June 20, 1996 Contact: Barbara McKenna (408) 459-2495; mckenna@ua.ucsc.edu

ANNUAL CONFERENCE WILL DRAW NEARLY FOUR HUNDRED COMPUTATIONAL LINGUISTS TO UC SANTA CRUZ

FOR IMMEDIATE RELEASE

If ewe can sea the errors inn hear your a head of my spell checker. Eye wont a bettor won.

SANTA CRUZ, CA--Anyone who uses a spell checker with his or her word-processing program knows that the tool, while quite handy, misses many of the worst errors because it isn't sophisticated enough to examine words in context. But a program currently in development by an Edinburgh University researcher offers hope for a smarter computer--one that would notice when the noun "ewe" is sitting where a pronoun should be and deduce that the pronoun is most likely "you."

The Edinburgh researcher, Andrei Mikheev, will present his work during the 34th annual meeting of the Association for Computational Linguistics (ACL), which takes place June 23-28 at the University of California, Santa Cruz. Close to 400 scientists, researchers, and scholars from business and academia are expected to attend the conference.

Advances in computational linguistics are rapidly bridging the communication gap between computers and their human users. At the weeklong ACL conference, many cutting-edge advances such as Mikheev's will be presented at some 40 workshops and lectures. Among the innovations conference attendees will report on are the following:

-- Mikheev will present a talk on his program, which is able to identify types of words (e.g., distinguish nouns from verbs) without being told anything about grammar. The program works out its own rules for classifying words by reading text and making tabulations based on what it has observed (rather than on what it has been told in advance to do). Not only can the program catch a correctly spelled word out of context (as in the above sentence) but it doesn't have to have the word stored in its memory banks before it begins the process.

-- An experimental setup devised by psycholinguist Michael Tanenhaus that helps researchers study how we understand spoken language. The device, which gives one an uncanny sense that it is reading the mind of the subject using it, tracks minute eye movements by means of a laser beam. That information is transmitted to a computer that figures out the exact spot the subject is looking at. The information from the computer is merged at high speed with a video camera attached to the subject's head. The result is a video image with crosshairs that shows exactly where the subject's gaze goes, even for periods as short as a fifth of a second. The tool is astonishingly effective at measuring how quickly someone comprehends specific instructions and which words in the instructions actually spark comprehension, second by second. This is the most effective approach yet for looking into a human mind to see how language comprehension is accomplished.

-- A program--run on a laptop computer--that can translate between English and Korean. Work on machine translation programs has been going on for three decades, but the programs usually involve European languages. It is only in recent years that there has been progress on automatic translation between thoroughly unrelated languages like Korean and English.

-- New methods of parsing--the process of reading a sentence and figuring out its grammatical structure and its meaning--that teach programs to eliminate irrelevant senses of words when necessary. (If your friend said she would wait for you at the bank, should you look down by the river or head toward the local Wells Fargo?)

-- Novel mathematical work on methods for parsing and for measuring the success of those methods.

This release is also available on the World Wide Web at UCSC's "Services for Journalists" site (http://www.ucsc.edu/news/journalist.html).

Press Releases Home | Search Press Releases | Press Release Archive | Services for Journalists

Maintained by:pioweb@cats.ucsc.edu