Excalibur Technologies Inc, McLean, Virginia, has set up a subsidiary in Windsor, Berkshire, to market what it claims is the only document image management and control software based on neural network technology. The new subsidiary will market the PixTex/Electronic Filing System worldwide except in the US, where it has been available for the past two years, and in Japan, where it is already licensed to several companies. PixTex/EFS document image management and control software is based on a proprietary pattern recognition technique. Normally a user searches on an individual word or stem of a word to retrieve text.
Fuzzy logic
PixTex/EFS, however, sees individual words as binary code, and indexes them automatically as a series of ones or noughts, so documents are indexed as a series of binary patterns. This eliminates the need to create key-word tables, topics or directories for data retrieval, which, in turn, eliminates discrepancies in the criteria used when indexing information. Information retrieval is carried out on two levels. When the user wants to find a particular text, he keys in a word or string of words, related to the file he requires. The neural network – intelligent software that emulates the way the brain works and learns from experience – compares the binary patterns of the key words with those contained in the index. So-called ‘fuzzy logic’ then locates roughly comparable binary patterns within any text stored on the database. So users can retrieve information from large data files even if a word is misspelt at either the search stage or in the original document, or if the document has been misfiled. The system will tell the user how many comparable binary patterns it has found, and list them in order of accuracy. Excalibur claims that between 80% to 90% of these ratings are correct, and that 200,000 pages of text can be searched in approximately 10 seconds. The user then simply goes through the texts that have been thrown up to find the one he wants. PixTex has a client-server design and when first released in the UK this March, ran only on Digital Equipment Corp’s VAX servers under VMS, and DECstations under Ultrix. Excalibur has a particularly close relationship with DEC. The company was set up in 1980 by biologist James Dowe, who used the research he had done on neural networks to develop a fuzzy keyboard to help him overcome his dyslexia. His findings were based, among other things, on the mechanism that enables frogs to learn from experience which fly they can eat and which they can’t.
By Catherine Everett
Dowe quickly brought in a team of software developers to help turn his ideas into bespoke applications such as the fingerprint recognition system currently used by police forces in the US. But about three years ago, Excalibur ran into financial difficulties and filed for Chapter 11 bankruptcy protection. The PixTex/EFS system was under development at the time, and Dowe won backing from various sponsors on the strength of it. DEC, among others, took shares in the company, which went public in 1986, and the two have a worldwide licensing agreement; DEC now sells PixTex/EFS as its own product under the Excalibur name. However, PixTex has run on other servers such as IBM Corp’s RS/6000’s under AIX, Hewlett-Packard Co’s HP 9000 range under HP-UX, and Sun Microsystems Inc Sparcstations since July this year (CI No 1,960). By the end of December, it will also run on Novell Inc NetWare servers. IBM and Hewlett-Packard will soon become distributors of the product, although they may well rebadge it. Other distributors include Zeta Systems Ltd in Bristol and Metrologie Internationale SA’s UK branch in High Wycombe. There are several more in France, Italy and Belgium, but Excalibur also sells its product direct. Any personal computer using Microsoft Corp’s Windows, Open Look or Motif graphical user interfaces as well as Apple Computer Inc’s Apple Macintosh can be used as a client. PixTex/EFS can be used in three different ways. First, it can link directly into Oracle Corp, Ingres Corp and Informix Corp
databases, by hooking into the database engine and passing an SQL statement directly to it. The server handles and stores the data. When a client machine requests a certain file, the server compresses it before sending it down the network. The client decompresses the information at the other end, and the file can be accessed immediately by concurrent users. Excalibur currently has marketing agreements with Oracle, Ingres and Informix (CI No 1,908) and is holding discussions with Sybase Inc at the moment.
Filters
Second, the product can index and retrieve unstructructured information, held electronically, by means of special filters that access word processing packages such as Microsoft’s Word for Windows and WordPerfect. The software is simply loaded onto a personal computer as a Windows application, and the user chooses the relevant filter from a menu. Users with one word processing package can view documents written using another package, but cannot modify them unless all of the ASCII information contained in the text is removed. Third, PixTex can be used as a document imaging system in conjunction with an industry-standard scanner, such as those manufactured by Fujitsu Ltd or Hewlett-Packard. The scanners act like electronic photocopiers, seeing the image as a series of dots, which are then stored on the server. The PixTex software indexes the text immediately. Electronic data can also be collected from computer disks, facsimiles and CD-ROM. PixTex is available in the UK now. Price ranges between $16,200 for a single user on a Unix box to $348,000 for 100 concurrent users. Running the software on a VMS machine costs approximately 10% more, but running it on a NetWare server is expected to cost substantially less. No sterling prices were available.
