Koji Inoue has a lot to tell. He sat down at the table, opposite a young woman. "I urgently need to talk to you," he begins, "it's the soccer World Cup right now, isn't it?" "Yes," the woman replies, raising her eyes expectantly. Inoue continues breathlessly: "Japan played against Germany. Now imagine: Japan won!" He's beaming.
In that second, something happens that computer scientist Ionue sees as tantamount to a revolution: The woman laughs and replies: "Wow! That's nice!" She is an android, a humanoid robot.
Situational laughter
"Erica" is her name. And recently she can do something that other robots can't: laugh. She does it depending on the situation, i.e. only when it makes sense in the conversation. It took the team at Kyoto University three years to teach Erica this skill. Humor can be a serious challenge.
Erica can best judge for herself. "Do you find it difficult to understand human humor?" The question goes straight to the android. Erica blinks her eyes, gently rocks her head back and forth. "Yes, yes," she replies, her expression reflecting a serious effort. "What people perceive as humor is different for everyone." For Erica it means decision-making processes in a very short time: "It's really difficult."
Pick up keywords using speech recognition
Erica has been wowing audiences in Japan for a long time. In 2015 she was presented – at that time without laughing competence. Speech recognition software enables it to pick up key words in conversations and provide appropriate responses. This works best if your programmers know the topic of conversation beforehand. Fascinating, but far from authentic entertainment.
"Laughing is essential for a warm, human conversation," says Inoue. In order to teach Erica that, the assistant professor first had to see through this phenomenon himself. He invited more than 80 students to speed-date with Erica. The conversations were examined for when and how the students laughed at Erica. Hundreds of bursts of laughter have been categorized.
Three Kinds of Laughter
From this, Inoue and his team filtered out three options for Erica: First, laughing happily at a joke – usually recognizable by the fact that the originator himself laughs the loudest. Second, the variant for advanced learners: "social" laughter. They are restrained sounds that signal approval or conceal embarrassment. Erica should reply. Third, the silent option, which is always in demand when laughter is more likely to provoke irritation. So they built a laughter algorithm.
"It was a lot of trial and error work," Inoue sums up today. Even though Erica looks more like a mannequin and her speech sounds like it's straight out of a tin can, it's hard to resist the appeal of a conversation with her. "Of course, in your head you know that you are talking to a robot," admits Inoue. "But when she laughs, it gets exciting. It's like talking to a person."
Laughter is silver – silence is golden
The conversation about football shows how sensitive Erika has become. She acknowledges Japan's victory over Germany with a happy laugh. When Inoue remarks, embarrassed, that the situation is uncomfortable because there are guests from Germany present – Erika only smiles mischievously and ends the exchange with a diplomatic "hm". Depending on the situation, silence is golden even for a robot.
Leave a comment