A neural network based software agent has learnt some semi-improvised dance phrases from Steph. Here Steph provides movements which the agent then uses as starting points for the generation of its own movement phrases as a response. Steph’s avatar is white, the Agent’s avatar is red. We are using an optical motion capture system to provide the agent with Steph’s movement stream both for learning and so it can discern what Steph is doing. Voice commands tell the agent when to find a close matching pose and then start creating a new phrase.