News Release
Teaching Computers to Learn Language"A child hears language and is eventually able to understand language and use it in context," says Dr. Brill, assistant professor of computer science in The Johns Hopkins University School of Engineering. "Computers are below child-like in their understanding, so we're trying to get them to act like we act with language and get them to automatically learn it." It's not an easy task, adds Dr. Brill, who is also affiliated with Hopkins' Center for Language and Speech Processing. "Language is hard to teach computers. It's an amazingly dynamic medium. People are coining phrases all the time -- like Jordanesque, for playing exceptional basketball -- that a computer won't comprehend." Today, despite the speed, power and storage capacity of most computers, information retrieval can often be difficult or laborious. If a learning algorithm to allow automatic induction of grammatical structure from text, or a spoken language system to access information from a computer, can be fully developed, it will have an endless array of applications. "Let's say you want to find references to newspaper articles on environmentalism, published at a certain time, that are negative in tone. You'd start with a key word, like environment, and get an enormous list to wade through. A smart computer would understand language, and do the filtering for you." Another potential application might be the development of an air travel information service, which would allow customers to acquire information, place reservations and pay for tickets by talking one-on-one with a computer. Dr. Brill, who came to Hopkins after working as a research scientist in the Spoken Language Systems Group at the Massachusetts Institute of Technology, was recently awarded a National Science Foundation Faculty Early Career Development grant to further his innovative research. For more information about Dr. Brill, or the Center for Language and Speech Processing at The Johns Hopkins University, call Ken Keatley at (410) 516-7907. His E-mail address is jkk@resource.ca.jhu.edu.
Go to Headlines@HopkinsHome Page
|