Developing Embodied Multisensory Dialogue Agents

摘要

A few decades of work in the AI field have focused efforts on developing a new generation of systems which can acquire knowledge via interaction with the world. Yet, until very recently, most such attempts were underpinned by research which predominantly regarded linguistic phenomena as separated from the brain and body. This could lead one into believing that to emulate linguistic behaviour, it suffices to develop 'software' operating on abstract representations that will work on any computational machine. This picture is inaccurate for several reasons, which are elucidated in this paper and extend beyond sensorimotor and semantic resonance. Beginning with a review of research, I list several heterogeneous arguments against disembodied language, in an attempt to draw conclusions for developing embodied multisensory agents which communicate verbally and non-verbally with their environment. Without taking into account both the architecture of the human brain, and embodiment, it is unrealistic to replicate accurately the processes which take place during language acquisition, comprehension, production, or during non-linguistic actions. While robots are far from isomorphic with humans, they could benefit from strengthened associative connections in the optimization of their processes and their reactivity and sensitivity to environmental stimuli, and in situated human-machine interaction. The concept of multisensory integration should be extended to cover linguistic input and the complementary information combined from temporally coincident sensory impressions.