Japan has made plenty of splashes in the robotics industry with its incredibly humanoid androids (think Yangyang). But despite how much robots can look like us, they can’t communicate like us, right?
Wrong.
Georgia Tech researcher, Dr. Crystal Chao interacts with Simon, a robot able to simulate extroverted and introverted speech patterns. (Image courtesy Georgia Tech.)
We humans typically converse in a very fluid manner. People often interrupt each other, shifting topics quickly. Tone of voice and complex body language both heavily influence perceived meanings behind words and phrases.
Conversely, human-to-robot conversations are quite uniform and much more structured.
For example, conversations with AI like Siri happen in a strict exchange where users voice a command, wait for Siri to register the command and then receive Siri’s response. Humans need to speak to robots slowly, clearly and coherently to avoid error.
To make human-robot interaction feel more natural, Georgia Tech researchers Chrystal Chao and Andrea Thomaz developed a model using the CADENCE engineering software application. With their model, Simon can operate within the parameters of the “conversational floor,” in which individuals take turns speaking using speech patterns and other cues to structure conversation.
Chao’s inspiration for the experiment stems from conversations with another group of researchers at Georgia Tech working on a computational system for improvisational theater.
“We were all investigating how dominance is expressed through cues in interaction,” Dr. Chao said in an article by IEEE Spectrum. “Some of these are nonverbal cues like body posture, but a large part of conversational dominance is how much a participant seizes and holds on to the speaking floor. A participant becomes more or less dominant based on how often she interrupts herself or others, how long her turn is, how long she waits to take a turn, etc. These cues have also been receiving more public attention as women strive to achieve equality in the workplace through how they communicate.”
Simon can simulate personality by speaking in one of two sets of speech parameters: “active” or “passive.”
Set to its active parameters, Simon seems more extroverted. It speaks for longer, at louder volumes and with an increased likelihood of interrupting others. Set to passive parameters, Simon is much more quiet. It speaks in shorter bursts, interrupting itself to allow other speakers to interject and making eye contact with humans before taking its turn to speak.
The video below depicts Simon using both parameters while speaking gibberish as humans try to teach it about objects.
“When the robot was active, people tended to respond and give feedback to whatever the robot was doing,” Chao said. “When the robot was more passive, people felt obligated to take more initiative. They taught the robot about the objects or told stories about them.”
This kind of technological advancement in the field of robotics could potentially bring fictional robots like the quirky Star Wars droid C-3PO to reality in the future. Voice-based assistants could be more personable and robots in service fields could become more human and trustworthy – if it doesn’t make them creepier.
Chao recognizes the downsides to “outgoing” personality types in robots. They could come across as stand-offish, which is counterproductive for robots who may be used in service industries.
“We expect that when the robot is more active and takes more turns, it will be perceived as more extroverted and socially engaging,” Chao explained. “When it’s extremely active, the robot actually acts very egocentric, like it doesn’t care at all that the speaking partner is there and is less engaging. This is why a balance of being active or passive is really needed.”
However, to efficiently emulate personality, advancements beyond speech patterns are necessary. Significant meaning in social interaction is articulated not through words, but body language.
With the human-like appearance of Yangyang and the spoken personality of Simon, simulated body language may be in the very near future.
Chao and Thomaz’s findings were published in the Journal of Human-Robot Interaction and can be read here.