Fifth January in a row, the same line sits at the top of the list: finally learn Spanish. The app is installed. The textbook is bought. A tutor was found, then quietly dropped. And on holiday you still point at the menu.
You can learn a language with an ordinary, unremarkable memory. Two decades of research agree on this: speed depends far less on memory capacity than on three things - which words you learn, how often you come back to them, and whether you are willing to speak badly. Memory is not the main variable on that list. Below is what specific studies show about each point, including the ones that contradict the most popular advice. One number that language coaches love to repeat turns out to be roughly three times too small. It comes up in the second section.
"My memory is terrible" - why that is not the problem
A bad memory is almost never the real reason a language fails to stick. The real reasons are usually different: people learn the wrong words, study rarely and in big blocks, and leave speaking for last.
Scale makes this easy to see. Ethnologue, the reference catalogue of the world's languages, counts more than 7,000 living languages in its 28th edition (2025). A large share of humanity switches between several of them every day, and those people are not memory prodigies. They are ordinary residents of ordinary countries.
One honest correction is needed here, because interviews endlessly repeat that "60% of the planet is bilingual" (bilingual means using two languages comfortably in daily life). No precise global figure for bilingualism exists. Psycholinguist François Grosjean, the source of the most quoted estimate, explains why: national censuses rarely ask about a second language, and when they do, they define it differently. Colin Baker, whose encyclopedia produced the 65% number, put it plainly: "I suppose the real answer is: we do not know."
Where people were surveyed carefully, we do have numbers. The 2024 Eurobarometer (the European Commission's regular public survey) interviewed more than 26,500 people across the 27 EU countries. Around 60% of Europeans can hold a conversation in a foreign language, rising to almost 80% among 15 to 24-year-olds. About half of Europeans speak English.
And a counterexample. India is usually held up as a nation of natural polyglots, but the 2011 census recorded 26% bilingual speakers, about 315 million people. A lot, but far from everyone.
What follows from this for you? Multilingualism runs on circumstances, not on memory: another language is nearby, and you have to use it for something every day. When the circumstances do not exist, you build them by hand. Where to start is next, because the very first step is usually the one people get wrong.
How many words do you actually need to speak
Roughly a thousand of the most frequent words will carry you through most everyday speech. That is measurable. The number usually quoted, however, is not.
In 2022 linguist Hung Tan Ha analysed a corpus of about 625 million words - subtitles from films, TV programmes and soap operas - and measured how much each additional thousand covers. He counted in "word families" (a word plus all its forms and obvious derivatives: go, goes, went, going are one family).
- The 1,000 most frequent word families cover about 83-85% of everyday speech
- 2,000 families cover about 86-89%
- 3,000 families cover about 87-90%
- Comfortable 95% coverage takes 2,000-3,000 families; 98% takes 4,000-5,000
So here is that promised number. Courses and interviews often claim that "350-400 words cover 90% of spoken language." The data does not support it. Ninety percent sits somewhere around a thousand word families and up, which is roughly three times more material. Three hundred words are not 90% of speech. They are your first confident sentences, which is still a real milestone.
But the idea behind the wrong number is right, and it matters more than the number itself. Words pay out very unevenly. The first thousand hands you 84% of speech; the next thousand adds four percentage points. It works like a keyring: one key opens the front door you use daily, the rest open a basement you visit once a year.
The practical takeaway: learning words in textbook order, or alphabetically, is the worst available option. Find a frequency list for your language (open lists of the first 1,000-2,000 words exist for almost every major language) and work down it. And learn short phrases rather than bare words. A word on a list gets forgotten; a word inside a sentence stays.
Good news: a thousand words is manageable even with a poor memory. Bad news: only if you spread it correctly across time. That is where most people lose a year.
Why five minutes a day beats two hours on Sunday
The same number of hours, broken into short frequent sessions, sticks noticeably better than the same hours in one block. This is one of the most solidly established effects in the psychology of memory.
In 2006 a team of psychologists led by Nicholas Cepeda pooled 184 research articles, 317 experiments and 839 separate measurements. Their work was a meta-analysis - a study that combines the results of many earlier studies and calculates the overall outcome. The finding: study sessions spread out over time produce better retention than the same sessions packed together. And the gap between sessions should grow along with how long you need to remember the material.
In plain terms: if you need those words in a month, reviewing them five times in one evening is close to useless. You need the same five reviews stretched out - today, tomorrow, in three days, in a week, in a month.
Think of watering a plant. Pour a week's worth of water at once and most of it runs past the roots; six days later the soil is dry again. The same bucket, spread across the week, gives a completely different plant.
What this means for a busy schedule:
- Ten to fifteen minutes daily beats ninety minutes once a week. Not because less is better than more, but because seven contacts a week beat one.
- A missed day cancels nothing. A missed month cancels everything.
- Return to old words instead of only adding new ones. It is the return, not the first encounter, that moves a word into long-term memory.
