ONCE upon a time, researchers dreamed of building a robot storyteller. It would regale young children at a level to help them learn well. The tales wouldn鈥檛 be too hard or easy, but just right.
This dream has come true. The robot is called Tega. It is cute, fluffy and appears to boost language skills in young children. Soon it will be trialled in a dozen classrooms in the US.
Developed by a team at the Massachusetts Institute of Technology (MIT), Tega sounds like a child, animates its body and face while storytelling and illustrates the tales on a screen.
Advertisement
In an eight-week trial, Tega read picture books to 67 children aged from 4 to 6 years in weekly one-on-one meetings lasting an hour.
During the sessions, Tega asked questions to gauge the listener鈥檚 opinion and comprehension, quizzing them on a word鈥檚 meaning or getting them to draw conclusions about a character. For example, asking: 鈥淲hat do you think will happen to the boy?鈥�
Tega also recorded the facial expressions and body positions of the child to see how engaged they were. For instance, leaning in was considered a good sign. After the book was finished, the child retold the story to Tega in their own words.
One group of kids played with a personalised version of the robot, which improved each time it interacted with the child, learning about their language skills from the conversations they had.
鈥淎ll the children who played with Tega the robot ended up with improved vocabularies鈥�
For this group, Tega chose which book to read based on what it thought would be most engaging and informative for the child鈥檚 language level. It did this by comparing the book鈥檚 sentence structures and vocabulary to its knowledge of the words the child was familiar with.
Every second session, Tega would replace several words in the story for these children with more complex synonyms, such as 鈥渃lamour鈥� in place of 鈥渘oise鈥�.
A second group interacted with a non-personalised Tega that remembered the child each week, but chose stories to read at random from its library of 81 picture books. For this group, the story difficulty increased every two weeks. A third group of kids didn鈥檛 interact with Tega at all.
Several weeks after the last session, the team found that all the children who played with Tega had improved vocabularies, but the personalised group learned the most words. Their error rate on a vocabulary test dropped by 23 per cent, almost double that of the non-personalised group. In the control group, there was almost no change.
The stories that the children told back to the personalised robot were also longer and more complex, says team member Hae Won Park. And the group showed the most engagement, as measured by body posture and other signs.
鈥淲e鈥檙e now trying to build a mechanism where the robot can ask questions more cleverly,鈥� says Park. The researchers want Tega to perk up children when it senses they are disengaged, or test for understanding on a concept they may be struggling with.
The team presented the work at the AAAI Conference on Artificial Intelligence in Hawaii.
Personalised robot teachers could be a boon for resource-strapped schools, as an effective teaching assistant in group scenarios or for students with special needs, says Cynthia Breazeal, who leads the MIT team.
The researchers will pilot Tega in schools in Atlanta and Boston, hoping to address early gaps in ability. 鈥淚f a child doesn鈥檛 start kindergarten ready to learn, it is very difficult and very expensive for them to catch up,鈥� says Breazeal.
