Researchers Use Machine Learning to Deconstruct Satire, Satire Approves

In a paper presented without irony at the AAAI Conference on Artificial Intelligence, researchers managed to deconstruct The Onion's headline, a staple of the satirical news site, to isolate the satire by feeding over a thousand headlines from the publication and closely-related, but serious, headlines to reverse engineer what it is that makes the headline funny.

A Serious Effort, All Kidding Aside

As a refined function of human cognition, humor is widely seen as a quintessentially human trait, something beyond simple play that relies on the relationships between complicated verbal and physical social cues to trigger a euphoric release of laughter.

No one has mastered this art form more than the writers of the headlines for The Onion, the satirical American publication that uses the conventions of serious, wire-service style news reporting to brilliant effect in satirizing current events.

[see-also]

Sometimes, The Onion headline itself is enough to fully convey the joke. Taking this as the starting point, researchers Robert West, a computer scientist at École Polytechnique Fédérale de Lausanne in Switzerland and Eric Horvitz, the director of Microsoft Research in Redmond, Washington, compiled over 1,000 The Onion headlines to analyze.

The researchers made an online game, Unfun.me, where a user tries to take a humorous headline from The Onion and turn it into a serious headline in as few words as possible.

West and Horvitz were able to compile about 2,800 serious, humorless alternatives to the original headlines in The Onion that they had selected. The edits made through Unfun.me, “put a finger onto the exact switch that induces the humor,” West said.

They ran these through a machine learning algorithm to try to identify a pattern in the changes made to produce the serious headlines that would explain what it is that makes The Onion headlines funny in the first place.

Deconstructing The Onion's Headlines

What the researchers found after deconstructing The Onion's headlines is that they follow a shared logical structure which is the mechanism that makes the headline funny.

The researchers call this mechanism the “false analogy.” As with normal analogies, the words switched share an important connection, but are nevertheless radically different in meaning.

An example cited [PDF] in the study, the original headline “Bush Picks Laser Background for Presidential Portrait,” and the edited one, “Bush Picks Rural Background for Presidential Portrait,” demonstrates this idea of a false analogy.

In this case, the background for the portrait connects the two words—the laser-background that every school kid in the 1980s and 1990s will be familiar with and a rural setting you might use for a more serious portraits—but laser backgrounds are the relics of childhood, which implicates the character and intelligence of the butt of the joke, former President George W. Bush.

Other Findings

Besides the false analogy, the switched words often fell into a relationship of opposition, which the researchers break down into several subcategories, such as High-Low stature, Good-Bad intentions, Reasonable-Absurd response, and similar pairings.

They also identified that the final block of a sentence was the most likely to be edited, indicating that headlines actually have a narrative structure embedded within them, with the final few words serving as the headline’s "micro-punchline."

The researchers hope that these insights might lay the groundwork for computer-generated humor, or even aiding an artificial intelligence in filtering out fake news from real reporting.