A June day in Annapolis, 2024. Reynolds Tavern, the 1747 Pub downstairs, the kind of place that has been serving tea in that town since before either of our countries looked the way they look now. My wife and I had just arrived at the baby shower for the daughter of old friends. And we were looking for a reason to leave that place where everyone talks only in English.
We came in and saw the room: small tables, set for four or five each. Which meant we would not be sitting together in a corner being quietly foreign. We would be sitting with Americans we had never met, in English, for two hours.
It's not that I don't speak English. But surely my English is not enough for small talk?
Two women sat down with us: an older mother and her adult daughter. The conversation started the way these things start — where are you from, oh, how interesting — and then it caught on something. We were Russian. And their husband and father had taught Russian for many years. At the Naval Academy, I believe, or somewhere like it. He had taught his daughter Russian when she was small. By her own account, by the time she started elementary school she spoke Russian better than she spoke English.
Over the next two hours, she did not say a single word of it. Not spasibo. Not do svidaniya. Not the tourist words, the ones people who have never studied a day of Russian somehow still have. Nothing. The language that had been more hers than English at seven years old was, at forty-something, simply gone.
And the conversation was wonderful. That is the part I did not expect. Word by word it took off, and somewhere in the first twenty minutes I stopped monitoring myself and just talked. I told them about losing my job. I told them about the year I had been driving for Uber Eats. We talked without stopping through the entire meal, and I forgot to be nervous, which is not a thing I can usually arrange on purpose.
When we were saying goodbye, the daughter asked me something I have thought about since.
"How do you manage to keep up a conversation in English so effortlessly?"
And then, before I could answer, she answered it herself:
"Ah — it must be the Uber work that forced you into it."
She was right. She did not know how right.
Two people, one table, opposite directions
Drive home from something like that and you turn it over. Here were two people at one small table in Annapolis, moving in opposite directions through the same phenomenon.
She had a language given to her early, by a parent, in the way languages are given to children — not taught at her, spoken with her. It worked. It worked better than English for a while. And then the person who spoke it with her stopped being the center of her days, and the language went somewhere she could not reach it. Forty years later there is no spasibo left.
I had the opposite. Years of English instruction — school, then university — that left me able to translate a page with a dictionary and unable to say a sentence to a person. What changed that was a job in IT where all the documentation was in English and I had no choice but to read it, and then, much later, a year of delivering food in Maryland where I had to talk to a stranger every fifteen minutes whether I felt ready or not. And a woman at a tea party asked me how I did it so easily.
Between the two of us we had one complete language education and one complete childhood immersion, and they were in different people. I got the schooling. She got the use. Only one of us could hold a conversation in the language in question, and it was not the one who had been taught.
So the difference is not study. It is something else, and I wanted to know what the research actually says about it, because I was about to build a product on the answer.
Can you forget your first language?
The short version: if you emigrated as an adult, essentially no. If you were a small child, absolutely yes — completely.
Start with adults, because that is the case most people worry about. Kees de Bot and Michael Clyne followed Dutch immigrants in Australia and re-tested them sixteen years after their first assessment. The finding was that after the first decade or so, first-language skills are remarkably stable. Not perfectly preserved — but the structure holds. You do not lose Dutch by living in Melbourne for forty years.
What does change is smaller and more specific than folklore suggests. Speech gets a little more hesitant. The range of words you reach for in free conversation narrows slightly. Low-frequency words come more slowly. Pronunciation can drift measurably — long-term migrants are sometimes rated as non-native in their own first language by monolingual listeners, though many are still rated as unambiguously native. Where grammar shifts, it shifts at the edges, in things like pronoun resolution, not in core syntax.
One thing worth naming, because it is the most common misdiagnosis: mixing languages is not evidence of loss. When you drop an English word into a Russian sentence because it is the word that fits, that is not attrition. That is what bilingual speech normally looks like when both languages are switched on.
Now the children.
In 2003, Christophe Pallier and colleagues published a study of Korean-born adults who had been adopted into French families between the ages of three and eight, and who had had no exposure to Korean since. They were tested fifteen to twenty years later. They could not pick Korean out of a lineup of five languages. Given a French word and two Korean options, they could not choose the correct translation above chance. Under fMRI, listening to Korean produced no Korean-specific response — their brains treated it like a language they had never heard.
