

Ten years later Spada and Tomita answered part of that. Their meta-analysis found the explicit advantage held even on free-production measures, not just discrete-point tests. And a surprise: it helped for both simple and complex structures, not only the ones that are easy to state as rules, as many had assumed. Goo and colleagues (2015) replicated the pattern.

So: explicit teaching genuinely works, and the effect is not merely an artefact of how it was measured.

And a decades-long natural experiment on immersion

On the other side we have something rare: the French immersion programmes in Canada, where English-speaking children take their entire schooling in French for years.

The results are striking in both directions.

On listening and reading, students reach near-native levels. They speak fluently and confidently and are not afraid to communicate. This is the strongest evidence there has ever been for the power of comprehensible input.

But Harley and Swain found something else: after years, students still made persistent grammatical errors — noun gender, verb morphology, tense distinctions. Those errors froze, and no amount of additional input corrected them.

From that observation Merrill Swain proposed the output hypothesis: input alone is not enough. When listening, you can extract meaning while ignoring grammatical form entirely — context supplies enough. Only when you have to produce are you forced to process syntax, and only then do you notice what is missing.

Roy Lyster later showed that adding attention to form inside those same immersion classrooms improved accuracy markedly — without costing fluency.

That is probably the strongest empirical answer to is immersion enough: enough to understand and speak fluently, not enough to speak accurately.

Why: it is a problem of attention

The mechanism rests on an idea from Richard Schmidt: the noticing hypothesis. For a feature to be acquired, the learner has to notice it in the input.

And here is the crux: a great many grammatical features are not salient. They are short, unstressed, and usually carry no information you need.

Take English third-person -s. If you hear he go to school, you understand it completely. The -s adds nothing the context has not already given. So your processing system — optimised to extract meaning fast — passes over it, thousands of times, without your knowing.

Immersion cannot fix this, because immersion runs on getting the meaning, and anything that carries no meaning never draws attention.

That is exactly why explicit instruction works: not by giving you a rule to apply while speaking, but by making a feature noticeable, so that next time you actually hear it in the input.

So, for whom?

This is where the answer gets concrete.

Age shifts the balance. Robert DeKeyser (2000) studied Hungarian immigrants in the United States and found a clean pattern: those who arrived as children reached high proficiency regardless of their language-analytic ability. Those who arrived as adults reached high proficiency only if that analytic ability was high.

His reading: children learn through implicit mechanisms; adults have largely lost access to them and must compensate explicitly. In other words, for an adult, studying rules is not a stylistic preference — it is the replacement mechanism.

Aptitude interacts too. A synthesis by Shaofeng Li (2015) found that language-analytic ability and working memory predict how much a learner benefits from explicit instruction. Wider working memory makes more use of an explanation; other learners need more encounters in context instead.

That carries a kind implication: if grammar explanations do not seem to help you much, it does not mean you are slow — it may mean your mix should tilt further toward input.

And at what stage?

This is the most neglected part of the question, and probably the most practical.

Early: input dominates. Explaining a rule to someone who does not yet have enough language to attach it to is close to useless. Rules need a scaffold of real linguistic experience. What matters here is a large volume of comprehensible input, plus enough vocabulary to make that input comprehensible.

Middle: explicit instruction has the highest leverage. You now have enough language for an explanation to grip, and — more importantly — your errors are becoming habits. This is the window where a well-timed explanation saves hundreds of hours.

Advanced: input dominates again. What separates good from very good — which words go with which, register, what sounds natural against what is correct but nobody says — is far too numerous and too subtle to state as rules. Only volume of exposure delivers it.

There is one more constraint teachers often forget. Manfred Pienemann argued that instruction only works when the learner is developmentally ready for that structure — you cannot skip two stages ahead. This explains a familiar experience: a grammar point is explained three times without landing, and then one day it is suddenly obvious. The fourth explanation was not better; the learner had arrived.

The synthesis the field converged on

Since the 1990s most researchers have settled on a middle position that Michael Long called focus on form.

Three things need separating. Focus on meaning only: pure communication, grammar never mentioned — the classic immersion classroom, which leaves errors frozen. Focus on formS: grammar studied in isolation from communication — the traditional classroom, which produces people who know rules and cannot speak. Focus on form: the main activity is meaningful communication, but attention is pulled briefly to form at the right moments — when an error has just happened, when a learner has just discovered a gap.

DeKeyser supplies the last piece through skill acquisition theory: an explanation gives you declarative knowledge — you know the rule. Practice in meaningful contexts turns it into procedural knowledge — you use it without thinking. Those two stages need different activities. Skip the first and you never notice the feature; skip the second and you own a rulebook.

Applied concretely for Vietnamese learners

The framework answers this fairly cleanly by category.

Learning Chinese. Characters: explicit and systematic, with no alternative — radicals and phonetic components have to be pointed out, because they do not reveal themselves through reading volume. Tones: both — input for the melodic contour, explicit correction where Vietnamese pulls you off. The particle 了 and the 把 construction: necessarily explicit, since Vietnamese offers no model and you will not notice them alone. Vocabulary: mostly input, plus spaced retrieval.

Learning English. Articles and inflectional endings: explicit, and repeatedly — these are exactly the non-salient features input will never teach you. Rhythm and stress: input and imitation, not rules — this is where shadowing earns its place. Academic vocabulary: exploit Greek and Latin roots explicitly, the way you would learn radicals.

In short

Asking which is more effective frames it wrongly, rather like asking whether eating or sleeping matters more.

Input is where the language comes in, and nothing substitutes for volume of it. Explicit instruction is the instrument for directing attention at what input alone will miss — and for an adult that list is considerably longer than it is for a child.

The mix is not fixed. It tilts toward input early, toward explanation in the middle, and back toward input once you are good. It tilts further toward explanation if you are an adult, and further still for features your first language does not have.

The only certainly wrong move is to pick a side and stay there for the whole journey.