How do you create hypertext systems? ==================================== Having already built several very large hypertext system (multi-megabyte), I've developed both techniques and tools to rapidly create hypertext systems . Here's the steps we go through to convert 150 pages of text into a comprehensive hypertext system : STEP 1 OCR (Optical Character Read) the text. We have machines that read both typewritten and typeset text. See . STEP 2 Clean up the text -- run spelling and grammar checkers, then proof the text to correct OCR errors STEP 3 Split the text into files -- Using a word processor, we add codes, a file name, and a file descriptor at each new idea unit in the text (generally every 1-4 paragraphs). Our SPLITTER program uses these codes to split the original files into smaller ASCII files. The 150 pages might become 500 ASCII files. STEP 4 Build the master cross-reference index to the ASCII files. Using our MARKER program , we mark the key words and phrases in each file (and add synonyms if necessary). We highlight perhaps 3-4 words/phrases per file, then assemble these word lists into a master cross-reference file. STEP 5 Put the cross-references into each ASCII file. We load the word/phrase cross-reference list into HOUDINI, then use this network to identify files that share common words and phrases. This information is added to each ASCII file. STEP 6 Integrity check. We use another program to make sure the ASCII networks are valid (no isolated files, no dead-end links, etc.) STEP 7 We use both a decision-tree builder and HOUDINI to build the knowledge hierarchies/networks that lead to the ASCII files. Although it appears to be complex and time consuming, the overall hypertext construction speed really depends on the user's knowledge of the subject area. However, we've had considerable experience in building systems for others. These systems typically range 2-4 megabytes in size, contain 1,000-plus files, and 2,000-3,000 hypertext links. Specifically, these hypertext systems have organized large amounts of procedural law arising out of special hearings. We converted all the minutes from many technical hearings into a comprehensive index by topic, issue, commentary, opinion, and consequences. As for performance, the systems we've built provide users, in an average of less than 10 keystrokes, with immediate references to any ruling, topic, or issue affecting certain areas of their profession. As for production standards, we believe the systems we've built are larger than any reported in the hypertext literature to date. In addition, while hypertext literature speaks of a year of effort to build modest systems (200 nodes), in practice we do that in a day. That leads to the question of...how do we do it? Well, we use a number of software tools to build such systems -- splitters, markers, reference-builders, hierarchical and network linkers, integrity checkers, compactors, and installer/verifiers. While we see these tools as essential to any hypertext efforts, we find no mention of such software techniques in the literature of hypertext, nor by companies or academia with hypertext software, nor from persons/reviewers claiming an understanding of hypertext. From all that, I think that you can draw your own conclusions. Just remember the tale of the six blind men and their efforts to desribe an elephant. Hypertext is much the same as most visions of the process do not match the methods or software needed to accomplish the task. Hypertext construction is a new field...and I'd say that users of MaxThink and HOUDINI understand more of the principles of hypertext than anyone else. Why? 旼컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴� � Good hypertext depends on abilities to clearly � � categorize information. � 읕컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴� That is something every MaxThink and HOUDINI user already knows and something users who are unfamiliar with hierarchical and network processors scarcely understand. Neil Larson 1/16/88 FILE30 44 Rincon Rd., Kensington, CA 94707 Copyright MaxThink 1988 -- Call 415-428-0104 for permission to reprint