Natural Language Computer Interface

Getting a computer to understand a request for information in everyday language may be more than a matter of convenience. Researchers with the Institute for Robotics and Intelligent Systems (IRIS) point out that the computer could show you things you could not have found any other way.

Computer scientists at the University of Regina and Simon Fraser University have developed two complementary approaches to helping a computer process a command in plain English terms. A natural language/ database interface, called System X, employs elaborate rules of grammar to parse a command into its essential components. A second system, called DBLEARN, enables the computer to use the attributes provided by a natural language instruction to find new linkages within a database.

The researchers refined a prototype of System X by trying it out on vice-president of Roger's Cablesystems Ltd., a key supporter of the work. He wound up keeping the equipment several weeks longer than anticipated because he found it so useful. "He was getting graphs of relevant data in two minutes that he had been taking six weeks to get," says Dr. Nick Cercone, vice-president of research at the University of Regina. "That meant he was able to make better decisions more quickly."

The system allowed the executive to enter into his computer terminal a statement like "Give me the Western region outage log for June," which would quickly yield a table or graph of the relevant data from the company's database.

Dr. Cercone says such an interface, which could be wedded to DBLEARN, provides much more than simple ease of use. In an experimental trial with the Natural Sciences and Engineering Research Council's Grants Information System, he and his colleagues demonstrated how DBLEARN "discovered" new groupings between various parts of the grants information database. "These are mined results classification information, categorization information, association information," he says.

For more information please visit the PRECARN Web site.