Your Grandchildren Will Not Speak Your Language, and No Website Is Going To Change That

Ruben Rumbaut, Douglas Massey and Frank Bean measured how long a language lasts after an immigrant arrives. In Linguistic Life Expectancies: Immigrant Language Retention in Southern California, published in Population and Development Review in September 2006, they took two large surveys of immigrants and their children in Los Angeles and San Diego, defined survival two ways (preferring the mother tongue at home, and speaking the mother tongue very well), and ran life-table methods over the answers. Their conclusion about Spanish, in the biggest Spanish-speaking enclave in the country: "its ultimate demise nonetheless seems assured by the third generation."

Generation three is your grandchildren. If you arrived, your children are generation two and your grandchildren are generation three. If your parents arrived, you are generation two, and the clock is one step further along than you think.

The survey data agrees. Pew Research Center reported in September 2023 that "only 34% of third- or higher-generation Latinos say they can carry on a Spanish-language conversation at least pretty well, with only 14% saying they can do so very well," and that "close to two-thirds (65%) of third- or higher-generation Latinos say they cannot carry on a conversation well in Spanish."

Families did not stop caring. The daily habit thins out, one generation at a time. Pew found in April 2018, using 2015 survey data, that 85% of Latino parents speak Spanish to their children: 97% of immigrant parents, 71% of second-generation parents, 49% of third-or-higher-generation parents. Almost all, then most, then half.

Not a small population and not an exotic one. The US Census Bureau reported in December 2022 that the number of people speaking a language other than English at home rose from 23.1 million, about one in ten, in 1980 to 67.8 million, almost one in five, in 2019. In Australia, the 2021 Census found the top five languages used at home other than English were Mandarin (2.7 per cent), Arabic (1.4 per cent), Vietnamese (1.3 per cent), Cantonese (1.2 per cent) and Punjabi (0.9 per cent). Millions of households, though the generational research is thickest on Spanish in the United States, so treat the third-generation figure as the best-measured case rather than a law that covers every language.

A language does not disappear at a border or in a census. A language disappears in one kitchen, when a grandmother asks a question in her own words and answers herself in English because the food is going cold.

Your children have not lost the language. They have become people who receive the language without returning any of it, and that state has been measured rather than guessed at.

Marina Sherkina-Lieber, Ana Perez-Leroux and Alana Johns studied Inuit receptive bilinguals of Labrador Inuttitut and described them as "heritage speakers who are capable of comprehension, but produce little or no speech." Their 2011 study in Bilingualism: Language and Cognition found real grammatical knowledge still in place. These speakers detected many morphosyntactic violations, though less reliably than fluent bilinguals. The grammar is in there. The speaking does not come out.

Which is the most plausible reason the chain snaps at generation three, and the reason looks like arithmetic rather than sentiment. A receptive bilingual parent produces little speech in the language. A child of that parent receives little. That child produces none. Nobody has followed that exact chain through three generations of one family, so read the sequence as a mechanism that fits the measured pieces, not as a measured result.

Barbara Zurer Pearson's review in Applied Psycholinguistics in 2007 reported that roughly a quarter of children in potentially bilingual environments do not become bilingual at all, and concluded that "the quantity of input has the greatest effect on whether a minority language will be learned," with language status and attitudes about language also playing a role. The five factors the review weighs are input, language status, access to literacy, family language use, and community support including schooling.

So how much your grandchild hears matters more than how correct your grandchild sounds. If you have been holding back because their accent embarrasses them, or because your dialect is not the prestige one, you have the trade backwards. Volume beats polish.

Almost every family in this position made a deliberate decision at some point, and most of the guilt attached to that decision is unearned.

Someone told you your child would fall behind. A school told you to speak only English at home. You wanted an easier run for your child than the one you had, and you were not wrong about what an accent costs in a job interview. Lily Wong Fillmore put the whole trade in the title of her 1991 paper in Early Childhood Research Quarterly: When learning a second language means losing the first. Families were asked to pick, and a great many of them picked the option that paid the rent.

The Pew generational figures are what that decision looks like from ten thousand feet. Ninety-seven per cent, then seventy-one, then forty-nine. Read across millions of households, that is not moral failure. That is a very ordinary response to a very consistent set of pressures.

What you can do now is a different question from what you could do then. Your grandchild is not sitting an exam in your language. There is no cost to them in hearing you, no risk of falling behind, no teacher to answer to. Recording now does not undo the old decision, and nobody is asking you to undo anything. Some of what you know exists only in that language and cannot be handed over in any other.

Two things go when the shared language goes. Everyday conversation, which families feel immediately. And the emotional register, which has no common name and has been measured in a laboratory.

