Skip to contents

National Science Foundation Logo

Search KDI Site

Banner: Multi-Disciplinary Research at NSF: Accomplishments of the KDI Initiative
 KDI Home    Contact Us   

Image: People Button

Image: Ideas Button

Image: Tools Buttong

 About KDI

 Behind the Scenes

 Taking Stock

 Links and Resources



Investigators Recognized for Innovative Work in Language Acquisition

Researchers at the University of Rochester have been recognized for their work in developing new approaches to understanding how children learn their first language.

University of Rochester professors Richard Aslin and Elissa Newport, along with then-graduate student Jenny Saffran, have developed a new way of looking at language acquisition called statistical learning. They hypothesize that, as a part of learning languages, humans compute how often particular sounds occur and the probability that those sounds will occur in particular sequences. The idea is that babies are like little computers, naturally keeping track of these statistics as they hear speech, and using them to learn.

Learning a language is much like breaking a code. When one hears a stream of speech in a new language, one hears pauses between sentences, but there are no pauses between words and no obvious signals that indicate how the words can be combined to form phrases and sentences. The task the learner faces is how to break this code, first finding the words and then acquiring the sequences in which the words can occur to express different meanings.

What the investigators at Rochester brought to this puzzle was the idea that people may actually be learning words by figuring out the probability that particular syllables will occur together, in a specific sequence. By keeping track of those statistics about the syllables, they can figure out what the words are. The investigators tested this hypothesis on a synthetic speech stream with infants and adults. They found that both infants and adults perform the kind of computation they had predicted and with remarkable speed. These results were replicated in studies using tone sequences, visual sequences, and motor learning tasks. In collaboration with Harvard Professor Marc Hauser, Newport and Aslin also tested their theory on tamarin monkeys and found that the monkeys can accomplish the most basic sequencing tasks as well. Taken together, these results suggest an ability to keep track of sequential statistics that is very general, appearing in humans for a wide variety of segmentation tasks, and even appearing in some nonhuman primates.

Of course monkeys do not acquire human languages. In order to explain why humans show such remarkable and unique abilities to learn human languages, there must be a further explanation. Newport and Aslin went on to more complicated learning tasks, where there is indeed a more complex set of results. They found that human learners were adept at complicated sequencing tasks that mirrored common language structures, but performed poorly at more complicated tasks that were not like sequences appearing in natural languages. Monkeys show a different pattern of successes and failures. In part, then, the nature of human languages may be shaped by these selectivities of learning: those structures that humans find easy to learn will appear commonly in languages of the world; those that humans find hard to learn will never appear in languages. Insofar as humans differ from other primates in the sequences they notice and learn easily, their communication systems may also develop differently.

The logical next step is to move from studying how people learn words to how they learn grammar, and it is on this task that the researchers are now focusing their attention. They are studying whether sentence structure is learned through a process of computing the frequencies of particular words and the probability of one word following another, just as words are learned by recognizing similar patterns with recurring syllables.

Why is this work important? Elissa Newport cites two reasons: "First, while we know a great deal about the structure of languages, we don't know much about the method by which languages are learned. We know that infants are remarkably good at learning any of the human languages they are exposed to. What we really don't know is exactly how they do this, what computations the brain is performing to learn a language. The main contribution of this work is to clarify the kinds of mechanisms that may be at play. Second, understanding this mechanism may help to explain why we find some of the recurring patterns that we see in human languages around the world, and why other patterns that seem perfectly reasonable don't show up in any human languages."

These researchers have been recognized for their important contribution in the area of statistical learning: Elissa Newport was named a fellow in the American Academy of Arts and Sciences, Elissa Newport and Richard Aslin were named fellows in the Association for the Advancement of Science, and Jenny Saffran received a Presidential Early Career Award for Scientists and Engineers. Saffran, who contributed to this research while a graduate student at the University of Rochester, is now an associate professor at the University of Wisconsin. Aslin and Newport are professors in the Department of Brain & Cognitive Sciences at the University of Rochester.


Back to Top of Page

People | Ideas | Tools
About KDI | Behind the Scenes | Taking Stock | Links and Resources

KDI Home | Contact Us | Site Map
NSF Home | CISE Home | Privacy Statement | Policies | Accessibility

National Science Foundation: Celebrating 50 Years Logo The National Science Foundation
4201 Wilson Boulevard, Arlington, Virginia 22230, USA
Tel: 703-292-5111, FIRS: 800-877-8339 | TDD: 703-292-5090