Out of the Mouths of Babes

Charles Yang shows that toddlers know their grammar.
June 1, 2013

Parents view their child’s first word as an amazing thing when, in fact, learning to speak is something every child does. They all do it in about the same way. And it’s something that no other species can do. It’s amazing.

“Everybody, regardless of differences in their personality and their intelligence and even their family background, goes through the same trajectory during our language learning process,” says Charles Yang, an associate professor of linguistics. There is a long-standing debate, however, about what is happening in the kids’ heads when they speak: Are they just mimicking adults, or do they understand and use a basic abstract grammar?

It has been a difficult question to test. “If you look at very, very young children’s language, which is barely putting two words together, there’s not a whole lot of rich and complex material you can use to draw an inference of what they know,” says Yang, who holds a doctorate in computer science and also has an appointment in the School of Engineering and Applied Sciences. He recently added to the debate with a paper, published in the Proceedings of the National Academy of Sciences, in which he used a mathematical model to help determine the answer.

He compares his analysis to trying to establish whether you’re pulling cards from a full deck. When playing poker, the probability of drawing a pair from a complete deck is about 44 percent. If you draw 10,000 times and find you’ve drawn a pair about 4,400 times, you can be virtually certain it’s a full deck. Yang predicted the diversity of children’s speech if they were drawing “from a full deck of grammar,” then used a database to analyze what they say. The result: “They are producing almost exactly the diversity of language usage you would expect mathematically if they were using an abstract grammar.”

He sees studies like this as part of making the study of language and cognition more fully integrated into the natural sciences. “Language as a biological capacity is very well behaved, such that you can work out a simple mathematical model and test it on two-year-olds,” he says. “Because language has some very mechanical ways of putting things together, ultimately rooting in our brain and our biological capacity, it is possible to integrate a study of language in a more quantitative form of science.”

How language is learned by children is intimately related to how language changes, another major topic in Yang and other Penn linguists’ research. Language change can be likened to evolution, as information—be it genetic or linguistic—is transmitted over generations of individuals. Understanding how language and speech is produced, perceived, and learned, he says, will help linguists construct and test models to predict change.

As an example, he cites a study in which children changed pronunciation of American English vowels after the community’s total population included only 20 percent of differently-accented speakers, who had moved there from a nearby city. “You might expect intuitively that you’d need more than half of the people [to speak differently] for the new pronunciation to take over, but it turned out to be much less,” he says. When children were within their window of language learning and with a peer group, the change happened in four or five years.

Yang hopes to be able to use this knowledge to understand why language changes, and ultimately to see how language behaviors can be related to the neurological and genetic levels. “Children learn languages in a pretty systematic fashion, which would seem to suggest that they are following some kind of rough blueprint—they’re not just trying out wild things,” he says. “Even the darnedest things they say are fairly well behaved.”