The Google Stalker

EPrécis is an API, or an application programmer interface—that is, a set of tools that helps a computer’s operating system better carry out low-level tasks.  EPrécis can be employed as a plug-in application by any other computer application that uses text—for example, with Microsoft Word documents, PDF files, QuarkXPress projects—to automatically prepare abstracts, executive summaries, or back-of-the-book indexes.

Creating and categorizing abstracts remains ePrécis’s chief use. Many text-based computer applications contain abstracting engines, so the fact that ePrécis can carry out these tasks isn’t terribly noteworthy. It’s how it does them that’s remarkable. EPrécis doesn’t just match strings of letters to other strings of letters. It analyzes words within the context of the sentence structures that contain them, taking into account issues of usage, grammar, and emphasis that would bewilder other search tools.

Very simply, here’s how it works. EPrécis first assigns a semantic weight to a word with the help of one of the two dictionary programs built into it. (Schultz prefers the term “semantic quotient” to “weight,” but weight is one way to understand what he and ePrécis are up to.) One dictionary filters out short, common words such as articles (a, an, the), leaving the rest of the words to be analyzed by the larger dictionary. That dictionary assigns a series of codes to the words based on their definition and how much information they contain.

EPrécis then assigns each word two numbers that determine its weight. One is based on its “level” of meaningfulness, with 7 being the least meaningful, and 1 the most. The other classifies the word based on how “tangible” or “intangible” it is. In Schultz’s system, this isn’t as clear a distinction as it might appear. By and large, a “tangible” word not only refers to something concrete and touchable (a cement mixer, say) but also “things” such as art and poetry. An “intangible” term refers to conceptual notions from business and law, such as profit and contractual agreements. In general, the more common the word’s usage, the less numeric weight it carries.

After the weight of a word is “calculated,” the program determines the word’s place in the text’s structure to ascertain its overall relevance. EPrécis then compares the weight of all the terms within a text and generates a coherent summary.

If you think that sounds like the work of something beyond the black-and-white, binary mind of a computer, you’re not alone. “It’s a rotated-concept index,” says Schultz, who resembles a more demonstrative and disheveled Carl Pohlad. “It takes the key indexable terms and rotates them to the front of the sentence, and then rewrites the sentence.”

Schultz, whose rambling discourse roams freely across a wide variety of topics—Karl Marx’s economic theories, the novels of German Nobel laureates Hermann Hesse and Thomas Mann, the “mind-blowing” opening verses of the Gospel of St. John—grows animated when talking about the underlying elements of language, casually citing ideas such as Noam Chomsky’s theories of transformational grammar as if they were See Johnny run. “When we converse with each other, we’re operating on a surface structure of language,” he says. “But below that surface, in what’s called the semantic base, there’s a kernel structure that goes through a series of transformations until it gets to the surface. And that’s where you find the whole sentence.”

Follow that? Think of it this way: EPrécis is not simply converting words to numbers, but trying to account for the way words interact, and how their meanings change depending on context. “Until you get underneath the surface and understand what goes on there, you’re not going to be able to communicate with a computer,” Schultz says. “You have to go further and further down until you can really grab those underlying syntactic structures.”

Perhaps the best way to understand all this is by contrasting a Google search with one performed via ePrécis. Search “beef barley soup” on Google, and you’ll get 553,000 hits. Underneath each result are two lines of text that usually contain little useful or relevant information. To follow up on the sites that might contain soup recipes, you must click on link after link after link to hunt down what you’re looking for. Searching on the same phrase via ePrécis brings up a page containing links to a dozen recipes whose pertinent details you can quickly scan, thanks to the four-line summary that appears with each link, before you decide to click on them.

Another distinction: EPrécis is the fruit of analysis and tabulation done over a period of many years not by a machine, but by a human being with a brain, a heart, and apparently endless reserves of patience and determination.