Listen to two bilinguals talk to each other for five minutes and you will hear a sentence that begins in one language and finishes in another, a noun dropped in from the other side, a whole aside delivered in the tongue the third person in the room does not speak. Almost every learner has been told that this is a failure — that mixing means you have not really learned the language, that it is laziness, or worse, evidence that the two languages are somehow tangled up in your head.

The evidence says something stranger and more interesting. Code-switching is one of the most rule-governed things bilinguals do, and the people who do it best are the ones who know both languages best. It is also cognitively expensive at the exact moment it happens. Both of those statements are true at the same time, and understanding why changes what you should do about your own study habits.

What Code-Switching Actually Is

The term was pinned down by the sociolinguist Shana Poplack, whose 1980 paper — its title taken from a line of her own recorded data, \"Sometimes I'll start a sentence in Spanish y termino en español\" — gave the field its working definition and its first serious set of constraints.

\"Code-switching: the alternation of two languages within a single discourse, sentence or constituent.\" — Poplack, Linguistics 18(7–8), 1980

Three kinds of switch fall under that definition, and they are not equally easy:

  • Tag switching — a fixed phrase from the other language slotted in: ¡Oye!, you know?, n'est-ce pas? Grammatically the cheapest move, because nothing has to fit.
  • Inter-sentential switching — one whole sentence in one language, the next in the other. The switch happens at a boundary, so neither language's grammar has to accommodate the other.
  • Intra-sentential switching — the switch happens inside a sentence or a clause, which means the two grammars have to be compatible at exactly that point. This is the hardest kind and the most interesting one.

There is also a category confusion that runs through nearly every conversation about this topic. A single word from the other language is not automatically a code-switch. Linguists separate borrowing — a word that has been absorbed into the receiving language, with its sounds and its endings adapted — from nonce borrowing, a one-off guest word still dressed in its own language's phonology, and from genuine switching, where a whole constituent from the other grammar appears. \"I sent you an email\" is not code-switching; it is Spanish or French or Japanese using an established loanword. \"I sent you a correo\" said inside an English sentence is something else entirely.

Poplack's 1,835 Switches: The Rules Nobody Breaks

The reason Poplack's study is still cited forty years later is what she found when she actually counted. She analysed recorded speech from twenty Puerto Rican bilinguals living on 102nd Street in New York City — some dominant in Spanish, some in English, ranging from people who admitted they could barely manage Spanish to people who moved between the two without seeming to decide anything.

In 1,835 switches, there were virtually no ungrammatical combinations, regardless of how good the speaker's bilingual ability was. People who felt insecure about their second language still respected the structural rules. Two patterns accounted for how this worked:

  • The equivalence constraint. Switches tend to land at points where the two languages' surface structures are compatible — so the resulting sentence is grammatical in both. Spanish and English both allow a determiner before a noun; that is a safe place to switch. Spanish marks adjectives after the noun where English puts them before, which is a bad place — and Poplack's speakers almost never switched there.
  • The free morpheme constraint. You can switch after any constituent except a bound morpheme — you cannot glue an English ending onto a Spanish verb stem and call it a switch.

Then came the finding that should be printed on every language learner's wall. The switch types traditionally dismissed as the most deviant — the dense, intra-sentential switches, the ones that look most like chaos to a monolingual listener — were the ones that demanded the most skill, and they were produced by the most fluent bilinguals in the sample. Speakers with lower proficiency kept to single-word and tag switches. Switching inside a clause is not what you do when you cannot manage a language; it is what you do when you can manage two.

Poplack later showed that even this is community-specific. Comparing Spanish–English bilinguals in New York with French–English bilinguals in the Ottawa–Hull region of Canada, she found that despite involving similar language pairs, the two communities had entirely different switching habits: the Spanish–English speakers produced frequent, varied, intra-sentential switches, while the French–English speakers restricted themselves to single-word and tagged switches. Code-switching is not a universal behaviour with universal norms — it is a local grammar people acquire by participating in a community.

