Even killer Terminators must start as infants and learn to talk before they can ride motorcycles and morph into just about any object they want, and that training has begun this week for the iCub robot.
These tests are part of the iTalk (Integration and Transfer of Action and Language Knowledge in Robots) project, the goal of which is to develop better ways to teach robots to communicate and work with humans (and each other), and develop complex behavioural, cognitive, and linguistic skills.
The cute little iCub robot (of which 5 have been built, with 4 being sent to the project's partners at various Universities), with sterile features and large chibi-eyes, was specifically designed to elicit the feeling that one was speaking to a child when communicating with it.
The iCub has 53 motors to operate its head and limbs, and teaching it to manipulate objects is a future goal of the project as well. The recent tests of the iCub focused on its capability to learn language though, and the results were positive.
The iCub, nicknamed DeeChee, started out rambling and babbling incoherently like a typical young child. As the researchers begin to talk to DeeChee, it slowly begins to pick up on commonly used syllables and words, and can begin repeating them. While DeeChee has no understanding of what the words mean, and therefore cannot string words together into sentences yet, it shows the same ability to pick up language (albeit at a much faster pace) than babies do when they finally utter their first "Daddy!" or "Mama!".
"It is known that infants are sensitive to the frequency of sounds in speech, and these experiments show how this sensitivity can be modelled and contribute to the learning of word forms by a robot" said Dr. Caroline Lyon, one of the researchers who worked with DeeChee for the study.
It is thought that the research can benefit not only the development of future robotics technology (such as Terminators that can ask "Where's John Conner?"), but also further our understanding of early language development. The results of the research appear in the most recent issue of PLoS ONE.
