17.4 is a model estimate, not a universal language-learning deadline

“You have to start before puberty.” The advice often presents one age as a biological deadline. The research does not support one universal cutoff that predicts an individual learner's outcome.

The number 17.4 is important, but it is a model estimate from one English syntax dataset, not a directly observed moment when language learning changes for everyone.

What the 2018 study estimated

Hartshorne, Tenenbaum and Pinker analysed data from 669,498 native and non-native English speakers who completed an online grammar test. Their computational model separated current age, age at first exposure and years of experience, then estimated that grammar-learning ability remained high until about 17.4 years before declining.

In the same dataset, native-speaker grammar-test performance approached its asymptote around age 30, although the authors said most improvement occurred during the first 10 to 20 years. That is a finding about performance on this English syntax test, not evidence that every part of a first language develops on the same schedule.

The paper also distinguished learning rate from ultimate attainment. Its authors estimated that learners generally had to begin around age 10 to 12 to reach native-like performance on the test, because learning takes years before the later decline in rate. The 17.4 estimate was never a promise that starting at 17 guarantees a native-like outcome.

Later analyses do not agree on one breakpoint

A 2021 analysis by Chen and Hartshorne expanded the dataset to more than 1.1 million participants, used improved psychometric measures and considered alternative models. It supported a sharp decline in syntax-learning ability near the end of adolescence.

A 2022 reanalysis by van der Slik and colleagues reached a different conclusion. When monolinguals, bilinguals and different immersion groups were modelled separately, continuous curves fit monolinguals, bilinguals and early-immersion learners better. Discontinuous models fit only non-immersion and later-immersion learners, with estimated breaks at 18.6 and 19.0, which the authors interpreted as possible schooling effects rather than one cognition-driven critical period.

The defensible conclusion is therefore not “the critical period ends at 17.4.” It is that age is related to English syntax learning, while the shape, cause and existence of one universal breakpoint remain disputed.

What these studies do not measure

The task tested English grammatical knowledge. It did not measure pronunciation, vocabulary, conversation, motivation or every skill covered by the phrase “learning a language.” The curves describe group averages and cannot predict an individual learner's ceiling or rate.

This is also why initial learning speed and long-run attainment can point in different directions without contradicting each other. Older beginners may progress faster at first, while younger starters can have a long-run advantage.

What follows for planning

Our model has no age coefficient because these studies do not provide one reliable multiplier that applies across learners, languages and skills. That does not mean age is irrelevant; it means FluencyClock cannot turn age into an honest personal correction factor.

If you are deciding whether to start, 17.4 is not a personal deadline. Choose a concrete target, count the time you can sustain, and treat the resulting estimate as a planning range rather than a prediction of native-like attainment.