Explain me something: How we learn what not to say

Adele Goldberg

Adele Goldberg, a professor of linguistics in Princeton’s Council of the Humanities, and her research team use nonsense words — such as ablim, adax and afec — to study how we learn which combinations of words are allowed and which are not. (Photo by Till Dreier. Illustration by Ilissa Ocko)

Explain me something. She considered to go. The asleep dog snored. We have learned to avoid using these phrases although it is difficult to say exactly why explain me is not allowed but tell me is acceptable. Somehow we have learned that consider to go is wrong but decide to go is right, and that although the asleep dog sounds odd, the sleeping dog sounds fine.

So how do we learn what not to say?

Through a series of experiments, linguist Adele Goldberg and her team believe they have found the answer: language learners create expectations about how words and phrases will be used, and if a different form is repeatedly heard instead, the previously expected form will be ruled out.

For example, when a young language learner expects to hear he disappeared the rabbit, but instead repeatedly hears he made the rabbit disappear, then the child learns that the second sentence is preferred, said Goldberg, professor of linguistics in Princeton’s Council of the Humanities. Linguists call this the “statistical preemption” theory because language learners record the frequency and context of phrases that then preempt the use of the expected phrases, Goldberg said.

This cycle of expecting one formulation followed by witnessing an alternate formulation helps language learners eliminate phrases that at first glance would seem to be allowable.

To test the idea of statistical preemption, in one study, Goldberg and postdoctoral researcher Jeremy Boyd looked specifically at adjectives that start with an unstressed a, such as asleep and afraid. (In Old English, asleep was a contraction of in sleep so the phrase, the in sleep dog was not used, explaining our modern day avoidance of the asleep dog. Many “a”-adjectives follow this pattern.)

Boyd asked 32 college-age volunteers to create new sentences using nonsense “a”-adjectives, such as afec, of which people have no prior knowledge. He found that when people had been led to expect to hear the afec fox (which is analogous to the asleep fox), but then heard the preemptive phrase the fox that was afec (which is analogous to the fox that was asleep), they quickly learned to avoid using the afec fox, a finding that affirms the statistical preemption theory.

Boyd also found that when the context was manipulated so a new set of volunteers did not form an expectation to hear the afec fox, the volunteers showed no evidence of avoiding it, even though they had heard the preemptive phrase the fox that was afec. The results were repeated using many other nonsense “a”-adjectives.

“From this study we concluded that learners use preemptive contexts and what is more, they are smart about what counts as a preemptive context,” said Goldberg, who receives support for her research from the National Science Foundation and the Free University of Berlin.

Goldberg’s work on statistical preemption helps dispel a competing theory: that the reason we don’t use phrases such as explain me is simply that we only use language in ways we have heard before. If this were the case, Goldberg said, we would never be creative with language.

“The verb sneeze has been heard a million times in a simple sentence such as, ‘She sneezed,’” Goldberg said. “But we can also think of new ways to use it, as in, ‘She sneezed the foam off the cappuccino.’ We clearly know how to use language creatively.”

–By Catherine Zandonella