Evaluating the Boundaries of Natural Conversation with Synthetic Personae for Multi-Modal Human Machine Dialogue Applications