Why Computers Don't Understand Sarcasm On 'TechStuff'


On this episode of TechStuff, host Jonathan Strickland wonders if we’ll ever build a supercomputer that can truly understand sarcasm. Many times, even humans don’t pick up on sarcasm, especially when it comes to written communication on the Internet, because we’re denied tone, body language, and other cues that help us pick up on it. In fact, a lot of people have started tagging posts “#sarcasm” or “/s” in order for their audience to fully understand what they’re trying to get across. And if we, the people programming computers, can’t pick it up, how are the computers going to be able to? Jonathan dives into the basics of computer programming and computer language, and the applications (and limits) of IBM’s Tone Analyzer software, to find out.

First, the basics: Computer code is the language of computers. At its lowest level, that language is binary, a series of ones and zeroes. “Unless you’re Neo from The Matrix, it means nothing to you,” he says. That binary language tells the computer how to respond to data inputs. Over time, developers have created “high-level” programming language that essentially allows computers to understand our natural way of talking. Alexa or Siri, for example, have high-level computing language that allows the smart assistants to respond to key phrases and answer questions. It also has the capability to tell you when it doesn’t understand you. There’s some impressive complexity there: You can ask, “How hot is it outside?” or, “Is it raining?” and the computer can understand you want information about the weather, even though you used very different words. 

But when it comes to sarcasm, it gets more difficult, because instead of just having to parse what you say, the computer also has to understand what you mean. A Tone Analyzer application, developed for IBM’s Watson, comes fairly close: It analyzes texts, searching for social, emotional, and written cues to figure out the overall tone of the statement. Particular words were pre-labeled as falling into certain categories, such as “happy,” “sad,” “angry,” or “helpful.” The analyzer has a little over a 75% precision rate – not bad for a computer! But when it was fed famously ironic or sarcastic passages from literature, it usually misinterpreted them as being straightforward statements, missing the nuance. Learn more about the difficulty of analyzing tone, and what it might take to get a supercomputer that can handle it, on this episode of TechStuff.

If you want to be sure you're listening to the podcasts everyone else is checking out, iHeartRadio has you covered. Every Monday, iHeartRadio releases a chart showing the most popular podcasts of the week. Stay up to date on what's trending by checking out the chart here. There's even a chart just for radio podcasts here, featuring all your favorite iHeartRadio personalities like Bobby Bones, Elvis Duran, Steve Harvey and dozens of others.

Photo: Getty Images