Nearly seven years ago, as Chris Yavelow sat down to write his first novel, he did the usual things. He took "a bunch of classes," joined some creative-writing groups and started writing. Then something unusual happened.
"I started having questions, like, Should I use the word 'seemingly'?" he says. "I was reading Congo by Michael Crichton. I wondered, Did Crichton use 'seemingly'?"
To find out, Yavelow scanned Congo into his computer and did a word search. The Congo scan led Yavelow, a computer expert who looks like a cross between Gene Wilder and Larry from the Three Stooges, to other questions about how to turn a phrase. And he wondered if certain turns of phrase would turn a buck. He hypothesized that all successful books had something in common – something that, if analyzed, could teach him to bang out best sellers like Crichton.
Surrounded in his mother's basement by books on writing and right-wing polemics, Yavelow started scanning popular novels into his computer, and he began writing a computer program to scan for patterns. In October 2005 he launched the result, FictionFixer (www.fictionfixer.com), a computer program and analysis service that, for about $380, Yavelow says, will help turn anyone's novel into a potential best seller.
"I'm in the middle of my second novel," he says. "I feel like the first one will be a best seller because of my use of FictionFixer."
That first novel is still unpublished.
FictionFixer compares any prose it is given to any or all of 210 best-selling novels Yavelow says he has scanned into his Mac G4. The software looks for patterns – certain words, usages, sentence structure, number of mentions of a character's name, and the like. Like a grossly swollen and complicated spelling/grammar check on a word processor, the program then prints a report with specific suggestions for changes to make the work being analyzed more like the best sellers.
Yavelow doesn't like the comparison to existing word-processing software. "When was the last time your word-processor advised you to bring in the antagonist earlier, between 2,832 and 4,218 words earlier," he asks on his website, "or to change the attributions in 259 dialog sentences; or to add at least 14 sensory triggers per thousand words; or to rewrite the openings of 436 narrative (non-dialog) sentences, telling the specific sentences and precise problems?"
Yavelow – whose CV boasts six degrees in music (including one from Harvard University), four operas and the invention of a music-composing machine that polled listeners' tastes and modified the melody accordingly – uses the pop-music industry as his model. "I realized I would never make any money as an opera composer," he says. "I could see that the Top 40 was molding the public." Like Yavelow's music machine, the music industry operates as a feedback loop, in which successful sounds and styles beget imitation. The money flows to those who stick to the formulas. Yavelow thinks the publishing industry should emulate that.
"The writing industry's track record, compared to the music industry`'s`, stinks," he says. "People get fired in the music industry if you `spend too much money to promote` something that ends up in the remainder bin at 70 percent off."
But as a very limited test we performed demonstrates, music yields to mathematical modeling much more readily than English prose.
Yavelow agreed to take some prose that we supplied and pump it through his machine. But the short sample caused some trouble. It didn't use quotation marks to set off characters' dialogue, for example, which caused FictionFixer to conclude falsely that the sample didn't contain enough dialogue sentences. The software also incorrectly deduced which character is the story's protagonist.
Yavelow had FictionFixer compare the sample to Dan Brown's The Da Vinci Code, and despite its limitations, the Fixer made some salient observations: "Told" and "called" are the top two attributions in both works. The sample uses "cried" much more than Brown. And the sample uses 14 times as many passive sentence constructions. (The sample text was the first six pages of James Frey's My Friend Leonard, with a few pages from a JT LeRoy short story grafted on when Yavelow said the software needed more text.)
Perhaps the Fixer's most useful service is its count of sensory descriptive phrases – words evoking touch, sight, smell, taste and sound. Writing coaches everywhere urge prose stylists to use more of these, and the successful writers usually do. But the computer program is imprecise.
FictionFixer listed Frey's use of the word "crack" among its descriptive verbs, and Yavelow praised it as a "good, sound word" before being told that, in Frey's work, "crack" is usually a noun meaning rock cocaine.
Yavelow deflects questions about FictionFixer's sales history so far. "That's confidential information," he says, but adds quickly, "There haven't been as many `paying customers` as I would have liked." He also says that he hasn't yet done much to promote the service.
Yavelow says he knows no computer program can write a novel – not yet, anyway. And his invention isn't supposed to homogenize the industry. "Then we'd all write Harry Potter over and over again with different names and places." No, FictionFixer is meant as a tool, something to define the boundaries of the salable novel for the serious, profit-minded writer – much the way music producers like Quincy Jones and Phil Spector honed pop music acts to fit within the bounds of the popular sound.
"This is not to make a formula for a good novel," Yavelow says, "but to define the space in which all good novels that communicate to people – that people enjoy – reside." The program and his service will help "pull your own novel into that space by sanding down the edges."
Yavelow says he's comparing his own novels to a single successful work. "I'm emulating a certain writer," he says. "But you would never know it."
And what about that Crichton book and the question that started it all? Did Crichton use "seemingly"?
Yavelow can't recall exactly. "I don't think he did," he says.
[email protected]