- Flashcard apps are useful for exactly this reason: they remember when your review is due. You cannot learn a language from them, but you can avoid forgetting what you learned.
So you know what to learn and how often. Yet plenty of people who get both right still stay silent. That problem sits in the throat, not the head.
What blocks you harder than a bad memory
Fear of making mistakes hurts your results more than memory ever will. That is not a figure of speech: the link between anxiety and language performance has been measured in hundreds of studies.
In 2019 Yasser Teimouri and colleagues pooled 97 studies and found a relationship between foreign language anxiety and achievement of -0.36. In 2025 Wei Chen repeated the exercise across 27 studies and 3,633 learners of Chinese and got -0.33. The minus sign means the higher the anxiety, the lower the result. The number itself is a correlation coefficient, where 0 means "no relationship at all," 1 means "a perfect one," and 0.33 means "clear and consistent."
An honest caveat: correlation does not prove cause. Anxiety may damage performance, or weak performance may generate anxiety. Most likely both, in a loop.
You have seen the scene. Someone attends classes for two years, knows the tenses better than the teacher, and answers a waiter abroad with a nod. Next to them stands a colleague with twenty words who is already sorting things out using hands, face and those twenty words. Six months later the second one speaks. The first one still nods.
The most damaging sentence a beginner can hear is "if you cannot say it correctly, say nothing." No child ever learned a first language under that rule. A two-year-old who says "I breaked the cup" does not get a failing grade, they get praise. Adults are not handed that allowance, so they have to write it for themselves.
What to do about it:
- Start sentences with the word "I" in your target language. Your brain finds it easier to continue a started sentence than to launch from nothing.
- Allow yourself the wrong word order and one verb tense instead of twelve. The person opposite you will understand.
- Change the goal. Not "pass," but "get understood." Nobody is grading you any more.
- If study has felt joyless for several weeks running, take a break instead of pushing. Material forced in under pressure does not stay.
Series with subtitles: learning or self-deception?
Watching films and shows in your target language does help, less than people assume, and the size of the effect depends heavily on the subtitles.
A meta-analysis by David Sutton and Stuart Webb in Studies in Second Language Acquisition (2026) pooled 56 experiments and 1,954 participants. Plain viewing with no on-screen text produced a measurable gain in vocabulary and listening (effect size g = 0.89), though the authors themselves describe the gains as modest. One more detail from the same review: single sessions and multiple sessions produced statistically indistinguishable results. Bingeing more episodes is not the lever.
What clearly does strengthen the effect is text on screen. Earlier meta-analyses by Montero Perez and by Kurokawa and colleagues found captioned video outperforms uncaptioned video. The subtlety is in which captions you switch on.
- Audio and subtitles in the same language. Then ear and eye receive the same thing, and your brain stitches sound to spelling. Native-language subtitles over foreign audio create two competing streams, and attention always drifts to the easier one.
- Rewatch what you already know. A familiar plot removes the load: you are not spending energy on "who are these people," so all of it goes to the speech.
- Take one line per episode. A single sentence written down and said out loud is worth more than forty sentences understood in passing.
Pronunciation, by the way, is trained with the mouth rather than the eyes: repeating lines after a character and copying their intonation works. We looked at that technique in detail in our piece on dubbing in your own voice.
Is 30, 40 or 50 too late to start?
Starting is not too late at any age. It is only too late for one specific goal: sounding exactly like a native speaker, with flawless grammar.
The largest study on this question came from Joshua Hartshorne, Joshua Tenenbaum and Steven Pinker (Cognition, 2018). They gathered data from 669,498 people, native and non-native English speakers, and built a model that separates a person's current age from the age they started and from years of practice.
The result: grammar-learning ability stays near its childhood level until roughly 17.4 years old, then declines. It declines gradually, it does not slam shut. There is no door closing at thirty anywhere in that data. The authors also note that reaching your own ceiling takes around 30 years for natives and non-natives alike, with most of the growth happening in the first 10 to 20.
Adults also have an advantage that rarely gets mentioned. Adults know how to learn: they understand what a rule is, they can take a structure apart, they know why they are doing it. In the first months, adults usually move faster than children. Children win on the long run and on accent, not at the start.
So if you are forty, drop the goal of sounding like a Londoner and keep the goal of speaking and understanding freely. That one is fully reachable, and it pays in more than satisfaction.
What a second language actually pays: money and brain
A second language does raise your pay and, according to part of the evidence, does delay dementia symptoms. Both numbers are heavily inflated in popular retellings. Here are the real ones.
Money
Economist Jacek Liwinski studied a sample of about 35,000 people in Poland (Empirica, 2019). Advanced command of a foreign language was worth roughly 11% in wages. By language, the spread was wide:
- Spanish - 32%
- French - 22%
- Italian - 15%
- German - 12%
- English - 11%
Notice the pattern: the scarcer a language is on the local market, the more it pays. English pays least precisely because so many people already have it.
