Developmental Language Learning from Human/Humanoid Robot Social Interactions: An Embodied and Situated Approach

Introduction: Developmental Learning

Teaching a multi-sensory artificial intelligence system to learn information concerning the surrounding world is a difficult task, which takes several years for a child, equipped with complex learning mechanisms, to accomplish. Indeed, the human body contains all sorts of multi-sensory elements very well adapted to the environment. Additionally, our brains are very complex and highly interconnected. Although a wide variety of models have been proposed to model its functioning, they often only address small parts of a much larger complex system. Consider, for instance, one small piece of the puzzle, object recognition: An object might have different meanings in different contexts; it might appear with various textures and colors, change shape, or be assembled with other objects. The function of an object within a task also varies significantly—for instance, a wooden rod when attached to a metal part, in one context might be a hammer, yet in another context, a walking stick.

However, infants have caregivers who lend a helping hand to facilitate their learning: changing interaction patterns according to each infant’s performance, so that the infant can learn useful information despite the complexity and noise in its surrounding world. Hence, infants’ functional development occurs simultaneously with the development of the caregivers' skills for socially interacting with infants (Sroufe, 1988). The importance of social interaction can be seen in developmental disorders such as autism (DSM-IV, 1994), which severely damages infants' social skills. Although autistic children often seem to have normal perceptual abilities, they do not recognize or respond to normal social cues (Baron-Cohen, 1995). This asocial behaviour puts serious constraints on the information that can be passed on by a caregiver to the autistic child, and severely limits the learning process.