Illustration: Helena Pallarés

For 50 years, educators and linguists have generally accepted that the critical period for mastering another language ends just before puberty begins. Yet most of the supporting research has relied on small samples, often of only a few dozen people, and has primarily tested subjects’ ultimate knowledge of a language, not their ability to learn, “which is like saying only that two runners finished a race, without saying how fast each ran,” says Joshua Hartshorne, an assistant professor of psychology and director of Boston College’s Language Learning Lab in McGuinn Hall. To study how the ability to learn languages changes over time, Hartshorne says, “You need a lot of people who’ve been learning a language for different numbers of years and who started at different ages. What better tool for that than the internet?”

While Hartshorne was a research assistant at Harvard in 2006, he created gameswithwords.org. Inspired in part by viral online quizzes such as “Which Harry Potter Character Are You?”, the site features short, concise tests “probing the nature of language and memory” (for example, a multiple-choice quiz on word meanings to examine how birth order affects vocabulary). More than two million visitors (including some revisitors) have participated in his crowdsourced studies, which have been funded by the National Science Foundation, the National Institutes of Health, and the American Psychological Foundation. Among the findings: Working memory (the ability to temporarily store and apply information) peaks at around age 30, a decade later than previously thought; and the ability to read and recall facial expressions peaks around age 45.

To lure participants into a more comprehensive study of language acquisition, Hartshorne devised a five-minute quiz that, he intimated, could deduce participants’ first language and English dialect. He spent a year and a half researching the idiosyncrasies of dozens of languages and English dialects (for example, “‘I am done dinner’ is a classic Canadianism,” he says) to hone the quiz. What the test actually measured, however, was command of English grammar, via 132 syntactical judgments of varying complexity (for example, fill in the blank with all correct answers: “I ____ for six hours by dinner time” [A] will have studied; [B] will have been studying; [C] will had studied; [D] will be studying”). After participants finished and received Hartshorne’s guess—accurate 80 percent of the time—Hartshorne collected information on their age, first language, location, and when they began learning English.

Hartshorne launched the quiz by posting a link on his personal Facebook page one night in May 2014. When he awoke in the morning, his site had crashed. Over two weeks, he garnered 669,498 subjects, ages seven to 89, representing more than 6,000 native languages and dialects from Abenaki to Flemish to Zulu. It was an unprecedented set of data that took more than three years to assess.

To help with the analysis, Hartshorne recruited his postdoctoral advisor, MIT professor of cognitive science and computation Joshua Tenenbaum; and cognitive psychologist Steven Pinker of Harvard. The results, which were published in the August 2018 issue of Cognition, indicate that people maintain a peak ability to learn English grammar until age 17.4 years, some 10 years longer than commonly thought. Thereafter, a “rapid” decline in that ability ensues. The results also suggest that because English takes many years to master (in fact, 30-year-old native speakers show the greatest proficiency) children must begin learning it around age 10 to reach native-like fluency, or else they “simply run out of time.”

So far, the research has uncovered the critical period only for learning English grammar. Hartshorne is now conducting several “massive online experiments” with other languages and other components of language, including vocabulary and pronunciation.

Raised in Kansas, Hartshorne has studied seven languages, including Classical Hebrew, Japanese, and Mandarin. After graduating with a degree in math from Oberlin College in 2002, he spent a year building nature trails in Siberia, so he could “wake up and speak Russian all day.” His investigations “may look overly multipronged,” he says, examining the “intersection between linguistics, neuroscience, artificial intelligence, and psychology” (his 19 undergraduate research assistants major in psychology, computer science, mathematics, biology, and economics). But all of his research, and much of his life, “is driven by one fundamental question: How is language possible? All scientific evidence suggests it’s impossible, except for the fact that we use it.”

Read more by Zachary Jason