A believable life-like agent must appear to understand what is said to it, or what is being said around it, even if this is not actually the case. It must be able to follow a conversation and to understand what is happening in a discourse, even though the verbal content of the dialogue may be too complex (or too noisy) to be recognised. This paper describes the platform design and details the description language being used for the ‘SCOPE' non-verbal speech-processing research currently being carried out at ATR for such a communicative robotic agent.