Across every major immigrant destination that measures it, the family language thins sharply in the children of immigrants and is close to gone in the grandchildren: in the United States, 93% of foreign-born Latinos can hold a conversation in Spanish, 69% of their children can, and 34% of the third generation can, according to Pew Research Center's 2022 survey; Statistics Canada found that immigrant mothers passed their language to 55% of their Canadian-born children in 2006, and that the Canadian-born daughters who had themselves received the language passed it on only 23% of the time. The pattern was first described by the sociolinguist Joshua Fishman in the 1960s and has since been confirmed in census data on both sides of the Atlantic and in Australia. The cause is not a lack of love for the language but a lack of input and need: once school, friends and screens arrive in the majority language, a child hears the home language from fewer people, for less of the day, and stops needing it. The research on what prevents the loss points, with unusual consistency, to one lever: sustained, interactive input from several people the child is attached to.
The three-generation pattern, and the data that confirms it
The sociolinguist Joshua Fishman set out the pattern in Language Loyalty in the United States (1966) and built his later work, including Reversing Language Shift (1991), on it: immigrants keep their language; their children grow up bilingual, learning the home language from their parents and the majority language from everything else; their grandchildren grow up monolingual in the majority language. Three generations, roughly seventy years, and a language that took a thousand years to form is gone from a family. Fishman's model has been criticised as too tidy, and the critics have a point: some groups shift faster, some slower, and the timing depends on who marries whom and how dense the community is. But the direction has never been seriously disputed, and the modern census data make it hard to dispute.
Start with the United States, where the measurement is best. A 2002 study in Demography by Richard Alba and colleagues, using 1990 census microdata, found that the grandchildren of Asian immigrants were switching to English-only homes at roughly the pace the grandchildren of the early-1900s European immigrants had, measured in the 1940 and 1970 censuses, and that Spanish speakers were shifting more slowly but shifting all the same. The main thing that decided whether a third-generation child still heard the family language at home was whether the parents had married inside the group and lived among other speakers. Four years later Rubén Rumbaut, Douglas Massey and Frank Bean applied life-table methods to two large surveys of immigrant families in Los Angeles and San Diego and computed what they called "linguistic life expectancies", the number of generations a mother tongue survives after arrival. They chose Southern California deliberately, as the hardest case for shift: the largest Spanish-speaking population in the country. Their conclusion, in the journal's own words, was that even there the language's demise "seems assured by the third generation".
Survey data from Pew Research Center put numbers on each step. In Pew's 2022 survey of 3,029 Hispanic adults, 75% of all US Latinos said they could carry on a conversation in Spanish pretty well or very well. Among the foreign born it was 93%. Among their US-born children, the second generation, it was 69%. Among the third generation and beyond it was 34%, and only 14% said they could do it very well; close to two-thirds of third-generation Latinos said they could not hold a conversation in Spanish. That is a language going from near-universal to a minority skill in two steps, in the group with the largest community, the most media and the shortest distance to the homeland of any immigrant population in the country.
Asian-American families, with smaller communities and more varied languages, shift faster. Pew's 2013 report on second-generation Americans found that only 41% of second-generation Asian Americans could speak their family's language very well (18%) or pretty well (23%); a third spoke "a little" and 26% did not speak it at all. The comparable figure for second-generation Hispanics was 80%. By Pew's 2025 analysis of census data, 68% of US-born Asian Americans spoke only English at home, against 14% of Asian immigrants. These are not fringe populations: the 2024 American Community Survey counts 74 million people aged five and over, 23% of the country, who speak a language other than English at home.
Canada and Australia show the same curve
Canada is the useful second witness because its census asks about mother tongue, and because Statistics Canada has studied transmission directly. René Houle's analysis for Canadian Social Trends compared immigrant mothers in the 1981 and 2006 censuses and asked how many of their Canadian-born children under 18 had received the mother's language, either as a mother tongue or as a language used at home. In 1981 the answer was 41%. By 2006 it had risen to 55%, a real improvement that Statistics Canada attributes mainly to the steady inflow of new immigrants. Persian was among the languages passed on most reliably, in a group above 70% that also included Armenian, Punjabi, Chinese, Turkish, Bengali and Urdu; Dutch, Italian, Creole and Tagalog were below 20%.