Vivian Tseng and Andrew Fuligni surveyed more than 620 adolescents from East Asian, Filipino and Latin American immigrant families for a study in the Journal of Marriage and Family in 2000. Adolescents who spoke a different language from their parents reported less cohesion and less discussion with both mother and father than peers who shared a language. Those who spoke the native language with their parents reported the highest cohesion and discussion of all.

Most articles stop at that sentence. The same study ran a two-year follow-up, and the follow-up ran the other way. Language use did not predict changes in the parent-adolescent relationship. The quality of the relationship predicted changes in language use. Closeness kept the language going more than the language produced the closeness.

Which reverses the usual advice. If what you want is a grandchild who talks to you, work on the talking and let the language ride along behind.

Ronald Cox and colleagues named the wider pattern Shared Language Erosion in a 2021 paper in the journal Children: adolescents improve in English while losing or never developing the heritage language, parents acquire English much more slowly, and the family ends up without a single language in which everything can be said. They argue that communication gap, rather than a clash of cultural values, drives a lot of what gets blamed on culture.

Then the register. Catherine Harris, Ayse Aycicegi and Jean Berko Gleason wired 32 Turkish-English bilinguals to skin conductance electrodes and had them rate words in both languages. Participants showed greater autonomic arousal to taboo words and childhood reprimands, phrases like "Shame on you!", in their first language than in their second. That is Applied Psycholinguistics, 2003, in a laboratory, with electrodes.

A second language argues. A first language stings. The language you were scolded in, sung to in and prayed over in is not interchangeable with the language you file your taxes in.

Start building your legacy Free build, 50 memories, no card required.

Which language you are asked in changes which memories arrive, so an interview conducted in English will surface fewer memories from the years you lived in another language.

Viorica Marian and Ulric Neisser interviewed Russian-English bilinguals and found what they called language-dependent recall: people retrieved more experiences from the Russian-speaking period of their lives when interviewed in Russian, and more from the English-speaking period when interviewed in English. Journal of Experimental Psychology: General, 2000.

Put that beside the skin-conductance result and the instruction writes itself. The Russian childhood comes back in Russian. The village comes back in the language of the village. Record your life story in English and you will get a real story, and you will never learn which parts failed to arrive.

Machine translation has become genuinely good, and it now reaches well past the big commercial languages. The No Language Left Behind team published Scaling neural machine translation to 200 languages in Nature on 5 June 2024: one model covering 200 languages, evaluated over 40,000 translation directions, with an average 44% improvement in translation quality over the previous state of the art as measured by BLEU. Two hundred languages is an achievement. There are more than 7,000 languages in the world, as the same paper says plainly, and the coverage gap falls hardest on exactly the languages families like yours are losing.

Even inside the covered set, the split is predictable. A translation keeps the facts, the sequence and the answer to the question. A translation loses the rhythm of your particular village, the word your mother used that nobody outside your family used, the joke that works because two words sound alike, the level of formality you use with an elder that English has no grammar for, and the sound of you. Australia's 2021 Census puts Cantonese and Punjabi in the top five home languages, and anyone who speaks both Cantonese and Mandarin can tell you what a system that treats "Chinese" as one thing throws away.

So keep the original. Let any translation sit beside the original as a door in, never as a replacement for the room.

No permission required, nothing to buy, and most of these take a minute or two each. The recipe takes as long as the cooking. Record yourself in your language, and not as a speech. These, in this order, because a small child meets them first and a grown grandchild cannot find them anywhere else.

• Counting to twenty, the days of the week, and the months. The first thing a child's ear learns to hold.

• The lullaby, or the nonsense rhyme, or whatever was sung at you. Sing it badly. Badly is fine and badly is more like you.

• The way your mother called you in for dinner. The actual words and the actual tune of them.

• The blessing, toast, prayer or thing said before eating, if your family has one.

• One recipe, spoken out loud while you cook, in your language, hands busy, including the part where you say how you know when the food is ready.

• The three or four sayings that do not translate. Say each one, then say in English what the saying means and who in your family said that line most. The second half is the part everybody skips and the part your grandchild will need.

• The names. Who is who, which side of the family, which town, which year, who was named after whom.

Then change one habit. Send your children and grandchildren a voice message in your language instead of a text, once a week, thirty seconds. The input research says the amount they hear is the single biggest factor in whether the language is learned at all, and a voice message is the only form of the language that survives distance and a busy week.

Two more free moves. Ask your adult child one question directly: "What do you wish you could still understand?" The answer is usually specific and usually not the one you would have guessed. And write down the words that have no English, the ones you find yourself explaining every time. That list is the real inheritance and it takes an afternoon.

