Heritage language loss

Why do the children of immigrants stop speaking the family language even when the parents never switch?

The pattern is so regular it has a name: three generations. The grandparents speak the family language; the parents understand it and answer in the majority language; the grandchildren have a few food words and an apology. What makes it worth explaining is that it happens in households where nobody switched. The parents kept speaking the language, every day, for twenty years. The child heard it constantly and still did not keep it.

That rules out the explanation everyone reaches for first. If the input never stopped, the loss cannot be about the parents' effort, and it cannot be about loyalty.

Ask a narrower question: what does a child actually need a language for?

Language is not learned as a subject. It is acquired in the course of doing things that require it — being understood, being included, getting what you want, not being the odd one out. So the useful unit is not hours of exposure but the domain: a bounded area of life with its own participants and its own language. Home is a domain. School is a domain. The playground, the workplace, the shop, the phone.

Now trace what happens to a child in a migration household. At four, there is essentially one domain, and it runs in the heritage language. At five, school opens, and it runs entirely in the majority language, along with everything attached to it: friends, playground negotiation, jokes, the register in which insults and secrets and status are transacted. Within a year the child has a whole life whose language is not the family's, and a family whose language covers a shrinking set of topics — meals, chores, affection, discipline.

Here is the step that matters most, and it is the one parents find hardest to believe. Children calibrate their speech to their peers, not their parents. That is why children of immigrants take the local accent rather than their parents', and it is why a language with no peer domain has no route into a child's productive use, however much of it they hear at home. Hearing builds comprehension. Speaking is what builds production, and speaking happens where refusing to speak would cost you something.

So look for the point where the cost disappears. It disappears at the moment the parent understands the majority language. From then on, a conversation in which the parent speaks the heritage language and the child answers in the majority language is perfectly functional — both parties understand both turns. Nothing breaks. And because nothing breaks, nothing corrects it. That two-language conversation is stable, and it can run for a decade, converting a child who could have spoken into an adult who only understands.

The same logic predicts the sibling effect, which is well documented: within one family, later-born children typically end up with weaker heritage proficiency than their older brothers and sisters. Nothing changed about the parents. What changed is that the older child now brings the majority language home and uses it in the one domain that was still holding.

By the third generation there is no domain left that requires the language, and it is not transmitted. Not refused — just never necessary.

Two clarifications, and one moral trap to step around.

First, receptive and productive competence come apart, and the gap is the whole phenomenon. The typical second-generation adult is not a failed learner. They are a fluent listener who never had a setting that required them to answer. That is a different thing, and it is often recoverable — heritage speakers relearn far faster than beginners, because the representations are there.

Second, the three-generation pattern is a strong tendency, not a law. It was described mainly in contexts like the twentieth-century United States, and it slows or stops where a language has a domain that survives school: dense neighbourhoods, continuous new arrival from the home country, religious practice conducted in the language, community schooling, regular return visits, or media and messaging that keep peer use alive. The mechanism predicts its own exceptions — protect a domain and you slow the shift.

The trap is moralising. Because the outcome looks like a choice, it invites a verdict: the child did not try, the parents did not insist, the family did not value it. But the child chose the language every four-year-old chooses, which is the one their friends speak. The loss is a structural fact about where a language is required, not a report on anyone's character.

• Why heritage speakers relearn so much faster than beginners, and what that says about what was retained.
• The policy version of the same mechanism: how a minority language is stabilised by giving it schooling, administration, and broadcast rather than by exhortation.

Every term the collection defines is gathered in the glossary.