Daniel sent us this one — he's been watching people around him in Israel learning Hebrew, immigrants of all ages from all over the world, and he's noticing something. Some people seem to absorb the language effortlessly, others struggle for years. He's asking whether the old truisms hold up under scrutiny — "I'm too old," "I'm just not good with languages," "math people can't do languages." And he wants to know what actually accounts for the wide diversity in outcomes, what you'd say to someone who's resigned to defeat, and whether AI tools could help with the embarrassment factor that holds so many people back.
That last part is where I want to land, but let's work our way there. And by the way, DeepSeek V four Pro is writing our script today — which feels appropriate for a conversation about language models.
A language model writing a script about language learning. So where do we start — the age thing? Because that's the one I hear constantly.
It's the one everyone defaults to, and it's not entirely wrong, but it's been flattened into something almost useless. The critical period hypothesis — the idea that there's a window for native-like language acquisition that closes somewhere around puberty — that's real. Eric Lenneberg proposed it in nineteen sixty-seven, and the broad contours have held up. But here's what gets lost: the hypothesis is specifically about achieving native-like phonological and grammatical competence. It's not about whether you can become fluent, functional, or even highly proficient.
The distinction is between sounding indistinguishable from a native speaker and actually being able to use the language effectively.
And most people, when they say "I'm too old to learn a language," they're collapsing those two things. They're not actually trying to pass as a native. They just want to have conversations, read the newspaper, navigate daily life. And there's essentially no age cutoff for that.
I've seen studies on older adults learning vocabulary — they sometimes outperform younger learners on explicit memory tasks. Give them word lists and they'll drill them methodically.
There was a study out of the University of Haifa a few years back — researchers looked at Russian immigrants to Israel across different age cohorts. The older learners, those over sixty, had slower acquisition rates in the first year, but by year three the gap between them and the twenty-somethings had narrowed substantially on comprehension measures. Where they continued to lag was pronunciation and grammatical gender accuracy — which tracks with the critical period. But functional communication?
That's encouraging. But it also suggests that if you're fifty-five and you've been at it for six months and still feel lost, your benchmark might be off. You're comparing your six months to a twenty-five-year-old's six months, when the trajectory is just different.
Right — and that's where the defeatism creeps in. People hit the six-month mark, they still can't follow a fast conversation, and they conclude, "See, I'm too old." When really what they're experiencing is a different slope, not a flat ceiling.
The message to someone feeling defeated isn't "try harder" — it's "adjust your expectations about the timeline.
Maybe also adjust your method. Because that's the other half of Daniel's question — the wide diversity in outcomes. Some of it is individual aptitude, sure. But a lot of it is mismatch between learner and approach.
Let's talk about aptitude, then. Daniel mentioned his family member, the retired math professor, and the stereotype that highly numerate people struggle with languages. Is that a real thing?
It's complicated. There's no evidence that mathematical ability directly inhibits language acquisition. If anything, pattern recognition — which is central to mathematics — should help with grammar. But here's where it gets interesting: people who are very strong in systematic, rule-based thinking sometimes approach language like it's a formal system to be solved. They want explicit rules for everything, and when they encounter irregular verbs, idiomatic expressions, or the sheer messiness of natural language, they get frustrated.
Whereas someone who's comfortable with ambiguity just absorbs patterns through exposure and doesn't need to understand why the exception exists.
And that tolerance for ambiguity is one of the strongest predictors of language learning success. It's not about being "good at languages" — it's about being willing to not understand everything and keep going anyway.
Which brings us to personality. I'd argue embarrassment is the single biggest barrier for adults. Kids don't care if they sound silly. Adults are terrified of it.
That's where Daniel's point about AI is really sharp. He's not talking about using AI because he's lazy — he's talking about using it as a low-stakes practice environment. A place to build confidence before you go out and make mistakes in front of actual humans.
The embarrassment factor is real, and it's not trivial. I've watched people freeze mid-sentence because they couldn't remember a word, and you can see the shame wash over them. That moment can set someone back weeks.
There's research on this — affective filter hypothesis, Stephen Krashen in the nineteen eighties. When anxiety is high, the brain's language processing effectively shuts down. You're not learning in that moment, you're just surviving. So if AI conversation partners can lower that filter, they're not a shortcut — they're an accelerator.
The AI tools now are genuinely good at this. You can have a conversation with Claude or ChatGPT in your target language, and it'll correct your grammar gently, suggest more natural phrasing, and — crucially — it doesn't judge you. You can make the same mistake fifty times and it'll correct you on the fifty-first without a hint of impatience.
