Even if an artificial intelligence could pass the bar exam, pass a medical exam, or read a bedtime story with emotion, without first mastering the art of sarcasm, it would be a wonder of the human spirit that will never be matched.
But that art may come next on technology’s dizzying list of capabilities. Researchers in the Netherlands have built an AI-powered sarcasm detector that can identify when people are showing their lowest wit and highest intelligence.
“We can recognize sarcasm in a reliable way and we want to develop it further,” said Matt Koller from the University of Groningen’s Speech Technology Laboratory. “We want to see how far we can push it.”
There’s a lot more to this project than teaching algorithms, and sometimes even the most passionate comments can’t be taken literally, but rather have to be interpreted as the opposite. Koller said sarcasm permeates our discourse more than we realize, and understanding it is important for humans and machines to communicate seamlessly.
“When you start studying sarcasm, you start to understand the extent to which we use it as part of our regular communication methods,” Kohler says. “But we need to talk to the device in a very literal way, like we’re talking to a robot. Because we’re talking to a robot. It doesn’t have to be this way.”
Humans are generally good at spotting sarcasm, but there are only so many clues we can pick up from text alone, so when speaking, tone, and facial expressions all reveal the speaker’s intent, it’s better than face-to-face interaction. It becomes difficult. In developing AI, researchers have discovered that multiple cues are also important for algorithms to distinguish between sarcasm and sincerity.
In research presented Thursday at a joint meeting of the Acoustical Society of America and the Acoustical Society of Canada in Ottawa, Xiyuan Gao, a doctoral student in the lab, said the group has described how he trained a neural network on emotional content. Clips from American sitcoms such as Friends and The Big Bang Theory. database, known as mustard was compiled by researchers in the United States and Singapore, who built a unique detector by labeling sentences from television programs for sarcasm.
One of the scenes the AI was trained on is: Leonard’s wasted effort To escape from the locked room in The Big Bang Theory, Sheldon says, “It’s just a privilege to be able to observe how your mind works.” In another episode of Friends, Ross invites Rachel to come over to his house and assemble furniture with Joey and Chandler, but Chandler commented, “Yes, I’m really looking forward to it.”
By training on text and audio, along with scores reflecting the emotional content of words spoken by actors, the AI was able to detect sarcasm in unlabeled interactions on sitcoms almost 75% of the time. Further research is being conducted in the lab using synthetic data to further improve accuracy, but the work is awaiting publication.
In addition to making conversations with AI assistants more fluid, Shekhar Nayak, another researcher on the project, hopes to apply the same approach to detecting negative tones in language and detecting abuse and hate speech. He said it could be used.
Gao said further improvements could be made by adding visual cues such as eyebrow movements or fake smiles to the AI’s training data. The question arises: how accurate is enough? “Is it possible to have a machine that is 100% accurate?” Gao said. “That’s not something even humans can achieve.”
By making the program more similar to how humans actually speak, humans should be able to converse with devices more naturally, Koller added. But what if, he added, machines adopted our newly acquired skills and started throwing cynicism at us? “If I ask, ‘Do you have time to ask a question?’ And it says, ‘Sure,’ I might think, do I or don’t I?”
Source: www.theguardian.com
