Rock by Numbers

Chicago Reader | July 20, 2004
The Internet is probably the best thing that's ever happened to the music geeks of the world. As easy as computers make it to ferret out new bands and trade songs, though, they're no closer to eliminating the subjectivity from fandom--an online record review is only as comprehensible and trustworthy as the critic who wrote it. If anything, the enormous number of competing voices on the Internet makes it harder than ever to be sure you've found good advice.

Loren Wilson wants to know if there's any rhyme or reason to the music writing on the Web--and he's taken a particular interest in Chicago-based indie-rock forum By day a computer network specialist at the University of Chicago, by night the 25-year-old plays guitar in the band Starlister and helps run Mr. Hyde Records, a label he hopes will catalyze the music scene in Hyde Park. He's also a BA student in interdisciplinary studies at the U. of C., and though he's on a leave of absence from school, as part of his final project he's written a series of computer programs to analyze the language of the record reviews on Pitchfork. On June 10 he posted the results at, including two MP3s of tunes he wrote to fit "a set of compositional guidelines based on the musical preferences" the critics expressed. Wilson's site was featured on Slashdot on June 16, and within 48 hours he'd logged 10,000 downloads of the songs. He's been blogged in at least seven non-English-speaking countries, and on July 1 a blurb on the project ran in the New York Times. So far he's counted more than 21,000 total visitors.

Founded in 1995 and originally run out of a bedroom, Pitchfork now posts five reviews and four or more news stories every day and claims to receive more than 500,000 unique visitors per month; revenue comes from ad sales, mostly to indie labels, and contributors are paid a nominal amount. Much of the writing on Pitchfork is undeniably passionate and informed, but like any successful indie institution it's attracted its share of sniping. Pitchfork is snarky, the complaints go; it's arbitrary, it's elitist, it's lily-white. On Sub Pop's Web site there's a parody page called "Popdork" that includes "news" headlines like "Indie cred flawlessly maintained" and "Pitchfork staffer says 'Hi' to real life woman." And's satire "Rich Dork" takes a poke at Pitchfork's alleged Radiohead fixation ("Radiohead nominated Time Magazine's 'Most Influential Anything Ever' by us").

It certainly looks like Wilson, too, is making fun of the site, but he claims he used Pitchfork because all of its roughly 5,600 record reviews were accessible online and accompanied by a numerical rating. "If I could have chosen Rolling Stone instead, I would have done that." On his site he maintains that straight face, explaining that he undertook the Pitchformula project to explore alternate ways of integrating computers into music making, in this case transforming rock criticism into a creative tool. (He also admits he's "just trying to graduate from college.") But the project has given him a framework in which to pose hard questions about music writing: "Is everybody in pop music criticism speaking the same language? Are pop critics doing a good job?"

Wilson's hardly the first to try to bring a little objectivity to the business of art. Recent Princeton grad Katherine Milkman was featured in the New York Times on June 1 for her own BA project, in which she used computer-driven statistical analysis to look for patterns among the short stories accepted by the New Yorker between 1992 and 2001. (Her conclusions confirmed a few suspicions in the literary community: after a change in the magazine's fiction editorship in 1995, there were more stories set in the New York area, more first-person narratives, and more works by men.) In the mid-90s, the Russian conceptual-art team of Komar and Melamid made waves by creating "Most Wanted" paintings for 14 different countries, which combined the elements that scored best in national market-research polls. (Denmark's "Most Wanted" is an impressionistic seaside pastoral that includes a pair of pirouetting ballerinas and a man in slacks who's planting the Danish colors like a golf caddy replacing a flagstick.) A follow-up project led to a CD of similarly constructed music--according to Komar and Melamid's Web site, "The Most Unwanted Song" will be enjoyed by "fewer than 200 individuals of the world's total population."

But a Spanish company called Polyphonic HMI has made what's probably the most disturbing attempt to apply market research or statistics to the artistic process. Employing proprietary "Hit Song Science" technology, the service analyzes the potential commercial viability of pop songs for the recording industry--according to Polyphonic's PR, this process identifies the "underlying mathematical patterns in unreleased music and compares them to the patterns in recent hit songs."

Wilson's project isn't nearly so mercenary--all it's brought him so far has been attention. A self-described "humanities guy," he enlisted an adviser and researched database theory for the better part of a year, hoping to give Pitchformula "some weight on the computer-science side of things." He applied content analysis (a computer-driven technique for compiling statistics about large texts) to the entire body of reviews on Pitchfork, creating a database of frequently used words. (The custom-designed scripts he used to build and manipulate this database are all on his Web site, in case anyone else wants to perform a similar experiment.)

He then grouped the most commonly used words--basic nouns like "melody" as well as slippery all-purpose adjectives like "ethereal" and "vibrant"--into five families: instruments, sound, structure, mood, and vocals. Each word was assigned a positive or negative score depending on how often it showed up in positive or negative reviews; the more positive the reviews, the higher the score. (Conveniently enough, Pitchfork has provided a plain-language key for interpreting its ratings--an 8.3 corresponds to "strong," while 3.7 means "definitely below average, but a few redeeming qualities.") Wilson also took pains to correct for context-related error--for instance, to determine whether "clouds" was used as a verb or a noun (or to make sure the phrase "this record is not shitty" wasn't mistaken for a true usage of the word "shitty"). To make database processing simpler, Wilson expressed Pitchfork's ratings as percentages, multiplying them all by ten and producing a total average rating of 67.4 for all the reviews on the site--if you think of Pitchfork as a classroom, the bands as students, and the albums as exams, that's a high D ("has its moments, but isn't strong").