\"Code-switching is a norm in specific speech situations that exist in stable bilingual communities. Satisfaction of this norm requires more linguistic competence in two languages than has heretofore been noted.\" — Poplack, 1980

The Cost: Why Your Brain Slows Down Exactly at the Switch

Now the other half of the story. In 1999, Renata Meuter and Alan Allport at Oxford ran an experiment that is still the reference point for everything after it. Bilinguals sat in front of a screen and named numerals out loud in whichever language a colour cue indicated, with the language changing unpredictably from trial to trial.

Responses on switch trials — trials where the required language differed from the previous trial — were reliably slower than on repeat trials. There was also a reversal that nobody had predicted. The cost was consistently larger when switching into the dominant first language than when switching into the weaker second one, to the point where responses in the \"easy\" language became slower than responses in the \"hard\" one. Meuter and Allport called it paradoxical, and it is the single most counter-intuitive result in the study of bilingualism.

The explanation, fleshed out in David Green's inhibitory control model, runs like this. To speak the weaker language, you do not merely activate it — you actively suppress the stronger one, because otherwise it would keep winning the race. That suppression is heavy, and it does not lift the instant you decide to switch back. So the next time you speak your native language, you are speaking it through the hangover of your own inhibition. The stronger the language you suppressed, the more it costs to bring it back online.

The pattern shows up in comprehension too, though less tidily. Bultena, Dijkstra and van Hell found Dutch–English readers slowing down when a sentence switched languages, with an asymmetry in reading that runs the other way from production. In eye-tracking, code-switched words attract longer first fixations. In EEG, a mid-sentence switch produces a late positive component — a signature of integration work — while a merely unexpected word in a single language produces a different response entirely. Your brain is not treating a switch as a weird word to be guessed at. It is doing real sentence-level repair.

But the Cost Is Not a Fixed Tax — It Depends on Habit

If switching were as expensive as the early lab studies suggest, its prevalence in bilingual communities would be an evolutionary puzzle, and researchers have spent the last decade dismantling the puzzle rather than the behaviour. What they found is that the cost is largely an artefact of how the lab switched languages: by cue, arbitrarily, on material the speaker had no reason to want to say.

  • Frequency matters. When researchers estimated how common each type of switch is in natural Spanish–English corpora and then had bilinguals read matched sentences, the switch types that occur most often in real speech produced the smallest costs. Bilinguals carry a fine-grained statistical model of where switching happens — and they charge you less where it is expected.
  • Naturalistic context matters. When bilinguals choose the switch themselves for communicative reasons rather than following a colour cue, the costs largely disappear. Studies using cued switching and studies using conversational switching have measured very different things.
  • Experience matters. In ERP work by Valdés Kroff, Román and Dussias, participants' self-reported habit of code-switching modulated their brain response, with less experienced switchers showing an earlier positivity that habitual switchers did not. In large studies of Montréal bilinguals, a person's everyday pattern of switching predicted how much control they engaged during comprehension.
  • Balance matters. Costa and Santesteban found symmetrical switch costs in balanced bilinguals, and the dominance asymmetry weakens as the two languages become more equal.

Which resolves the paradox. The cost is real, it is paid at the switch point, and it is the residue of inhibition rather than evidence of confusion. It shrinks as the behaviour becomes demographically normal for you — because you have practised it thousands of times in exactly the contexts where it is useful. Fluent code-switchers describe their mixed variety as the most comfortable way to speak, and they are not wrong. They have amortised the price.

The Child Data: Not a Sign of Confusion

The oldest objection to code-switching is developmental: surely a child raised with two languages mixed together will end up with a muddle? The research is unusually clear. Bilingual children as young as two adjust their language choice to their interlocutor — speaking the majority language to a monolingual stranger and the home language to a parent — which requires them to track who knows what. Genesee, Nicoladis and Paradis showed that patterns of early switching follow the conversational partner's language rather than random leakage.