The voice mode on these models has gotten remarkably fluid. You can have a spoken conversation, stumble through a sentence, and it'll wait. No native speaker has that kind of patience after the first five minutes. And the AI can adjust its level — speak more slowly, use simpler vocabulary, then gradually ramp up as you improve.
That's the other thing — adaptive difficulty. A human conversation partner might not know how to meet you at your level. They either speak to you like a child or they speak at full speed and you drown.
Let me flesh out the diversity question more concretely. What accounts for the wide range in outcomes? I'd say there are at least five distinct factors. First, age of arrival — the critical period effect is real for phonology and some syntactic structures. Second, aptitude — and aptitude isn't one thing. It breaks down into phonetic coding ability, grammatical sensitivity, rote memory, and inductive language learning ability.
That's the work of John Carroll and Stanley Sapon, right? The Modern Language Aptitude Test?
Yes, developed in the nineteen fifties and still used. And what's fascinating is that someone can be high on phonetic coding — they can hear and reproduce sounds accurately — but low on rote memory. Or vice versa. So you get the person who has a beautiful accent but can't remember vocabulary, and the person who has a huge vocabulary but a thick accent.
That explains Daniel's observation about people struggling with one aspect and not another. It's not a single "language ability" — it's a profile of sub-skills.
And the third factor is the learner's first language. If your native language shares features with the target language, acquisition is faster. A native Arabic speaker learning Hebrew has a structural head start over a native English speaker, because both are Semitic languages with similar root systems and grammatical patterns.
That's something people don't always account for when they compare themselves to others. "My friend picked it up in a year, why haven't I?" Well, your friend's native language might be much closer.
Fourth factor: motivation and identity. Integrative motivation — wanting to belong to the culture — tends to produce better long-term outcomes than instrumental motivation, which is just wanting to pass a test or get a job. Though both work in the short term.
I think embarrassment ties into identity. If you're worried that speaking imperfectly makes you look less intelligent or less competent, you're not just learning vocabulary — you're managing a threat to your self-image.
That's profound, actually. For someone like Daniel's mother, who's convinced she can't learn even basic phrases — that conviction isn't just a belief about language aptitude. It might be a protective mechanism. If you never try, you never confirm the fear that you'll sound foolish.
Part of the answer to "what do you say to someone who's resigned to defeat" is to lower the stakes. You don't need to become fluent. You don't need to sound native. You just need to learn five phrases and use them. Success at that level can break the cycle.
The fifth factor — the one people overlook most — is sheer exposure time. Foreign Service Institute estimates it takes about eleven hundred class hours for an English speaker to reach professional working proficiency in a Category Four language like Hebrew. That's almost thirty weeks of full-time study. Most adult learners are doing an hour a week on Duolingo and wondering why they're not fluent.
Right — it's not that they're bad at languages. It's that they're dramatically underestimating the time investment required. And then they blame themselves.
There's a researcher named Francois Grosjean who's done great work on bilingualism. He makes the point that we hold adult learners to an impossible standard — we compare them to native speakers and call anything short of that a failure. But a native speaker has had tens of thousands of hours of exposure. An adult learner who can function professionally after a thousand hours has achieved something remarkable.
The truism "I'm too old" — when we actually test it — is mostly false for functional proficiency. The truism "I'm just not good with languages" is probably a mix of mismatched method, insufficient exposure, and an overly narrow definition of success.
The truism "math people can't do languages" is more about learning style than ability. The math professor who struggles might thrive with a grammar-heavy, rule-based approach that a more intuitive learner would hate. The problem isn't that he can't learn — it's that the immersion method everyone recommends might be exactly wrong for how his brain works.
That's an important corrective. The language learning industry has been dominated by the communicative approach for decades — lots of conversation, minimal explicit grammar instruction. And for many people that's great. But for someone with a highly analytical mind, it can feel like being thrown into the deep end without being taught how to swim.
I've seen this in medical contexts. Some medical students learn best by shadowing and absorbing, others need to understand the underlying physiology before anything clicks. Neither is wrong — but if you force the second type into the first type's learning environment, they'll fail and conclude they're not cut out for it.
Let's get concrete. If someone's listening right now and they're in that defeated place — they've tried, they've stalled, they've concluded they just don't have the gift — what do you tell them?
First, I'd say: the fact that you're frustrated means you care, and that's actually a good sign. Apathy is the real enemy. Second, I'd ask them to diagnose where they're actually stuck. Is it vocabulary? Most people say "I can't speak the language" but that's too vague to act on.