Then came the second step, and it is the one that matters for this page. Houle compared the Canadian-born daughters of the 1981 immigrant mothers who had received the language, aged 25 to 42 by 2006, and asked how often they in turn passed the language to their own children. The answer was 23%. Multiply the two steps (41% of the children, then 23% of their children) and roughly one grandchild in ten shares a mother tongue with the grandmother. The 2021 census shows the stock the curve is working on: 9 million Canadians, one in four, have a mother tongue other than English or French, and 68.8% of them already speak English or French at home at least regularly. "Persian languages" were spoken predominantly at home by about 180,000 people in 2021, up 26% in five years, and some 326,000 people in Canada could hold a conversation in one.
Australia's census asks a different question, whether a language other than English is used at home, and the answer is the same shape. A 2023 study in Frontiers in Psychology by Lubei Zhang, Linda Tsung and Xian Qi tracked "language shift", meaning English-only at home, across the 2006, 2011 and 2016 censuses by birthplace. In 2016 the share of overseas-born people who had shifted ranged from 2.6% for the China-born to 65.4% for the Netherlands-born. For their Australian-born children the figures were far higher: 94.7% for Dutch background, 89.7% German, 83.4% Filipino, 82.0% Polish, and even the most retentive group, Vietnamese, had shifted 22.5%. For most groups the gap between first and second generation was more than 30 percentage points. The Australian Bureau of Statistics reports that of the 5.8 million people (22.8%) who used another language at home in 2021, 71.8% were themselves immigrants, which is another way of saying the language is being held mostly by the people who brought it.
Why it happens: input, need and the arithmetic of a child's day
Immigrant parents usually want the language kept. In Pew's 2022 survey, 42% of foreign-born Latinos said it was extremely important that future generations of US Latinos speak Spanish; by the third generation only 13% said so. Wanting is not the mechanism. The mechanism is exposure, and exposure collapses on a schedule that has nothing to do with the family's feelings.
Three things happen at once when a child of immigrants starts school. First, the majority language takes over the hours: school, friends, television, the street. Barbara Zurer Pearson's 2007 review of the social factors in American childhood bilingualism concluded that the quantity of input has the greatest effect of any factor on whether a minority language is learned, ahead of language status, attitudes, literacy and community support, and that about a quarter of children raised in potentially bilingual homes do not become bilingual at all. Second, need disappears. The Linguistic Society of America's guide for parents, by the linguists Antonella Sorace and Bob Ladd, names the two levers as exposure and need; a child who can get everything they want in English has no reason to keep producing Persian, and comprehension outlives speech by years. Third, the parents switch. Lily Wong Fillmore's nationwide survey of immigrant families, published in 1991, found that children's early learning of English in preschool changed the language patterns of the whole household, as the household follows the child into English.
The heritage-language researcher Silvina Montrul, who studies why the second generation's version of a language drifts from the homeland's, is blunt about where the drift comes from: not the quality of what parents say, but the quantity, and the fact that children use the language too little in early childhood, the period that matters most. The census work agrees from the other direction. In Houle's Canadian data, the single biggest factor reducing transmission was a parent married to someone with a different mother tongue; the mother's education, her prior knowledge of English or French and the years since immigration also reduced it. The Australian study found the same link between marrying outside the group and shifting to English. Put plainly: the language survives where it is spoken by more than one adult, needed daily and not competing with a spouse's.
The Persian-speaking diaspora: large, educated, dispersed
Nobody has a defensible count of the Iranian diaspora as a whole. Figures of "four to six million" appear in conference papers and government statements, and the Clingendael Institute, a Dutch think tank, has quoted six to eight million while noting that reliable, up-to-date numbers do not exist; those estimates count descendants, partial ancestry and emigrants from before 1979 in different ways. What can be counted is national census data, and it is considerable.
In the United States, Pew Research Center's March 2026 analysis of the 2024 American Community Survey puts the number of Iranian Americans at about 750,000, of whom about 59% (445,000) are immigrants and 310,000 were born in the US. Nearly half live in California (375,000), with 230,000 in metropolitan Los Angeles; 65% of adults hold at least a bachelor's degree, and 79% have at least some college, against about 58% of other US adults. A separate estimate by the UCLA sociologist Kevan Harris, published in 2023, put the Iranian-ancestry population at 400,000 to 620,000; the gap illustrates how much depends on how ancestry is asked. The 2024 survey counts 570,013 people aged five and over who speak Persian at home, a category that also includes Dari speakers from Afghanistan. In Canada, as noted above, 326,000 people could converse in a Persian language in 2021 and about 180,000 spoke one predominantly at home.