And the mixed input was never as mixed as parents fear. In a study of bilingual parents talking to their toddlers, code-switching appeared in about 15.8 per cent of all utterances on average — but with enormous variation between families, ranging from 0.4 per cent to 58.5 per cent in one household. Children growing up inside the heaviest-switching environments do not end up language-confused; they end up acquiring the local variety, which happens to include switching.

What This Means for a Learner

Here is where the learner's situation genuinely differs from the community bilingual's, and where the research has practical teeth.

Community code-switchers switch inside a system they already own in both languages. Learners mostly switch by gap: the L2 word is missing, so the L1 word fills the slot. Those are not the same behaviour with different quality. Poplack's equivalence constraint describes people who could have finished the sentence in either language and chose to switch; the gap-filler finishes the sentence in one language because it is unavailable in the other. The first is a sign of skill. The second is a sign of a missing word — which is fine as communication and corrosive as practice, because the moment you take the L1 escape hatch, you have taught yourself that the sentence is achievable without the L2 unit that was missing.

Four things follow from the evidence:

1. Sort your switches into deliberate and default

Keep a note for a week. Every time you catch yourself mixing, mark it chosen (you could have said it in the other language — a tag, an aside, a quotation) or gap (the word simply was not there). The chosen ones are harmless and often rhetorically useful. The gap ones are your syllabus: every gap is a specific, dated, high-value flashcard, collected in the act.

2. Switch at the boundary, not through the middle

If you are going to mix, do what the fluent switchers do and put the switch where both grammars can survive it — at a clause or sentence boundary, or on a fixed tag. This keeps your L2's grammar intact inside the L2 material. The learner version of an illegal switch is building an L2 sentence on L1 word order and calling it code-switching; that is not a switch, it is a transfer error wearing a costume.

3. Protect one hour of monolingual mode

François Grosjean's most useful idea for learners is the language mode: a bilingual sits somewhere on a continuum between a monolingual mode, where only one language is active and the other is largely switched off, and a bilingual mode, where both are live and switching is expected. Most learners spend almost all their practice time in bilingual mode — their tutor explains in L1, the app's interface is L1, the self-talk is L1, and the L2 words are tourists passing through. If you never practise in monolingual mode, you never practise the state you will need in the exam, the meeting, or the emergency. One hour a day of no-exit L2 is worth more than three hours of bilingual comfort.

4. Train the switch itself

Switch costs fall with practice, so practise the movement directly. A three-card drill takes ten minutes:

  • Rapid alternation. Take twenty known words. Say each one in L2, then L1, then L2 again, against a metronome or a one-second timer. You are not learning vocabulary; you are pricing the switch and lowering it.
  • Interference pairs. Choose words where your L1 will shout loudest — cognates, false friends, and anything with the same initial sound in both languages. Those are the trials where inhibition has the most work to do, so they are the trials with the most to gain.
  • Sentence-level switching. Practise pairs of sentences where one clause changes language at a legal boundary, in both directions. This is the same skill Meuter and Allport measured, moved from digits to meaning.

The Bottom Line

Code-switching is not a corruption of bilingualism, and it is not a magic fluency hack either. It is a skilled, rule-governed act — Poplack's 1,835 switches were overwhelmingly grammatical regardless of speaker ability, and the densest switches came from the strongest bilinguals — that nonetheless costs you real processing time at the moment of the switch, because speaking one language means suppressing the other and the suppression takes a moment to lift.

The cost falls as the habit grows, and it falls fastest when the switch is chosen rather than forced. That gives you a clean rule for your own study. Mix languages on purpose, at points where both grammars survive, and log every switch you did not choose — because each of those is a word your flashcards are missing. Then spend an hour a day in monolingual mode, where the escape hatch is locked. Your two languages are not tangled. One is simply doing the work while the other waits — and the waiting is what you are training.