Break it down.
Break it down. If it's vocabulary, that's the most fixable problem — spaced repetition systems like Anki are effective, and they work for almost everyone regardless of aptitude. If it's listening comprehension, you probably need more hours of input at a level you can mostly understand — what Krashen called "comprehensible input." If it's speaking, the issue might be anxiety, not ability — and that's where AI conversation practice could be transformative.
If it's pronunciation, that's the one where age really does play a role. But even there, you can improve. You might never sound native, but you can sound clear and comprehensible, which is what actually matters for communication.
There was a study from the University of Illinois that looked at late learners of English — adults who arrived in the U.after age twenty. After five years of residence, about twenty-five percent of them scored within the native speaker range on a grammaticality judgment test. Not the majority, but a substantial minority. The ceiling isn't universal.
Even on the dimension where the critical period is strongest, there's individual variation. Some people do get there.
We don't fully understand why. It could be neurological — some people's brains retain more plasticity. It could be behavioral — they sought out more input, more correction, more practice. It could be genetic. But the point is, the blanket statement "you can't achieve native-like proficiency after childhood" is statistically true for most people but not universally true.
Which means if you're fifty and starting Hebrew, you shouldn't expect to sound native. But you also shouldn't assume it's impossible to become highly functional. The range of possible outcomes is wider than the truisms suggest.
Let me add one more piece on motivation, because it connects to Daniel's situation. He's in Israel — he's immersed. And immersion is powerful, but it's also exhausting. You're using a language you're still learning for every transaction, every bureaucratic interaction, every social event. That cognitive load is real, and it can lead to burnout.
I've experienced that. After a full day of functioning in a second language, your brain feels like it's been through a workout. You're tired in a way that's hard to describe.
That's called cognitive fatigue, and it's well-documented in bilinguals and second language learners. The prefrontal cortex is working overtime to suppress the first language and manage the second. Over time, that gets easier as automaticity develops, but in the early stages, it's draining.
Which is another reason people give up — not because they can't learn, but because it's exhausting and they don't see the progress to justify the exhaustion.
The advice has to include: pace yourself. You're not lazy for needing breaks. You're not failing because you're tired. The brain is literally building new neural pathways, and that takes energy.
Let's pivot to the AI piece more directly. Daniel mentioned vocabulary acquisition specifically — using AI to build confidence. How would that work in practice?
Imagine you're learning Hebrew and you need to go to the bank to open an account. You know the vocabulary — cheshbon, p'kad, ribit — but you're terrified of freezing up when the teller asks you something unexpected. An AI conversation partner could simulate that exact scenario. You practice it five times, ten times, with variations. By the time you walk into the actual bank, the script is so practiced that your anxiety drops significantly.
The AI can throw curveballs. The teller asks about something you didn't prepare for. You learn to handle the unexpected in a safe environment.
That's the key — safe failure. You need to fail to learn, but failing in front of a real person has social consequences. Failing in front of an AI has none. So you can push your boundaries further, take more risks, and learn faster.
There's also the correction aspect. A human conversation partner, especially a stranger, will often let your mistakes slide to avoid awkwardness. They understood what you meant, so they don't correct you. The AI, if instructed properly, will give you consistent, gentle feedback every time.
You can customize it. You can say, "Please correct every grammar mistake but ignore pronunciation for now," or "Focus on my use of prepositions." No human tutor can track that precisely in real time.
Daniel mentioned embarrassment specifically — that he loves the idea of using AI to build confidence before speaking with native speakers. I think that's the killer app for language learning AI. Not replacing human interaction, but scaffolding it.
Scaffolding is exactly the right word. The AI is the training wheels. You use them until you don't need them, and then you take them off. But for a lot of adults, they never had the training wheels — they were just thrown into conversations and expected to pedal.
The training wheels analogy works on another level. Kids learning their first language have years of silent listening before they start producing. They're absorbing patterns, building a mental model. Adults rarely give themselves that silent period. They feel pressure to speak from day one.
AI could provide that low-pressure input phase. You can listen, read, interact at your own pace without the social demand to produce before you're ready. Then when you do start speaking, you've got a foundation.
Are there any downsides to the AI approach? I'm thinking about over-reliance — someone who gets so comfortable with the AI that they never make the leap to human conversation.
