Bilingual education research: two stories, one answer.

Ask what the research says about bilingual education and you will get two very different answers, often in the same conversation. One answer is about brains: raising or teaching children in two languages is said to strengthen attention, self-control and mental flexibility — a cognitive upgrade that pays off everywhere. The other answer is about schools: programs that teach half the day in English and half in a partner language, and what happens to the children inside them.

These are separate literatures. They use different methods, run on different samples, and — this is the part that matters — they are not equally healthy. The cognitive-advantage story has spent a decade in the replication wars, with documented publication bias at its centre. The program story has quietly produced some of the best causal evidence in education, including randomized lotteries across whole districts.

Mix them up and real damage follows, in both directions. Schools oversell immersion with brain claims that may not survive the next meta-analysis. And when the brain claims wobble, sceptics dismiss dual-language programs whose actual evidence never rested on them. This article takes the two literatures in turn, and gives each the verdict it has earned.

Two literatures wearing one name

The cognitive question asks whether managing two languages trains the mind in general — whether bilingual children get better executive function, meaning the attention-control machinery that regulates thought and behaviour. Its natural home is the psychology lab: reaction-time tasks, matched samples, small groups.

The schooling question asks something a district can act on. When children — from English-speaking homes and from homes speaking the partner language — are taught the curriculum in two languages, what happens to their reading, their maths, their progress toward English proficiency? Its natural home is the field: program evaluations, admission lotteries, state test records.

Keep the two questions apart and the evidence sorts itself cleanly. The rest of this article walks the cognitive literature first, because its collapse is instructive — and then the program literature, because its strength is usable.

From deficit to advantage

For the first half of the twentieth century, researchers mostly scored bilingual children as duller. The studies behind that verdict tested immigrant children in their weaker language, ignored poverty, and then blamed the gap on bilingualism itself (Peal & Lambert, 1962). The deficit conclusion travelled into schools and into advice given to parents: drop the home language.

Elizabeth Peal and Wallace Lambert’s 1962 Montreal study broke that consensus. Comparing ten-year-old French-English bilinguals with monolingual peers, with controls for background the earlier work had skipped, they found the bilinguals ahead — on verbal measures and nonverbal ones alike (Peal & Lambert, 1962).

The correction carried its own flaw, and the authors’ successors knew it: the study enrolled balanced bilinguals — children already strong in both languages — which builds selection into the sample. But its historical work was done. Bilingualism stopped being treated as a handicap, and the field spent the next fifty years asking the opposite question: is it an advantage?

The advantage era

The modern version of that question took shape around executive function. In an influential 2004 study, Ellen Bialystok and colleagues gave bilingual and monolingual adults the Simon task — a speeded test of resisting interference — and found bilinguals faster and less distracted, with the gap widest in older adults (Bialystok, Craik, Klein & Viswanathan, 2004). The proposed mechanism was elegant: a bilingual mind suppresses one language every time it uses the other, and decades of that suppression train general control.

The idea was plausible, hopeful and testable, and it drew a wave of studies. By 2010, a meta-analysis could pool dozens of them: it reported small-to-moderate positive links between bilingualism and attention control, working memory and awareness of language itself (Adesope, Lavin, Thompson & Ungerleider, 2010).

But an association is all a design like this can deliver. Nobody is randomly assigned a second language. Bilingual and monolingual samples differ in migration history, culture, income and schooling. Every one of those rides along in the comparison. The advantage era rested on small samples and unrandomized groups — a structure that works fine, right up until someone tests it at scale.

The replication turn

The stress test arrived in 2013. Kenneth Paap and Zachary Greenberg ran bilingual and monolingual students through a battery of executive-function tasks — Simon, flanker, antisaccade, colour-shape switching — yielding fifteen separate indicators of executive processing (Paap & Greenberg, 2013). If bilingualism trained general control, it should show up somewhere across fifteen measures.

It showed up nowhere. No indicator favoured the bilinguals reliably, and the handful of significant differences ran the wrong way. The paper’s title announced a turn in the field’s tone: there is no coherent evidence for a bilingual advantage in executive processing (Paap & Greenberg, 2013).

Other large-sample failures followed, and the debate turned to methods. Which tasks actually measure executive function? Why did effects keep appearing in studies with a few dozen participants and keep vanishing in bigger ones? That pattern — significance clustering in small samples — has a familiar, uncomfortable explanation, and in 2015 someone went looking for it directly.

The file drawer opens

Angela de Bruin, Barbara Treccani and Sergio Della Sala did something simple and devastating: they followed the results that never became papers. Conferences are where findings first surface, published or not. The team collected 104 conference abstracts on bilingualism and cognition from 1999 to 2012, sorted each as supporting or challenging an advantage, and then checked which ones later appeared in journals (de Bruin, Treccani & Della Sala, 2015).

The two kinds of result had entered the pipeline in similar numbers. They left it very differently. Roughly 63% of the results supporting a bilingual advantage were eventually published. For results challenging the advantage, the figure was roughly 36% (de Bruin, Treccani & Della Sala, 2015).

