Contextual factors and adaptative multimodal human-computer interaction: Multi-level specification of emotion and expressivity in Embodied Conversational Agents