That's a legitimate concern. The AI, for all its sophistication, doesn't replicate the unpredictability of human interaction. It doesn't interrupt, it doesn't speak over you, it doesn't use slang that hasn't appeared in its training data, it doesn't have an accent you weren't expecting. So if you only practice with AI, you might get a false sense of readiness.
The solution is probably to treat AI as a stepping stone, not a destination. Use it to get to a baseline of confidence, then deliberately seek out human interaction.
There's a middle ground too — AI-assisted human interaction. Imagine having an earpiece that gives you real-time vocabulary suggestions during a conversation. Not translating everything, just surfacing the word you're reaching for. That technology exists now, and it's getting better.
That feels like it could either reduce anxiety or increase it. Having an AI whisper in your ear while you're trying to listen to someone else sounds cognitively overwhelming.
Probably depends on the person. Some people would find it reassuring, others would find it distracting. But the point is, we're moving toward a world where the boundaries between "learning a language" and "being assisted in a language" are blurring.
Which raises a philosophical question. If you're using AI assistance to function in a language, have you actually learned it? Or are you just operating a system that makes you functional?
I'd say it's a spectrum, not a binary. If you're using AI to translate every sentence in real time, you haven't learned the language — you've learned to use a tool. But if you're using AI to supplement gaps in your knowledge — the word you can't remember, the grammatical construction you're unsure about — that's not fundamentally different from asking a human "how do you say X?" It's just more efficient.
I think the embarrassment factor Daniel mentioned is key here. If the AI assistance reduces the anxiety enough that you actually engage in more conversations, you're getting more practice, and over time you'll need the assistance less. It's a virtuous cycle.
The opposite of the vicious cycle a lot of learners are in — they're anxious, so they avoid speaking, so they don't improve, so they become more anxious.
Let's talk about Daniel's mother for a moment, because I think that's a really common profile. Someone who's convinced at a deep level that they simply cannot learn even basic phrases. What do you do with that?
You start with something so small that it's almost impossible to fail. Not "learn Hebrew" — that's overwhelming. Not even "learn ten phrases." Start with one. " That's it. You already know it. Now use it. Use it every time you enter a shop. Feel what it's like to say it and be understood.
Then add "todah.Now you've got a complete interaction — greeting and thanks. That's a success. You've just communicated in Hebrew.
The psychological shift from "I can't learn languages" to "I just communicated in Hebrew" is enormous, even if the actual linguistic content is tiny. It breaks the identity of being a non-learner.
The identity piece is underrated. Once you see yourself as someone who can learn a little, the door is open to learning a little more. The barrier wasn't aptitude — it was self-conception.
There's a concept in psychology called self-efficacy — your belief in your ability to succeed in a specific situation. Language learning self-efficacy is one of the strongest predictors of actual outcomes. And self-efficacy isn't fixed. It's built through small successes.
The advice to someone like Daniel's mother isn't "you're wrong about yourself, you actually can learn." It's "let's find something so small you can't fail at it, and then build from there.
Crucially, don't frame it as "learning a language." Frame it as "learning to say hello and thank you." That's a finite, achievable goal. Once she's done that, she's not a language learner — she's just a person who knows how to be polite in Hebrew. That's a much less threatening identity.
I want to circle back to something you mentioned earlier — the different sub-skills of language aptitude. Because I think this is where a lot of the "I'm not good at languages" self-diagnosis comes from. Someone struggles with pronunciation and concludes they're bad at languages, when they might be excellent at vocabulary acquisition.
The education system often reinforces this by grading language ability as a single thing. You get one grade in French class, and it collapses pronunciation, grammar, vocabulary, reading, writing, and speaking into a single number. If you're weak in one area, your overall grade suffers, and you internalize "I'm bad at French.
When really you might be great at reading French literature and just have a terrible accent.
And for most real-world purposes, the terrible accent matters much less than being able to read and understand. But in a classroom setting, the accent gets graded, so it feels equally important.
One practical takeaway for someone feeling defeated is: identify your strengths. What aspect of the language actually comes more easily to you? Lean into that. Build confidence there. Then work on the weaker areas from a position of some success.
Match your method to your strength. If you're a visual learner, use written materials heavily. If you're an auditory learner, podcasts and conversations. If you're analytical, grammar books. The "best" method is the one that aligns with your cognitive profile, not the one that's currently trendy.
There's an irony here. The language learning world has spent decades arguing about which method is best — grammar translation versus audio lingual versus communicative versus task-based. And the research pretty consistently shows that no single method is best for everyone. But the industry keeps trying to sell one-size-fits-all solutions.