So how do you get an A from a Pitchfork writer? Wilson's data suggest that they're really into "drones" (+4003.46) but not "repetition" (-107.88). "Reverb" beats "silence" by a nose (+1261.12 to +1242.36). Anything distorted, melancholy, or explosive rates high, and "wailing" (+779.08) is much better than "yelling" (-675.20). Of course, you shouldn't be "formulaic" (-1539.16) or "predictable" (-1397.99), and you'll also want to sound less "polished" (-56.84). While you're at it, work on those "lyrics" (-3141.48), and don't trade in your "guitar" (+17833.85) for a "sampler" (+2917.08) just yet. By all means try to "groove" (+3388.23).

Wilson also discovered that female pronouns scored much lower than male pronouns--"his" and "he" are associated with 18 times more rating points than "she" and "her." (And this doesn't just mean that far more men than women play on the records Pitchfork covers--female pronouns are more likely than male pronouns to show up in negative reviews.) Albums deemed substandard are more likely to be evaluated in terms that refer to the intelligence of the band or its fans ("retarded," "vapid," "frat") or reflect associations with marketing or business ("sell" scores a -1432.56). Critics use more concrete language when talking about music they like--in fact, they're "3.8 times more likely to use a meaningless value judgment word" when panning something. Wilson wonders if a related phenomenon explains the very negative score attached to "lyrics"--perhaps writers take the trouble to quote from lyrics they like, rather than using the general term.

Wilson doesn't care to draw conclusions about pop-music criticism at large based on this project, but he does believe that Pitchfork critics are indeed "speaking the same language": the repetition of certain words on Pitchfork, combined with the relatively predictable association of these words with either good or bad reviews, creates an internally consistent aesthetic. Perhaps too consistent--Wilson finds it difficult to read Pitchfork now, after studying its language so closely. "All those 'critic's words' jump off the page, left and right," he says. "It's totally distracting." (His Pitchfilter program, which highlights positive and negative words on Pitchfork's site in green or red, is posted on the "bonus features" page.)

The songs Wilson wrote by keying on these "critic's words" show their seams a bit, like Frankenstein's monster--and his online commentary encourages us to think of them as the patched-together results of an experiment. "Sad and dark moods are preferred in the word lists," he writes, "so I had a couple of brainstorming sessions and made a list of depressing past experiences." Even the performances were subject to the tyranny of his method: "I tried to sound delicate and avoid screaming; even though I thought that screaming might fit at times, the word 'screaming' gets a very low score." Despite the artificiality of Wilson's process, the tunes aren't too bad. "Kissing God" features a stuttering drum loop reminiscent of the Notwist, a swaying wall of cello and violin in the bridge ("unpredictable"), and shockingly loud choruses loaded with distorted, dissonant guitar ("dynamic shifts").

Online reaction to Wilson's music has been mixed. A visitor to the Saddle Creek message boards pronounced it "pretty good," but the consensus at seems to be that it's "shitty." One poster to the I Love Music discussion forum complained that it's "just emo"--apparently expecting Wilson's method to have produced songs so preternaturally excellent that they'd transcend all previously established genres. Another visitor cracked, "Finally, scientific evidence that Pitchfork's ideal music sounds like the Postal Service."

The project as a whole has provoked an equally wide range of responses. At the message board for WXRT's Sound Opinions, a poster (apparently a Pitchfork staffer) declared it an "admirable undertaking," but visitors at called Wilson a "lunatic" and his work a "silly enterprise." Someone at named him "the least creative human being ever not to be a member of the Beautiful South." It's fair to say Wilson has touched a nerve in the online indie community--as of July 2, a Google search turned up 573 hits for "pitchformula," and discussions about the project have sprung up in at least 50 different forums. Many of the threads spiral outward into larger debates about art criticism, statistics, and the business of popular music.

A few commentators have accused Wilson of cooking up the whole project just to get his music out into the world. He begs to differ. "I mean, if that was the case, I could have gotten a second job for all the time I spent on this," says Wilson, "and then used the money to hire someone to promote me." Fewer than 600 Pitchformula visitors have continued over to the Starlister site to check out Wilson's "real" songs.

The project seems to have benefited him in other ways, though. The Pitchformula tunes are the first Wilson has written in a year--he says he needed the project to push himself to rethink the way he's approached music so far. "Because otherwise," he says, "things get stagnant and you get bored and you stop." He's also talking about developing a public online forum for computational studies of popular music, using the same kind of home-brew open-source code that Pitchformula did.

His online notoriety, if it lasts, will certainly help him launch that venture. He's heard that the folks at Pitchfork have read his study, and that even they were impressed by his obsessive work. "I got invited to the Pitchfork staff party, but I couldn't go," he says. "My band had a show that night."

Chicago Reader

Chicago Reader, a pioneer of the weekly alternative press, is one of the largest and most successful AAN publications, recognized throughout the industry for both editorial excellence and advertising impact. It publishes an average of 100 pages per week in...
More »
Contact for Reprint Rights
  • Market Served: Metropolitan Area
  • Address: 350 North Orleans Street, 10th Floor, Chicago, IL 60654
  • Phone: (312) 222-6920