They performed like French people who had never encountered Korean in their lives. A language that had been their only language at three was, functionally, not there.
The susceptibility fades gradually rather than switching off. Emanuel Bylund's work on maturational constraints describes a decline through childhood that is largely finished by roughly age twelve — and, importantly, far more variable before that age than after it. Two children with the same history can end up in very different places.
And one honest caveat that matters for how far you can push any of this: adoption is a total severance. All input stops on a single day and never resumes. Ordinary emigration is nothing like that — there is a parent, a phone call, a community, a song. So "age" and "complete cut-off" are tangled together in exactly the cases that show the most dramatic loss. The woman at the tea party sits somewhere between those two worlds — nobody took the language away from her, and it went anyway.
Except it does not entirely vanish
Here is where it gets strange, and where I think the interesting thing is.
Pallier's adoptees showed nothing on tests that asked them to use Korean. But a different question — what happens if you try to teach it back to them? — gives a different answer.
In 2017, Sunyoung Choi, Mirjam Broersma and Anne Cutler trained Dutch adults who had been adopted from Korea on Korean consonant sounds, over roughly ten days. The adoptees learned the notoriously difficult three-way Korean stop contrast faster than Dutch controls with no Korean background. Including — and this is the part that stopped me — adoptees who had left Korea at three to five months old. Before they could speak. Before they could remember anything at all.
Similar results elsewhere. Oh, Au and Jun found that Korean adoptees in a first-semester Korean class were already better at identifying certain Korean sounds by week two. Pierce and colleagues scanned Chinese adoptees in Montreal — adopted around thirteen months, French-speaking ever since, no conscious memory of Chinese — and found their brains processing Chinese lexical tone the way a bilingual's does, not the way a monolingual's does.
So something is still down there. It is not language you can use. It is not spasibo coming back at a tea party. It is a residue in the sound system, invisible until you try to build on top of it, at which point it turns out the ground is not empty.
I want to be careful here, because it would be easy and wrong to tidy this up. The picture is not settled. Pallier's group was adopted later — three to eight — than the groups showing the relearning advantage, most of whom were infants, which is awkward for any clean story about age. Researchers who work on this say so openly. Nobody should read "adoptees keep their birth language." They do not. What they keep is smaller and stranger than that.
The shape underneath
What struck me is the shape of it. Something is present. It cannot be reached. Under the right kind of contact, it becomes reachable again.
If you have read anything else on this blog, you will recognize that shape, because it is the shape of the intermediate plateau. You know thousands of words. You cannot produce them at conversational speed. They are present; they are not reachable. And what closes that gap is not more memorizing — it is meeting the words again, in enough varied situations that they come when you need them.
Now, an honest caution. It is tempting to say these are the same mechanism. There is a well-known proposal in the field — Michel Paradis's activation threshold hypothesis — that says roughly this: disuse does not delete what you know, it raises the cost of getting to it. Use lowers that cost again. It is an elegant fit for everything above.
It is also contested. Monika Schmid, who has done as much empirical work on attrition as anyone, has repeatedly failed to find the relationship the hypothesis predicts — people who use their first language more do not reliably show less attrition. She argues the symptoms come from somewhere else entirely: difficulty suppressing the second language during live speech, rather than the first language fading out of reach.
So I will not tell you these are the same machine. What I will say is that they are the same shape, and that the shape has a practical implication I trust regardless of which mechanism wins: language does not live in what you have been told about it. It lives in encounters, and it stays alive for as long as the encounters keep happening.
Both halves of that sentence were sitting at the table in Annapolis. Her half, and mine.
I studied English for years. Then I learned it.
I want to be accurate here, because the tidy version — immigrant teaches himself English, no classroom involved — would be a lie, and it would also be a weaker story.
I studied English at school. The teaching was poor, but it was not nothing: I came out with grammar, a small stock of words, the ability to work out what a sentence was doing. Then at university, like every Soviet student, I did the thousands — a quota of thousands of characters of technical text to be translated each term, in writing, with a dictionary. Nobody spoke. Speaking was not part of the exercise. You sat with a text and a dictionary and you produced a translation, and that was the whole of it.
So by the time I was done being taught, I had years of English behind me. I could translate a technical paragraph given enough time. I could not say anything to anyone.