Because it's hard to market "it depends on your individual cognitive profile, motivation, first language, learning environment, and goals." That doesn't fit in an app store description.
People buy the app, it doesn't work for them, and they conclude they're the problem. When really the app was designed for a different kind of learner.
Now with AI, we actually have the capability to personalize at scale. An AI tutor could diagnose your learning style, adapt its approach, focus on your weak points, and do it all without the social anxiety of a human tutor. That's not science fiction — it's being built right now.
That's the optimistic take. The pessimistic take is that AI language tools will get so good that people stop bothering to learn languages at all. Why spend a thousand hours learning Hebrew when your earpiece translates perfectly in real time?
I've thought about this. And I think the answer is that language isn't just about information transfer. When you speak someone's language, even imperfectly, you're signaling something. You're saying, "I'm making an effort to meet you in your world." That's a social signal that no translation device can replicate.
There's a difference between "your device translated what I said" and "you learned my language.
A profound difference. And I think people intuitively understand that. The technology will keep improving, but the human desire to connect directly, without mediation, isn't going away.
The AI isn't a replacement for language learning — it's a tool that makes language learning more accessible to people who would otherwise be too anxious or too time-poor to attempt it.
That's my position. And it circles back to Daniel's point about embarrassment. If AI can lower the barrier enough that more people actually start learning, that's a net gain for human connection, not a loss.
Let's talk about one more thing Daniel raised — the pronunciation versus vocabulary split. Why do some people have a great ear for sounds but terrible retention for words, or vice versa?
Phonetic coding ability — the capacity to hear and reproduce unfamiliar sounds — seems to be partly neurological and partly related to early exposure. If you grew up hearing multiple languages, even if you didn't learn them, your auditory system got trained on a wider range of phonemes. That gives you an advantage later.
Someone who grew up monolingual in a linguistically homogeneous environment is at a real disadvantage for pronunciation.
Yes, and it's not their fault. Their brain literally pruned away the ability to distinguish sounds that weren't present in their linguistic environment during infancy. By the time they're six months old, they've already started specializing.
Which explains why some people seem to "have an ear" and others don't. It's not a gift — it's early exposure.
Or lack of early pruning. But here's the encouraging part: you can retrain this to some degree. High-variability phonetic training — where you listen to many different speakers producing the same sound — has been shown to improve adult learners' ability to distinguish and produce unfamiliar phonemes. It takes work, but it's not hopeless.
On the vocabulary side?
Vocabulary acquisition is heavily dependent on memory strategies and exposure frequency. Some people naturally use effective strategies — they create associations, they use the word in sentences, they connect it to words they already know. Others just try to memorize by rote repetition. The strategy matters more than some innate "memory for words.
If you're struggling with vocabulary, it might not be your memory — it might be your technique.
Spaced repetition, active recall, using words in context, creating vivid mental images — these are learnable skills. The person with a "good memory for words" is often just someone who stumbled onto effective strategies without realizing it.
Which brings us back to the core message: most of what people attribute to fixed ability is actually a combination of strategy, exposure, and mindset. The truisms are mostly wrong.
Not entirely wrong — the critical period is real for certain aspects, and there are genuine individual differences in aptitude. But the range of what's achievable for the average person is vastly wider than the truisms suggest.
If I'm talking to someone who's convinced they can't learn a language, I'd say: your belief is probably the biggest thing holding you back, not your age, not your brain, not some missing language gene. Try a different approach. Try a smaller goal. Try an AI conversation partner. But don't conclude the ceiling is low based on a failed attempt with one method.
I'd add: be patient with yourself. Language learning is a marathon, not a sprint, and the early miles are the hardest. The fact that you're struggling doesn't mean you're failing — it means you're in the hard part.
Now: Hilbert's daily fun fact.
Hilbert: The Pacific barreleye fish has a transparent head filled with fluid, and its tubular eyes can rotate inside the dome to look upward through the top of its own skull.
...right.
Here's an open question to sit with: if we build AI language tools that are adaptive, patient, effective at lowering the affective filter — what happens to the language learning industry? Do the apps that treat all learners as interchangeable survive? Or do we finally get the personalized approach that the research has been pointing toward for decades?
I suspect we're heading toward a world where the default is AI-mediated language practice, and human instruction becomes a premium option for those who want cultural nuance and authentic interaction. But I also suspect that more people will actually learn languages, because the barrier to entry will be so much lower.
That's a hopeful note to end on. Thanks to our producer Hilbert Flumingtop. This has been My Weird Prompts. Find us at myweirdprompts.See you next time.