Read that as a systems failure, not a fraud story. No individual researcher need do anything wrong: journals prefer positive findings, authors shelve nulls, and the published record drifts away from the evidence. But the consequence is mechanical. Any meta-analysis built only on published studies inherits the tilt — which set up the decade’s final accounting.

The meta-analytic reckoning

The accounting came in 2018. Minna Lehtonen and colleagues pooled the adult literature — 152 studies and 891 effect sizes comparing bilinguals and monolinguals on executive-function tasks (Lehtonen et al., 2018). Taken at face value, the pooled advantage was tiny: roughly g ≈ 0.06, a difference no classroom or clinic could ever detect.

Then they corrected for publication bias — exactly the distortion the conference-abstract study had documented — and the tiny advantage went to approximately zero (Lehtonen et al., 2018). The pattern held across task domains: inhibition, shifting, monitoring, working memory. For adults, the field’s own best accounting finds no reliable bilingual advantage in executive processing.

The claim is not fully dead — childhood and old age are still argued over, and task-measurement problems cut in every direction. But the arc mirrors what happened to music training and IQ: an appealing far-transfer story, an era of small supportive studies, then shrinkage toward zero as samples grew and bias was priced in. Brains do not hand out general upgrades so easily. What bilinguals reliably get is the thing itself — two languages.

Meanwhile, in real classrooms

Here is the turn this article has been building toward: none of the above settles what schools should do. The schooling question was always different — how to teach children who live in two languages, and what happens to any child who learns the curriculum through two of them. On that question the evidence runs the other way, and it is strong.

Start with reading instruction for English learners. Robert Slavin and Alan Cheung’s systematic review compared bilingual approaches against English-only immersion across the qualifying experiments (Slavin & Cheung, 2005). Most favoured bilingual instruction, especially paired approaches that teach reading in both languages across the day. None of the rigorous studies favoured English-only. Teaching a child to read in the home language did not crowd out English. It fed it.

Then came the strongest design in the program literature. Portland Public Schools had more applicants for its dual-language immersion programs than places. Seats went out by lottery — random offers, thousands of applicants, seven cohorts. Jennifer Steele and colleagues tracked the lottery winners and losers through state records (Steele et al., 2017). Winning an immersion place raised reading performance by roughly 0.13 standard deviations at grade 5 and roughly 0.22 at grade 8.

The quieter findings matter as much. Despite spending up to half of each school day in another language, lottery winners showed no reliable losses in maths or science. And English learners in immersion were more likely to be counted English-proficient by grade 6. The clearest gains went to children whose home language matched the classroom’s partner language (Steele et al., 2017). Two languages came at no measured academic price — the second language was simply added.

What the evidence doesn’t show

Both literatures come with boundaries, and an honest account keeps them visible. Here is what the evidence does not say.

• Bilingualism harms children. The old deficit findings dissolved once background was controlled (Peal & Lambert, 1962). Today’s nulls concern extra cognitive perks — no serious result shows a cost.
• The advantage debate is over. The firm null is for adults (Lehtonen et al., 2018). Childhood and old age are still contested, with measurement problems cutting both ways.
• Every immersion program works. The lottery evidence comes from one district with mature programs and real demand (Steele et al., 2017). Implementation quality and staffing are part of the treatment.
• Maths and science gained too. Portland’s maths and science estimates were imprecise, not zero (Steele et al., 2017). “No reliable difference” means no measured cost — it is not proof of no effect either way.
• Publication bias means fraud. The 63%-versus-36% gap is a system property — journals prefer positives and nulls stay in drawers (de Bruin, Treccani & Della Sala, 2015). No individual misconduct is implied.
• Brain claims decide school policy. The program evidence stands on attainment alone (Slavin & Cheung, 2005). Dual-language schooling needs no cognitive-upgrade story to justify itself.

Where the evidence stops

1. 1Bilingualism harms children
2. 2The advantage debate is over
3. 3Every immersion program works
4. 4Maths and science gained too
5. 5Publication bias means fraud
6. 6Brain claims decide school policy

© 2026 FUTURE PROOF™

Choosing programs by the evidence

For school leaders, the two-literatures split turns into a short set of operating rules.

Buy immersion on attainment, not on brains. The defensible case is the lottery evidence: reading gains, no measured cost elsewhere, faster English-proficiency progress for English learners (Steele et al., 2017). That case survives every future meta-analysis of the Simon task. A program sold on executive function is one bad news cycle from a funding fight.

Protect the home language. The deficit-era advice — drop the home language to speed up English — was wrong on the evidence then available, and the reading reviews point the other way: literacy in the home language feeds English literacy (Slavin & Cheung, 2005). Parents deserve to hear this from their school, plainly.

Treat implementation as the treatment. Portland’s results came from mature programs with trained bilingual staff and real community demand (Steele et al., 2017). A district copying the label without the staffing copies nothing. Partner-language teacher supply is the binding constraint, and it belongs in the plan before the press release.

Measure in both languages, and mind the seats. A dual-language program measured only in English is half-audited. And where demand exceeds places, lotteries are not just fair — they are how Portland turned a program into evidence. Districts that randomize scarce seats leave behind the data that settles the next debate.