That is the exact condition this entire blog is about. I knew a great deal about English. I had almost no English.
What changed it came in two stages, and neither of them looked like studying.
The first was work. I was in IT, and the documentation was in English, all of it, and there was no Russian version coming. So I read it. Badly at first, then less badly. I was not trying to learn English — I was trying to make something compile. The English arrived as a side effect of wanting something else.
The second was Uber Eats. I lost my job, and I drove, and driving meant talking: to restaurant staff, to customers, to people at doors, dozens of times a day, in unpredictable order, with no time to compose anything. You cannot translate in your head at a restaurant counter with a line behind you. You either produce the sentence or you stand there. So I produced the sentence.
That is the thing the woman at the tea party identified in one guess. Real-time pressure removes the option of routing through your first language. There is no time for the detour, so you stop taking it, and after enough repetitions the direct path is the only one you have.
But let me refine her diagnosis slightly, because the refinement is the whole reason LAFwords exists.
What Uber Eats gave me was not compulsion exactly. It was frequency. Compulsion just happens to be the most reliable way humans ever get frequency, because it removes the decision. A hundred forced encounters a day will beat five willing ones every time — not because force is magic, but because a hundred is more than five.
Which means the question for anyone who is not about to lose their job and start delivering food is: where do the encounters come from?
Learn as first
This is the principle the whole product is built on, and it took me a long time to be able to say it in one line.
You should meet a language the way you met your first one: through use, before explanation. That is what "learn as first" means — learn it as you learned your first language.
Nobody taught you what a preposition was before you started using prepositions. Nobody explained polysemy to you before you understood that a word can mean several things. You met the same word in enough different situations that its shape settled, and the explanation, if it ever came, came years later and changed nothing about your ability to speak.
In practice, inside LAFwords, that principle looks like a set of refusals.
We do not use grammatical vocabulary in the learning flow. No "polysemy," no "lemma," no "inflection," no "derivation." If a word you have entered turns out to have two unrelated meanings, we do not present you with a menu labeled noun and verb. We show you two short sentences and let you point at the one that matches what you actually read or heard. The sentence does the teaching. The label would only get in the way.
And we give every word ten of them. Ten short sentences spanning different senses, registers, and situations, delivered across spaced repetition rather than all at once. Not because those ten situations are the ones you will need — you almost certainly will not need any of them specifically. Because ten varied encounters give a word enough range that you can deploy it in the hundredth situation, the one nobody wrote a sentence for.
That is the whole design. Encounters instead of explanations, and enough of them, spaced out, so that a word stops being a translation you retrieve and becomes a word you have.
Here is the limit, and I would rather say it than have you discover it: an app cannot compel you. Uber Eats compelled me. No software will. What software can do is make the encounters reliable — varied, well-timed, and there tomorrow whether or not you feel like it. That is a real thing and it is not nothing. It is also not the same thing as a restaurant counter with a line behind you, and any vocabulary app that tells you otherwise is selling something.
If you want the full comparison of what the other tools in this category do and do not close, I wrote that one honestly too.
What to do with this
If you are somewhere on the intermediate plateau, the practical reading of all of the above is short.
Stop trying to learn about the language. You are past the point where that helps. The rules you are missing are not the bottleneck; almost everyone at B1 and B2 knows more grammar than they can execute.
Build encounters instead. Read things written for adults who share your interests, not for learners. Listen to the same twenty minutes five times rather than five different things once. Say sentences out loud before you feel entitled to. And re-meet the words you half-have — not to memorize them harder, but to see past the first translation into the range a word actually covers, across enough varied situations that it stops needing to be retrieved and starts simply arriving. That is how you expand your vocabulary without adding new words.
Then keep doing it. That is the second half of the sentence, and it is the half the woman in Annapolis paid for. Language lives in encounters and it stays alive for as long as the encounters continue. Her father stopped, and the Russian went. My deliveries did not stop for a year, and the English stayed.
I still think about that afternoon. I walked into that room afraid I would not be able to hold a conversation, and I walked out having been complimented on exactly that, by a woman who had lost a language I still have. What I have was not built by the years I spent being taught. It was built by the years I spent using. We arrived where we each arrived on the same mechanism, running in opposite directions.
That is what the app is for. Not to teach you about words. To keep introducing you to them until they are yours.