Never mind that it can pass the bar exam, pass medical tests, and read bedtime stories with emotion, artificial intelligence will never match the wonder of the human mind without first mastering the art of sarcasm.
But that art, it turns out, might be next on the list of the technology’s dizzying capabilities. Researchers in the Netherlands have built an AI-powered sarcasm detector that can see when the lowest form of wit, and the highest form of intelligence, is being deployed.
“We can recognize sarcasm in a reliable way, and we are keen to grow it,” said Matt Coler at the University of Groningen’s speech technology lab. “We want to see how far we can push it.”
There is more to the project than learning algorithms, that sometimes even the most exuberant comments cannot be taken literally and must instead be interpreted as the diametric opposite. Sarcasm permeates our discourse more than we might appreciate, Coler said, so understanding it is essential if humans and machines are to communicate seamlessly.
“When you start studying sarcasm, you become hyperaware of the extent to which we use it as part of our normal mode of communication,” Coler said. “But we have to talk to our devices in a very literal way, like we’re talking to a robot, because we are. It doesn’t have to be that way.”
People are generally skilled at spotting sarcasm, although the limited cues found in text alone make it more difficult than in a face-to-face interaction when delivery, tone, and facial expressions all reveal the speaker’s intent. In developing their AI, the researchers found that several cues are also important for the algorithm to distinguish the sarcastic from the sincere.
In work presented Thursday at a joint meeting of the Acoustical Society of America and the Canadian Acoustical Association in Ottawa, Xiyuan Gao, a PhD student at the lab, described how the group trained a neural network on text , audio and emotional content of video clips from American sitcoms including Friends and The Big Bang Theory. The database, known as Mustardwas compiled by researchers in the US and Singapore, who annotated sentences from the TV shows with sarcasm tags to build their own detector.
One scene the AI was trained on was Leonard’s futile attempt to escape from a locked room in The Big Bang Theory, prompting Sheldon to remark, “It’s just a privilege to watch your mind at work.” Another one from Friends has Ross inviting Rachel to come over and put together some furniture with Joey and Chandler, prompting Chandler to comment, “Yeah, and we’re really excited about it.”
After training on the text and audio, along with scores reflecting the emotional content of words spoken by the actors, the AI was able to detect sarcasm in unmarked exchanges from the sitcoms almost 75% of the time. Further work at the lab used synthetic data to further increase the accuracy, but that research awaits publication.
As well as making conversations with AI assistants more fluid, the same approach could be used to detect negative tones in language and detect abuse and hate speech, said Shekhar Nayak, another researcher on the project.
Gao said that additional improvements could come from adding visual cues to the AI’s training data, such as eyebrow movements and smiles. Which begs the question, how accurate is accurate enough? “Are we going to have a machine that is 100% accurate?” Gao said. “It’s not something that even humans can achieve.”
Making apps more familiar with how people actually speak should help people converse more naturally with devices, Coler adds, but he wonders what will happen if machines embrace their newfound skills and start throwing sarcasm at us. “If I ask: ‘Do you have time for a question?’ And it says: ‘Yes, sure’, I might think: yes or no?”