For a longer set of prompts, how to interview yourself sets out the method properly, and the things only you know how to do covers the skills and recipes that live in your hands rather than in sentences. For the plain mechanics of getting your voice on record, recording your life story from the beginning is the practical version. If the fluent speaker in your family is your mother or father rather than you, how to interview your grandparents is written for sitting across the table from someone else, and watching a parent forget is the one to read when the memory is going and the language is going at the same speed. Language is one branch of a larger question, which is what your child will want to ask you at thirty, and that page is the place to start if you are recording for a whole family rather than for the language.

At Afterlife.ai™ you talk to a guide called Idy, in your language. Idy asks questions and follows what you say, so a session runs like an interview and not like a form. The app and the guide run in sixteen languages: Arabic, Chinese, English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Polish, Portuguese, Russian, Spanish, Turkish and Vietnamese.

What you say becomes memories. The version of you built out of those memories, the one your family can put questions to, is what we call a "Persona". Personas answer from what you actually said, so where a subject never came up, the answer still comes out of your own material instead of out of invention. That discipline matters more in a second language than in a first, because a guessed answer is exactly where the wrong idiom and the borrowed accent creep in.

Your voice is cloned from your own recordings, so answers can be heard and not only read. The speech model behind that covers 29 languages, checked 30 July 2026. Those two lists are not identical, and the honest gap is Vietnamese: the app runs in Vietnamese, and Vietnamese is not among the 29 languages the default speech model supports.

Now the weak join, with the obvious bias of the man who built the thing. The second language boundary is where this is least good. When someone answers in a language the guide is not running in, a session still works, but the questions do not land the way the same questions land in a first language, and the follow-ups come out more generic. So if your language sits outside the sixteen, record the fixed pieces first: the counting, the lullaby, the sayings, the names. Those carry whatever language asked for them, and those are the pieces a grandchild reaches for.

Language on the reader's side is a setting, one of those sixteen. Whichever one a grandchild picks is the language the answer comes back in, spoken in your cloned voice. Set the app to your language and your language is what they hear. Underneath, your memories keep the words you actually said, in the language you said them in.

Two limits on that, said plainly. The answer a grandchild reads is composed in their language out of your material, so it is your substance and not your sentence carried across word for word. And there is no side-by-side view today, no original beside a version marked "translation". If that distinction matters to you, and on this subject it should, keep your own audio of the sayings and the names, exactly as you said them, and never rely on any service to hold the only copy.

You nominate who gets access, and no clock releases anything on its own. Release runs through Executor Lock™: a nominated person requests release, and a seven-day evidence-verified hold runs before anything opens, which exists so that a release cannot be triggered quietly or by mistake. Until then the memories are yours to add to, change or remove.

Your first fifty memories are free, no card is asked for, and a free build does not expire. That last part is deliberate. No clock should be running while a grandmother works through the sayings and the names in her own language, and nobody should be charged at the moment they are least able to refuse. If you want to see what the fuller plans include, the plans are listed here.

Recordings only do their job if a grandchild can reach them without knowing which drawer to open. Start with the counting rhyme, in your own language, and let your grandchild find the rest later.

Will this teach my grandchild the language?

No. Personas are not language courses, and hearing you for a few minutes will not make anyone a speaker. What a child gets is exposure to the real sound of the language as one specific person speaks it, plus the ability to reach your words later in life, which is when a lot of heritage speakers go looking. If teaching is the goal, daily input, community and school do that job.

My language is not one of the sixteen. Is this any use to me?

Partly, and the split is worth spelling out. You can record and speak in any language at all, because a recording is a recording. The guide's own questions and the app's screens exist in sixteen languages, so outside that list you would be led in another language while answering in yours. Voice synthesis covers 29 languages, and those are not the same sixteen, so a language outside that set may not sound right when spoken back. Vietnamese is the known gap: the app runs in Vietnamese, the default speech model does not. Ask before paying if your language is not on both lists.

Can my grandchild ask in English and get an answer in my language?

Partly, and here is the exact shape. The app's language is a setting covering the sixteen languages listed above, and the answer comes back in whichever of those the reader has chosen, spoken in your cloned voice. A grandchild who sets the app to your language asks in your language and hears your language. A grandchild who leaves the app in English hears English, in your voice. What the app does not do today is show your original sentence and a translated version side by side, or mark anything on screen as a translation. Your memories are held in the words you said, and the answer on screen is composed in the reader's language out of that material.

Who owns my voice and my recordings?

You do, and that answer should be in writing before you record anything anywhere, including here. What a voice clone is legally, and who may use one after you die, is genuinely unsettled in most countries, which is why who owns your voice clone is worth reading before you upload audio to any service, ours included.

