Towards a Grounded Dialog Model for Explainable Artificial Intelligence