The age 17.4 finding does not mean what you were told

In 2018 a grammar quiz went viral on Facebook, and 669,498 people took it. Hartshorne, Tenenbaum and Pinker used the results to model how grammar-learning ability changes with age, in the journal Cognition. The model found a rate change beginning at 17.4 years. That number is where the modern "you missed the window" comes from, and it got turned into something the paper never said.

The window closed and you missed it

Every person took the quiz once. That is a cross-sectional design: nobody was followed over time, so 17.4 is an inflection point fitted to a single snapshot, not a cliff anyone watched a learner fall off. The authors say plainly that measuring how good people end up "cannot by itself reveal changes in underlying learning ability."

The finding concerns the rate of learning, not a ceiling on it. The same paper reports that its native speakers did not reach their own ceiling until around age 30, and that non-native immersion learners who started in their late twenties eventually scored above the youngest native speakers in the sample. Grammar kept improving for decades after the supposed window shut. These studies also measure whether someone ends up indistinguishable from a native speaker on a written grammar test. Dinner in Lisbon is a lower bar: a few hundred hours, against the thousands behind a native score.

Accent is the genuine age effect. Dollmann, Kogan and Weißmann rated the German of 1,843 immigrants in Applied Linguistics in 2020, and found a critical period up to around age 10, after which speaking without a detectable foreign accent becomes less likely. Among people who arrived later, contact with native speakers still predicted a weaker accent: using German with family and friends carried a coefficient of about -0.92 on accent strength for arrivals after 16, and each extra month living with a German partner about -0.06 for arrivals between 10 and 16. If you start Spanish at 45, you will probably always sound foreign in it. Being understood, being fluent and sounding like a local are three separate things, only the third has a hard age effect, and talking to people is the part you control.

Children learn languages faster than adults

Watch a 6-year-old move to a new country and pick up the language while the parents struggle, and this looks obviously true. The controlled comparisons say the opposite about speed. Snow and Hoefnagel-Höhle followed English speakers who had just moved to the Netherlands and were learning Dutch from daily life, in Child Development in 1978. They tested every person three times across that first year on pronunciation, vocabulary, sentence repetition and grammar rules, in age bands from 3 to 5 up to adults. The 12-to-15-year-olds and the adults made the fastest progress in the early months, and at the end of the year the 8-to-10s and 12-to-15s had the best Dutch, with the 3-to-5-year-olds lowest on every test. One caution: the groups were small, a handful each.

Marinova-Todd, Marshall and Snow put that study alongside the rest of the literature in TESOL Quarterly in 2000, in "Three misconceptions about age and L2 learning," and their first named misconception is exactly this belief. Their summary: older learners are generally faster and more efficient in the initial stages of learning a second language. So a 40-year-old learns grammar more slowly than a 14-year-old and gets there anyway, with more hours, counted in hundreds rather than years.

Adults have advantages inside those hours: you can be told a rule and apply it the same afternoon, where a 5-year-old reconstructs the grammar from raw exposure. What children have is hours and stakes, and an adult gets whatever minutes are left after work.

Your memory is worse now, so vocabulary will not stick

Adults who say their memory has gone are comparing today against school, and school was a machine for spaced practice: the same material daily for months, tested, then reviewed before an exam. Read a word list twice on a Sunday and forget it by Wednesday, and the fair conclusion is that reading a list twice does not work at any age.

Two things reliably move vocabulary into long-term memory. The first is retrieving the word rather than rereading it. Karpicke and Roediger tested this in Science in 2008 with 40 Swahili-English word pairs. Students who kept being tested recalled about 80% of the pairs a week later. Students who kept restudying them, but were dropped from testing, recalled 36%. Everyone predicted about half, so rereading feels as good as it works badly. Roediger and Karpicke found the same crossover in Psychological Science in 2006: rereading wins a test five minutes later and loses a week later. Cover the answer, produce the word from your own head, and only then check.

Here is one rep. The prompt side reads "the bill, in a restaurant." Before you look at anything, say out loud the Spanish word, its article, and a full sentence using it. Target answer: "la cuenta", feminine, and "La cuenta, por favor." Then turn the card over. Full credit means you produced the noun, the correct article and a sentence, in under about five seconds. Partial means you got "cuenta" but said "el cuenta" or froze on the sentence, and a partial counts as a miss: schedule it again tomorrow. Most learners instead see "la cuenta" among four options, pick it, feel correct, and count that as a rep. Choosing between four printed words is closer to rereading than to retrieval, because the answer sat on the screen and your head never had to produce it.

