The Endless Quest To Understand Sarcasm On The Internet

Sarcasm is often misinterpreted on the internet, and there are no shortage of technologies trying to fix that

Illustration: Tara Jacoby
Sep 03, 2016 at 8:00 AM ET

For as long as people have been chatting online, sarcasm has been a breeding ground for miscommunication. Unless you’re finely attuned to someone’s sensibilities, it’s almost impossible to detect sarcasm without some sort of implicit warning in advance. As a result, conversational mishaps as a result of dry humor are far too frequent an occurrence, and trolls have learned to use it to their advantage many times over.

In 2015, linguist Gretchen McCullough explored the various ways we attempt to translate those sarcastic eye rolls into words on the internet through the use of hashtags and asterisks, deliberately misspelled words, or even the faux coding expressions, like “/sarcasm.” And we’ve all employed our fair share of smiley faces and LOLs to indicate offsetting a harsh line of text with a winky counterbalance.

But the problem of misinterpreted sarcasm still persists, and because of this, a group of researchers at the University of Turin say they’ve developed a machine learning algorithm that can analyze internet banter and detect sarcasm with accuracy that ranges between 80 and 89 percent.


More What Your Instagram Filter Use Says About Your Mental Health

This isn’t the first attempt at sussing out a facetious statement, however (and it probably won’t be the last). For over a decade, numerous companies and institutions have attempted to remedy what has seemed an unsolvable problem. In 2006, the Signal Analysis and Interpretation Laboratory at the University of Southern California came up with an automatic sarcasm detector using recordings of human statements that examined pitch, duration, and energy. In 2011, researchers at Hebrew University in Jerusalem developed a sarcasm detection method analyzing 6 million tweets and over 60,000 Amazon reviews which it deemed to be 91 percent accurate.

At the beginning of 2016, Cornell students participated in a tech challenge to identify sarcasm in product reviews, and came up with a software detection system that hit 71 percent accuracy. And this past August, Silvio Amir at the University of Lisbon developed a method (that he says was 87 percent accurate) for detecting sarcasm on Twitter in response to a request from the Secret Service.

This research might all seem frivolous, but when sarcasm is misunderstood or taken out of context, it’s often interpreted as an act of hostility, latent anger, or insecurity, which can compound other issues within a social dynamic.

Even before the internet, linguists sought out solutions to aid us in picking up on it in writing. Ideas for various punctuation to indicate when someone is having you on — from the backwards question mark, to the upside down exclamation point — have their roots in the 16th and 17th centuries (and were later revisited in 2004).

And it’s clear that sarcasm isn’t waning as a mode of expression on the internet despite the challenges in communication it poses — there are any number of memes about the sarcasm that celebrate sarcasm’s potential for verbal chaos as they do criticize it — so why not understand it on a semantic level as much as possible?

Newer research even suggests sarcasm actually has its benefits, like strengthening bonds and boosting creativity if the two people engaging in it trust each other. (If they don’t, it can still make for a hell of an entertaining exchange nonetheless.) But while most people start to understand sarcasm intuitively around age 6, it’s not exactly in everyone’s wheelhouse: Redditors have earnestly asked fellow Redditors how to tell when someone is being sarcastic.

But even if computers can figure out sarcasm better than we can, then what? Scientific American notes that the bulk of the research into online sarcasm usage — and studies indicate we use way more of it online than in person — goes beyond novelty, and could actually have a number of potentially helpful scenarios.

It could help everyone from law enforcement agencies, who can better figure out if an online threat is real; it can assist customer service centers, who can better determine whether you’re really upset or just trolling; or it could even aid political figures in figuring out if their message is really landing. Other researchers note that such data could help track our emotions, predict stock market trends, or improve services like Apple’s Siri so she knows when you’re just horsing around.

But it’s just as likely that marketers and corporations would want to get their hands on this research so they can tell whether we really like them or are just mocking them, or make us think they’re cool. All so we will trust them more and give them even more of our money. And isn’t that the coolest use of sarcasm of all?