How children use the power of prediction to learn new words

Having a seamless conversation with another person is an ability we often take for granted. But for cognitive scientists, this type of rapid turn-taking demonstrates the remarkable dexterity of the brain. The gaps between turns are short – around 200 milliseconds or so – but the latencies involved in language production are much longer, over 600 milliseconds. This implies that the listener must make predictions about the speaker’s content and timing.

Adults can easily make such predictions after a lifetime of language experience, and some researchers have even proposed that prediction plays a key role in learning language during childhood. For instance, infants as young as 18 months use context to infer the possible meaning of new words. If they hear “It’s a dax,” they assume the word “dax” refers to an object. This phenomenon, called syntactic bootstrapping, could contribute to children’s ability to learn the meaning of new words.

Naomi Havron, a postdoctoral researcher at the Laboratoire de Sciences Cognitives et Psycholinguistique at the École Normale Supérieure, wanted to find out whether prediction can drive language acquisition by presenting 3- to 4-year-old children with novel words. By first presenting these novel words in a certain context and sentence structure, she tested whether children’s expectations of these words upon hearing them again update rapidly enough to allow for prediction-based learning. The findings were published by the journal Child Development in July 2018.

“In the current study, we were interested in how flexible these expectations are, and how quickly young children change them,” said Havron. “We wanted to know whether children constantly track what words precede nouns or verbs, or whether their expectations are quite fixed.”

Participants included 45 monolingual French-speaking children aged 3 to 4 years old from middle- to high-socioeconomic status homes. In French, la petite is most often followed by a noun (e.g. la petite grenouille, “the little frog”) but can also be followed by a verb (e.g. la petite dort, “the little one is sleeping”). The researchers exposed one group of children to a verb condition in the induction phase, where la petite was consistently followed by familiar verbs, and the other group to a noun condition, where la petite was consistently followed by familiar nouns.

In the testing phase, they observed whether children used their expectations based on the induction phase to infer the meaning of novel nouns and verbs. For instance, instead of “the little frog,” the researchers would present “the little aye-aye,” and instead of “the little one is sleeping,” they would present “the little one is yielding.” This was done by recording their eye gaze while watching two videos placed side-by-side on the screen. One video represented a verb by showing a girl performing an action, while the other represented a noun by showing the same girl holding an object.

“Children are very fast trackers of structures in their language.”

After only four induction trials, children from the verb condition looked more at the action video than children from the noun condition, and vice versa. The results suggest that children in the verb condition were more likely to interpret the novel word as referring to the action than children in the noun condition, despite la petite being much more commonly found in front of a noun.

“Using only four items was a result of a constraint of the experiment’s duration, and we were quite nervous that four examples might not have been enough,” she said. “So the results did put us at ease — children are very fast trackers of structures in their language.”

The children in the verb condition not only changed their expectation of sentence structure, but they also used the adapted prediction to guide their learning of an unfamiliar word. Havron has some initial ideas about how this kind of research could lead to training videos or other such interventions for children with language learning difficulties — but she stresses that more work needs to be done first.

“I would be very interested in studying how we could use these findings to support interventions for autistic children, language impaired children, and second language learners,” said Havron.