Public Release: 

Smarter Than You Think: Babies are Amazingly Quick Studies, Psycholinguists Find

University of Rochester

One of the hardest things in understanding a new language is to pick out words from the rushing stream of speech. Babies master this challenge around the age of 8 months and do it, surprisingly enough, by thinking like little statisticians, according to a University of Rochester study published in the Dec. 13 issue of Science.

The study offers powerful evidence that infants are extraordinary learners. After hearing only a two-minute sample of speech, babies speedily detect clear patterns in the sounds of language. Such vigorous "computational abilities" can help explain how very young humans learn so much about their world so quickly, the authors say. In a second article called "Learning Rediscovered" in the same issue of Science, two UC San Diego psychologists explain that the Rochester study is an important new argument for the "learning" side of the ledger in the long scholarly debate over whether language is "innate" or "learned."

Linguists have known for years that recurrence of sound patterns plays some role in learning language. Consider the four- syllable phrase, "pretty baby." Because "pretty" is a word, the first sound "pre" is often followed by the second sound, "ty." Similarly, the sound "ba" is often followed by the sound "by." However, the pattern "tyba," consisting of the end of one word and the beginning of another, is less common. As babies hear language, they become aware that sound combinations like "pretty" and "baby" occur more frequently than other sound combinations like "tyba."

The new study shows that babies can use this statistical information alone to learn where one word ends and another begins, according to Jenny R. Saffran, a doctoral student in Brain and Cognitive Sciences at the University, and first author of the study. Professors Richard N. Aslin and Elissa L. Newport are co-authors.

"In real life, babies use lots of cues to tell what's a word and what isn't," said Saffran. "They listen to pauses, to changes in pitch, stress and rhythm to figure it out. But we wanted to see if babies could learn using only statistical information, so we made it really tough on them: We made up a nonsense language, and took out all the cues except statistics to see if babies could still learn. Amazingly, they did."

To test whether babies could extract individual words from continuous speech by relying upon statistical information alone, the authors designed a study using a nonsense language spoken by a voice synthesizer. The synthesizer produced flat, monotone speech with no pauses between words. Twenty-four 8-month-old infants participated.

To familiarize the babies with the nonsense language, the investigators played a two-minute sample of four nonsense words, repeated in random order. The speech sounded something like the following: "bidakupadotigolabubidaku...."

Then, to assess what the babies had learned, the investigators let the babies listen to either words from the language ("bidaku bidaku bidaku...") or sequences that the babies had heard which were not words ("dakupa dakupa dakupa....") These "part-word" sequences spanned word boundaries, like "tyba" in the "pretty baby" example.

In order to see what the babies had learned, the investigators took advantage of a simple fact about infants. "As every parent knows, babies are restlessly curious and get bored quickly," Saffran says. "They'll explore a new toy longer than one that's been around them all the time, so if you want to hold their attention longer, you need to keep coming up with something new and fresh. We thus asked whether the infants would listen longer to the part-words, because if the infants had learned and remembered the words from the two minute speech stream, the part- words should seem relatively new and interesting."

By keeping track of the amount of time infants listened to the words and part-words, the investigators saw clear evidence that babies recognized the difference between the words and the part-word sequences crossing word boundaries. The babies listened to the unfamiliar part-word sequences longer than the now- familiar words - demonstrating, as infants will, a longer attention span for something new.

The investigators don't claim that the babies understood the words in the way adults would -- as being a collection of sounds attached to meaning. Instead, their study shows that after only two minutes of hearing nonsense words, 8-month-old infants can remember familiar sequences of sounds, and distinguish them from unfamiliar ones, and that they can make these distinctions based on statistical recurrence alone -- without the help of other normal cues in speech, such as pauses, or changes in pitch or inflection.

Aslin emphasized the importance of these results: "For many years, language acquisition has been thought to involve a strong innate component in part because children learn language so rapidly. No one believed that young infants could be such rapid learners. Now we have evidence that months before infants begin to produce words, they can very rapidly learn which sounds are likely to go together to form words."

Newport places these results in a somewhat broader perspective. "We know that infants must somehow learn the words of their language, and many other aspects of their particular language's structure. We also know that some aspects of language acquisition are probably innate, and not learned from experience. But our lack of understanding of infant learning, and our expectation that young infants would not be especially gifted learners of complex statistical information about auditory patterns, may have led researchers to underestimate the importance of learning in the language acquisition process."

This research was supported by the National Science Foundation and the National Institutes of Health.


Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.