Adele Goldberg, a professor of linguistics at Princeton, and Jeremy Boyd, a postdoctoral researcher now at the University of Illinois, asked adult speakers to produce sentences containing made-up adjectives, and the researchers found that the speakers carefully evaluated the input they received, and that learning only occurred when the input was deemed informative. The subjects discounted input that seemed irrelevant.
“If a child sees an adult with full casts on both arms turn on a light switch with his head, do they assume that the light can only be turned on with one’s head? If children learned purely by imitating others then this would be the case, but learning is much more sophisticated and creative than this,” Goldberg said.
“This is especially true when it comes to language, where the fact that children routinely say things that they’ve never encountered before indicates that they are not simply imitating what they hear,” she added.
The researchers found that people avoided using a made-up adjective before the noun it modified, unconsciously treating it like existing adjectives that sound similar. The research subjects did not learn restrictions when seemingly patterned use could be attributed to some irrelevant factor. The results of the research will be published in the March issue of the journal Language.
Goldberg explains their results as follows:
“Speakers routinely say things that they’ve never encountered before, and yet at the same time they know not to use certain semantically sensical and syntactically reasonable formulations. This fact has bedeviled language acquisition researchers for a long time. How can children learn that ‘The magician vanished the rabbit’ is odd, given that ‘The lion king banished the rabbit’ is fine (if a bit unusual)? And how can they learn that using certain adjectives like ‘afraid’ prenominally (as in, ‘the afraid boy’) is dispreferred relative to predicative use (‘the boy that’s afraid’), despite the fact that most adjectives do appear prenominally, e.g., ‘the scared boy’ is perfectly natural? After all, adults do not reliably point out these fine details to children.
“In a set of new studies conducted at Princeton, we had speakers produce sentences that contained made-up adjectives like ‘ablim’ to see whether they would generalize the restriction against using ‘afraid’ prenominally to new adjectives that sound similar. We found that speakers did generalize the restriction to some extent, demonstrating that people can generalize not only on the basis of what they witness, but also on the basis of what they fail to witness.
“But how is this restriction learned in the first place? Additional results demonstrate that witnessing ‘ablim’ used predicatively instead of prenominally (e.g., ‘the cow that’s ablim’) makes participants even more likely to avoid its prenominal use in their own utterances. While this may sound like learners are simply imitating the adjective uses they see in their input, a final experiment revealed that the learners are quite savvy. When a new group of participants witnessed ‘ablim’ used predicatively, but this time in a context in which there was a reason for its predicative use that had nothing to do with ‘ablim’ itself, participants did not learn a restriction against ‘ablim’s’ prenominal use. This indicates that learners carefully evaluate the input they receive, and that learning only occurs when the input is deemed informative.
“This research establishes that language learning goes well beyond simple imitation, and is in fact quite creative and remarkably smart. Learners are able to generalize grammatical restrictions to new members of a category, and they do not learn restrictions when seemingly patterned use can be attributed to some irrelevant factor.”
The work was supported by the National Science Foundation and the Freie University of Berlin.
Goldberg is available to discuss the research with interested members of the news media, and a copy of the study is available upon request. Members of the media wishing to contact Goldberg are encouraged to e-mail her at firstname.lastname@example.org.