The research specifically on Persian as a heritage language is thin but consistent with everything above. Khadij Gharibi and Frank Boers tested 30 Persian-English bilingual children aged 6 to 18 in New Zealand against 30 monolingual children in Iran matched for age, sex, siblings and family background, and reported in Language Acquisition (2017) that the heritage speakers knew fewer Persian words, with the gap widest among children who had grown up with both languages from birth; for children who emigrated later, age at emigration was the strongest predictor, and for those born abroad it was the parents' attitude to maintaining the language. In interviews with the same families, Gharibi and Corinne Seals found parents committed to Persian at home, struggling mainly with literacy, and getting almost no institutional help. Both are single studies and small. The large-scale signal is Houle's: in Canada in 2006, Persian was one of the languages immigrant mothers passed on most reliably, above 70%. The first step, in other words, is one Iranian families mostly manage. It is the second step, the one that ran at 23% for the daughters in Houle's data, that the next decade will decide.
What it costs a family
The economic literature on bilingualism is large and not the point here. The cost that matters for this thesis is inside the family, and it has been measured. Vivian Tseng and Andrew Fuligni surveyed 620 immigrant parent-adolescent pairs of East Asian, Filipino and Latin American background for the Journal of Marriage and the Family (2000). Adolescents who spoke a different language with their parents than their parents spoke with them reported less cohesion and less discussion with both mother and father; the closest relationships were those where parent and child both used the native language. Over the following two years, the quality of the relationship predicted which language the family used, and language use did not predict the relationship, so this is not a simple causal arrow, but it is a tight knot.
Janet Oh and Fuligni followed up with 414 adolescents from Latin American and Asian immigrant families in Social Development (2010) and found that it was heritage-language proficiency, how well the young person could actually speak it, rather than how often they chose to, that went with better parent-adolescent relationships and a stronger ethnic identity. Wong Fillmore's conclusion from her 1991 survey, in the words of the abstract, was that "primary language loss can be very costly to the children, their families, and society"; her 2000 follow-up in Theory Into Practice described what the loss looks like in a kitchen: parents who can no longer correct, comfort or argue with their children in a language either side fully owns.
Grandparents are the sharpest edge of this, because they rarely learn the new language and the child rarely learns the old one. The evidence here is suggestive rather than large. Hiromi Ishizawa, using the Census 2000 Supplementary Survey in Sociological Perspectives (2004), found that grandchildren living with grandparents who did not speak English were markedly more likely to use the minority language, and that a grandmother in the house had a stronger effect than a grandfather; she argued the three-generation model had missed that a language can skip from grandparent to grandchild directly. That is a finding about the minority of families who share a home. For the far larger number with a grandmother in Tehran, Toronto or Hamburg and a grandchild in another country, the relationship runs through video calls, and the child's side of it runs out of words.
What the evidence says prevents the loss, and the implication
The prevention research is unusually consistent, and it points away from most of what is sold to parents. Pearson's review puts quantity of input first. Annick De Houwer's survey of 1,899 bilingual families in Flanders, in Applied Psycholinguistics (2007), found that every child spoke the majority language but not every child spoke the minority one, and that the best odds came in homes where both parents used the minority language and at most one also used the majority one; the much-advertised "one parent, one language" rule was, in her words, neither necessary nor sufficient. Erika Hoff's work at Florida Atlantic University adds that acquisition depends on the quantity and the quality of a child's language experience and on their opportunities to use it; in one small study of 29 Spanish-English two-year-olds, she and Silvia Place found that the number of different people a child heard a language from, and how much of it came from native speakers, predicted skill beyond the raw hours, at least for English. Houle's Canadian data say the same thing in demographic language: two parents with the same mother tongue, and a community around them, roughly double the odds.
Three conclusions follow. The language has to come from people, not from media alone; comprehension is not enough, the child has to need to speak; and more speakers beat one speaker. The weakest link in an immigrant family's chain is the second step, the one where the bilingual second generation, now parents themselves and often married outside the group, cannot supply enough Persian hours on their own. The people who could supply those hours, and who have the strongest motive to, are the grandparents, aunts and uncles, usually in another country or another time zone.
That is the gap most language products ignore. Apps teach vocabulary to a child who already understands; cartoons supply input without need or relationship; classes supply an hour a week. None of them adds a loved person speaking the language to a child's ordinary day. Azizan is built on exactly this thesis: relatives record themselves on video reading illustrated Persian stories page by page, so the input comes from a known face and voice, repeatedly, from several people, in a form a small child will ask for again. Whether that is enough to move the second step is an empirical question, and the honest position is that it has not yet been measured. What has been measured, in four countries over sixty years, is what happens when nobody tries.
Originally published at azizan.app, where every figure links to its primary source. Azizan (azizan.app) is a family storytelling app: relatives record themselves on video reading illustrated children's stories in the family's mother tongue, page by page.
Top comments (0)