For comparison, in the United States Albert Saiz and Elena Zoido found a premium of only 2-3% among college graduates. The reason is simple: in an English-speaking country a second language is a nice extra, while in a country with a local language it is part of the job. So the popular formula "a language adds 15-30% to your income worldwide" does not hold. The accurate version: there is a premium almost everywhere, and its size depends on your country and on how scarce your language is.
Brain
This is where popular retellings drift furthest from the evidence. Interviews throw around "languages push Alzheimer's back by 10 to 20 years." A systematic review by Maurits van den Noort and colleagues (Behavioral Sciences, 2019) pooled 34 studies covering 2,794 bilingual and 4,207 monolingual people with dementia. The results:
- Symptom onset was delayed by 4 to 5.5 years in bilinguals, not 10 to 20.
- About 53% of studies supported a protective effect, another 12% partly supported it, and 35.3% found no effect at all.
- Retrospective studies (which look back through old medical records) show the effect; prospective ones (which follow people forward in time) do not. That pattern is a classic warning that part of the effect may be an artefact of method.
- The authors' own conclusion: no firm conclusions can be drawn today.
How to hold this sensibly. Language study is not medicine and not insurance against dementia. It is a cheap, pleasant, harmless load on the brain where the worst case is no effect and the best case is something like four extra years of a clear head. You take that bet without needing a guarantee.
If AI translates for you, why learn at all?
Machine translation solves "understand this text" very well and "be understood by this person" very badly. The gap between those two is the answer.
A telling case came with the Korean series Squid Game in 2021. Bilingual viewers said the English version lost almost everything the original had. The story spread online as "AI translated a show and ruined it." That is not accurate. As ABC News reported, the complaints were not about machine translation at all: they targeted the closed captions of the dubbed version, while the regular English subtitles stayed close to the Korean. And the deeper cause was not technology. Translation lecturer Jinhyun Cho explained that Korean has a system of address encoding a speaker's age and social standing, and English simply has no tools for it. Neither a human nor a machine can translate something a language does not contain.
The practical conclusion follows. The translator in your pocket handles menus, train stations and washing-machine manuals. It does not handle a joke, a hint, polite distance, or the moment a stranger decides to speak to you informally. Speaking someone's language does something to a relationship that no perfect translation reproduces.
There is a second, less obvious reason. The more we hand over to external tools, the less we hold ourselves: phone numbers, routes, dates. We looked at what the research says about that in is AI making us dumber. A language is one of the few skills that works with no battery and no signal.
Your first week
Everything above reduces to seven actions. None of them requires a good memory.
- Name your reason. Not "self-improvement," but something concrete: understand song lyrics unaided, talk to a colleague in another office, read a hobby forum. Without an answer here, no method works.
- Download a frequency list of the first thousand words in your language. That is your route, instead of a textbook.
- Cover the basic structure in one evening: word order, pronouns (I, you, he, my, your) and three forms of the main verb - present, past, future. All of it fits on one sheet of paper.
- Book 10-15 minutes a day and attach it to an existing habit: after coffee, on the commute, before bed.
- Learn phrases, not words. Not "to be late," but "sorry, I am late."
- Speak from week one - out loud, to yourself if there is nobody else. Start your sentences with "I."
- Add background: a playlist, a podcast, a familiar series with same-language subtitles. Not as study, but so the language stops sounding foreign.
After a week you will not have a language. You will have something more useful: proof that seven days in a row is possible. That proof is exactly what the January line has been missing for five years.
Today it takes two minutes. Open the notes app on your phone and write down one sentence in your target language that you would like to say out loud to a real person within the next month. One sentence. Your first thousand words start there.
Sources
- Ethnologue: Languages of the World, 28th edition (2025), via Our World in Data
- François Grosjean. Chasing Down Those 65% (2014)
- Eurobarometer 2024: Europeans and their languages (European Commission survey, 26,500 respondents)
- Census of India 2011 data on bilingualism
- Ha H.T. Vocabulary Demands of Informal Spoken English Revisited. Frontiers in Psychology (2022)
- Cepeda N.J. et al. Distributed Practice in Verbal Recall Tasks. Psychological Bulletin (2006)
- Chen W. Foreign language anxiety and academic achievement: a systematic review and meta-analysis. Frontiers in Education (2025)
- Sutton D., Webb S. The effects of audiovisual input on second language learning. Studies in Second Language Acquisition (2026)
- Hartshorne J.K., Tenenbaum J.B., Pinker S. A critical period for second language acquisition. Cognition (2018)
- Van den Noort M. et al. Bilingualism, Cognitive Decline, and the Onset of Dementia: a systematic review. Behavioral Sciences (2019)
- Liwinski J. The wage premium from foreign language skills. Empirica (2019)
- ABC News. Squid Game subtitles and dubbing prompt online debate over Netflix's translation (2021)
