5/04/2005

Automating ACW writing

Struggling through James McPherson's carefully hedged nonsense yesterday while doing Jean Smith plagiarism analysis, I asked myself "Where have I seen this kind of writing before?"

In style and tone it matches, SCIgen, a famous and very good Internet joke.

SCIgen is a program that churns out technology papers. You plug in the names of some authors, and SCIgen's engine spits out carefully hedged insights in the blink of an eye. Some pranksters have successfully submitted these papers to conferences.

Here's a paper I generated under the name of James M. McPherson; we've made him a computer scientist. "Dr. McPherson," SCIgen version, writes:

Unified wearable methodologies have led to many key advances, including kernels and replication. Even though prior solutions to this problem are encouraging, none have taken the pseudorandom approach we propose in this paper. To put this in perspective, consider the fact that famous analysts largely use von Neumann machines to solve this question. However, suffix trees alone can fulfill the need for signed models.

Dr. McPherson, acclaimed Civil War writer, actually penned this:

It is hard to separate fact from fiction in this matter. Many wartime stories of Grant's drunkenness are false; others are at best dubious. Grant's meteoric rise to fame provoked jealousy in the hearts of men who indulged in gossip to denigrate him. Subject to sick headaches brought on by strain and loss of sleep, Grant sometimes acted unwell in a manner to give observers the impression that he had been drinking. But even when the myths have been stripped away, a hard substratum of truth about Grant's drinking remains.

Eerie, is it not? A random and meaningless string of assertions that somehow hang together. The paragraphs also end similarly. "But even when the myths have been stripped away, a hard substratum of truth about Grant's drinking remains" = "However, suffix trees alone can fulfill the need for signed models."

If the publishing industry gets hold of SCIgen and makes a few changes for Civil War nonfiction, there won't be enough Pulitzers in all creation to award to the robotic prizewinners.