Conceptual development from the perspective of a brain-inspired robotic architecture

Concepts are central to reasoning and intelligent behaviour. Scientific evidence shows that conceptual development is fundamental for the emergence of high-cognitive phenomena. Here, we model such phenomena in a brain-inspired cognitive robotic model and examine how the robot can learn, categorise, and abstract concepts to voluntary control behaviour. The paper argues that such competence arises with sufficient conceptual content from physical and social experience. Hence, senses, motor abilities and language, all contribute to a robot's intelligent behaviour. To this aim, we devised a method for attaining concepts, which computationally reproduces the steps of the inductive thinking strategy of the Concept Attainment Model (CAM). Initially, the robot is tutor-guided through socio-centric cues to attain concepts and is then tested consistently to use these concepts to solve complex tasks. We demonstrate how the robot uses language to create new categories by abstraction in response to human language-directed instructions. Linguistic stimuli also change the representations of the robot's experiences and generate more complex representations for further concepts. Most notably, this work shows that this competence emerges by the robot's ability to understand the concepts similarly to human understanding. Such understanding was also maintained when concepts were expressed in multilingual lexicalisations showing that labels represent concepts that allowed the model to adapt to unfamiliar contingencies in which it did not have directly related experiences. The work concludes that language is an essential component of conceptual development, which scaffolds the cognitive continuum of a robot from low-to-high cognitive skills, including its skill to understand.