Now there's a computer program that will thumb through the files

A three-drawer filing cabinet, as most office workers know, eats files. It snatches them from the fingertips and, as soon as the drawer is shut, shuffles them off to some obscure location and label. Then try finding something, especially if it's a file someone else has put away. Computer filing systems are an improvement. They can yank a file in a split-second -- but only if asked correctly. There's no way to reach inside and thumb among the memory chips until you find the needed document. Multiply that necessity by the hundreds of thousands of pages companies store on computer tape and disk drives and you face a serious problem.

But a tiny software company in Boulder, Colo., thinks it has a solution, one that can open the door into those massive data banks in an unusual way. Information Access Systems Inc. has been slowly but steadily marketing a program design that, in essence, makes a mini or mainframe computer think like a person when rummaging for files.

The program, called J-Space, allows a computer user to send the computer in search of documents that pertain either generally or specifically to a subject. The system is unique, its owners assert, in its ease of use and ability to duplicate human judgment in determining what sort of material to find.

Earlene Busch, a behavioral psychologist and president of IAS, offers the word ``hypography'' to illustrate how J-Space works. A technical term that describes the use of graphics to document activity at a mining site, the word is part of the system IAS installed at the Amax Molybdenum division headquarters in Golden, Colo.

When the computer is asked for whatever it has on hypography, she says, J-Space delivers, in priority, descriptions of documents that describe graphics as well as those that describe mine sites. It then summons requested documents.

``It's very likely, though,'' says Dr. Busch, ``that many of those documents won't contain the word `hypography.' But J-Space knows what the word means, can make a judgment about what is being requested, and find the appropriate material.''

Few other text retrieval programs can be used with plain English commands and make inferences beyond the specific command, says Dr. Busch.

Law firms use these sorts of programs to research precedents. Most systems can find cases on topics as broad as breach of contract suits in Fortune 500 companies or as narrow as defamation of character suits against a particular publication. However, these programs often require extensive coding on every document entered into the data banks and a trained operator to run the system. Documents are identified in a ``key word'' manner. When those documents are sought later, the operator uses the key words to locate them.

Documents typed into the IAS system need no coding, which means anyone who can type can store them. And the J-Space system gives a user not just documents with the key word ``hypography'' in them but also looks for information that relates to the meaning of the word.

The heart of the process lies in J-Space's attempt to define human behavior. ``Turning that into a program was actually fairly simple,'' says Peter Ossorio, a University of Colorado psychology professor who developed J-Space. ``It can't be done using computer science approaches. It can be done using a behavioral or artificial intelligence approach. . .. What I've done is drop the mathematical model and use a person model. I use that model to design systems so they do the same things people do.''

Essential to the systems' design are the similar types of judgments made by people in similar lines of work. Police, for instance, attach the same values to certain subjects -- vehicles, guns, marijuana -- as other police do. Dr. Ossorio studies those values and designs a system that will respond appropriately.

Seattle's King County Police Department recently bought J-Space to track the information on a series of related homicides. There are now some 80 volumes of documents plus 17,000 tip sheets packed with information related to the crimes, says police Lt. Dan Nolan, project manager for the computer system.

``A homicide generates a tremendous amount of textual material,'' Lieutenant Nolan says. These range from formal police reports to letters from the community. ``We needed a way to manage that. . .. Investigators can become experts with an hour's training.''

Drs. Ossorio and Busch sifted through documents related to the case, assembled a long list of relevant subjects and an even longer vocabulary of often used words. They then gave a ``relevance rating,'' on a scale of zero to nine, to each vocabulary word as it applied to the subject matter. The process took five months. Though a suspect has not as yet been apprehended, police say they now at least have a better means of relating fragmented pieces of information.