The attempt works even when it fails. Kornell, Hays and Bjork ran six experiments in 2009 and found that unsuccessful retrieval attempts still improved learning, as long as the answer came right after. Their materials were word pairs and general-knowledge questions, not foreign vocabulary, so read it as the mechanism.

The second is spacing those attempts out. Cepeda, Pashler, Vul, Wixted and Rohrer reviewed 317 experiments on distributed practice in Psychological Bulletin in 2006: the gap that produces the best retention gets longer as the delay before the test gets longer. Cepeda and colleagues put numbers on that in Psychological Science in 2008, testing more than 1,350 people: the best gap was about 20 to 40% of a one-week delay, falling to about 5 to 10% of a one-year delay. In days, that is one to three for a word you want next week, six to twelve for a word you want in a month, and about two and a half to five weeks for a word you want in a year. Each successful retrieval earns a longer gap. An app that shows you a word four times in one sitting feels productive and produces very little.

You need years of full-time study before it is useful

The number people quote comes from the US Foreign Service Institute, which trains American diplomats. Its Category I languages, the group closest to English, take 24 to 30 weeks of full-time instruction, or 552 to 690 class hours. Spanish sits at the top of that band at 30 weeks, with French, Italian and Portuguese in the same category. What it buys is Professional Working Proficiency: arguing policy and running a meeting, a much higher bar than yours.

Cambridge English publishes a rough hours table for the Common European Framework, the six-level A1-to-C2 scale used across Europe: A1 at about 90 to 100 hours from a standing start, A2 at about 180 to 200, B1 at about 350 to 400. The framework's own can-do descriptors say an A1 speaker can introduce themselves, can ask and answer questions about personal details such as where someone lives, and can interact in a simple way when the other person speaks slowly and is willing to help.

Say what that table is: published guidance from an exam board, not a measured study with a sample, and the weakest-sourced figure here. Every other number on this page comes from peer-reviewed work with participants attached.

Now do the division. Thirty minutes a day, six days a week, is 156 hours a year, so A1 arrives in seven to eight months, A2 at 14 to 15 months, and B1 at 27 to 31 months. An hour a day halves all three: A1 in under four months, A2 in seven, B1 in about 14.

If your trip is in six weeks

Six weeks is too short for the arithmetic above. Write your own list under five headings: greetings, ordering and paying, directions, numbers and times, and asking someone to slow down or repeat.

Ordering and paying, in Spanish: "Una mesa para dos, por favor." "¿Qué me recomienda?" "La cuenta, por favor." "¿Puedo pagar con tarjeta?"

Slow down or repeat: "Más despacio, por favor." "¿Puede repetir?" "No entiendo." "¿Cómo se dice esto en español?"

That is the granularity: whole sentences you say out loud, not single words to translate later. I have no study telling you how long the finished list should be, and anyone who gives you a round number is guessing. Make it as long as your trip requires and drill it daily with the retrieval rep above.

What to do this week

Pick the language, then write down the specific thing you want to do in it. "Speak Italian" is not a target. "Understand my mother-in-law at Sunday lunch" is, and it tells you which words to learn first. Thirty minutes a day, six days a week, is the 156 hours a year above, and it beats two hours on Saturday, because the gaps are the mechanism.

For the next seven days, learn ten words a day from something you actually want to understand, a song, a show, a real conversation. Give every word an expanding schedule from the ratios above. Monday's words get retrieved on Tuesday, then the following Monday, then three weeks after that, then about three months later. Tuesday's words follow one day behind. Each retrieval you get right pushes the next gap out, and each one you miss pulls that word back to a one-day gap. One word's first year is five reviews: day 1, day 2, day 9, day 30, and around day 120.

On day seven, test yourself cold on all seventy words and write the score down. This checks the method rather than retention, because the longest gap you have exercised is six days: it tells you whether ten a day is a load you can carry. The thresholds that follow are my judgment calls, not findings from the studies above. Retrieving 50 of the 70 means the load is right. Under 40 means you are meeting too many words rather than losing your memory, so cut to five a day and run the week again. Between 40 and 49, hold at ten a day and repeat the seven days on the same seventy words before adding new ones.

The real test is day 21, after the three-week gap has run. Retest the same seventy cold. That score is the one the spacing research is about.