What happens if the company shuts down?

No company can promise permanence, and any service that claims otherwise is selling something. Keep your own copies of your own recordings, on your own drive, in a format you can play without special software. Treat any service, this one included, as a convenience layered on top of files you control.

Can I delete all of this later?

Your memories are yours to delete, and before Executor Lock has run nobody else has access, so the decision is entirely yours. On the cloned voice specifically, ask any service, ours included, exactly what happens to a voice once it exists, because the ability to remove one is not universal and is not always immediate.

Start building your legacy Free build, 50 memories, no card required.

• What your child will want to ask you at thirty is useful if you are recording for a younger family and want to know which questions actually come back.

• The family historian's next step after Ancestry is for readers who have the dates and the documents already and have worked out that not one ancestor left a voice.

• Turning sixty, and what to do with your story is the page to read if you have the time to do this properly and no idea where to start.

• Watching a parent forget is for readers whose fluent speaker is losing the words already, and who need to know what is worth recording first.

• Who owns your voice clone is the legal side, and worth reading before you hand recorded audio to anyone.

The generational numbers come from peer-reviewed demography and linguistics (the Rumbaut, Massey and Bean life-table study, Pearson's input review, Sherkina-Lieber and colleagues on receptive bilinguals) and from primary survey publications by Pew Research Center, the US Census Bureau and the Australian Bureau of Statistics. The claims about what the app does are taken from the product as built, checked against the running code rather than the marketing copy, and from published platform documentation on language coverage, checked on the date above. Where the app does less than a reader might assume, on side-by-side originals and on Vietnamese voice synthesis, that is written into the body of the page rather than left out.

What is not settled: almost all the well-measured evidence on shared language and family closeness is about immigrant parents and their teenage children, not grandparents and grandchildren, and the Tseng and Fuligni follow-up found closeness predicting language use rather than the reverse. So treat the grandparent case as a reasonable extension and not as a measured result. The legal status of a voice clone after death is unsettled in most countries.

If something here is wrong or out of date, tell us through contact and we will correct the page.

1. Rumbaut, R.G., Massey, D.S. and Bean, F.D. (September 2006). Linguistic Life Expectancies: Immigrant Language Retention in Southern California. Population and Development Review 32(3): 447-460.

2. Pew Research Center (2 April 2018). Most Hispanic parents speak Spanish to their children, but this is less the case in later immigrant generations. Data from the 2015 National Survey of Latinos.

3. Pew Research Center (20 September 2023). Latinos' Views of and Experiences With the Spanish Language. Survey conducted 1 to 14 August 2022.

4. US Census Bureau (December 2022). Nearly 68 Million People Spoke a Language Other Than English at Home in 2019. American Community Survey.

5. Australian Bureau of Statistics (28 June 2022). Cultural diversity of Australia, 2021 Census.

6. Tseng, V. and Fuligni, A.J. (May 2000). Parent-Adolescent Language Use and Relationships Among Immigrant Families With East Asian, Filipino, and Latin American Backgrounds. Journal of Marriage and Family 62(2): 465-476.

7. Cox, R.B., deSouza, D.K., Bao, J., Lin, H., Sahbaz, S. and Greder, K.A. (25 March 2021). Shared Language Erosion: Rethinking Immigrant Family Communication and Impacts on Youth Development. Children 8(4): 256.

8. Sherkina-Lieber, M., Perez-Leroux, A.T. and Johns, A. (2011). Grammar without speech production: The case of Labrador Inuttitut heritage receptive bilinguals. Bilingualism: Language and Cognition 14(3): 301-317.

9. Pearson, B.Z. (2007). Social factors in childhood bilingualism in the United States. Applied Psycholinguistics 28(3): 399-410.

10. Marian, V. and Neisser, U. (2000). Language-dependent recall of autobiographical memories. Journal of Experimental Psychology: General 129(3): 361-368.

11. Harris, C.L., Aycicegi, A. and Gleason, J.B. (2003). Taboo words and reprimands elicit greater autonomic reactivity in a first language than in a second language. Applied Psycholinguistics 24(4): 561-579.

12. Wong Fillmore, L. (September 1991). When learning a second language means losing the first. Early Childhood Research Quarterly 6(3): 323-346. Cited here for the trade-off named in its title.

13. NLLB Team (5 June 2024). Scaling neural machine translation to 200 languages. Nature 630: 841-846.

14. ElevenLabs, Models documentation, language coverage for Eleven Multilingual v2. Platform documentation checked 30 July 2026.

Last reviewed: 30 July 2026. Language coverage and platform claims on this page are on a 90-day recheck cycle.