Robots that teach one another new words through interaction with their surroundings have been demonstrated by UK researchers.
The robots, created by Angelo Cangelosi and colleagues at Plymouth University, UK, currently exist only as computer simulations. But the researchers say their novel method of communication could someday help real-life robots cooperate when faced with a new challenge. They could also help linguists understand how human languages develop, they say.
The simulated bots each have a humanoid upper-torso and a wheeled chassis. One has the job of explaining to the other how to perform a simple task, like lifting an object using its arms. It communicates verbally and has the capacity to teach new words or phrases by combining words previously learned.
Advertisement
The teacher bot first demonstrates basic actions, like 鈥渃lose left elbow鈥� or 鈥渓ower right shoulder鈥� which the learner picks up by copying. The teacher then repeats the performance while saying words that describe the actions. Then the teacher combines words to describe more complex movements, like 鈥減ush鈥� or 鈥済rab鈥�. The learner uses its understanding of these words to perform the movements.
A video shows the robots at different stages of the language-learning process (Windows Media format, 4.5MB). The teacher bot is always on the left and the learner is on the right.
Listen up
Existing voice recognition systems only have a limited 鈥渄ictionary鈥� of words or phrases, Cangelsi says. 鈥淎 pre-loaded dictionary has no reference to the outside world,鈥� he explains. 鈥淚f the words are grounded in experience it is possible to recombine them to understand new concepts by drawing on that grounding.鈥�
Cangelsi developed the simulated robots with colleagues Emmanouil Hourdakis and Vadim Tikhanoff. He told New 杏吧原创 that the bots 鈥渃an help us understand how natural linguistic systems develop, and develop new technologies for communication between humans and robots.鈥�
The latest versions of the simulated robots have vocabularies including more than 100 words and Cangelsi plans to give real-life robots the same abilities. He says this could be useful for future interplanetary exploration, for example.
Untried skills
鈥淵ou could use purely linguistic instructions to have a robot perform new tasks,鈥� he explains. 鈥淚t would understand them using the basic concepts it already knows.鈥�
Stevan Harnad, who researches artificial intelligence at Southampton University, UK, agrees. 鈥淭he idea that words that can eventually be recombined to describe new things or untried motor skills is promising,鈥� he told New 杏吧原创.
Harnad adds that, although there is a big difference between simulation and application, it should be a feasible development. He suggests that robots could even teach humans how to perform new tasks.
