State-of-the-art AI planning systems are able to generate complex plans thanks to their efficient reasoning engines. In a large number of application domains, the plans are automatically executed by systems such as autonomous robots. In this context, it is not necessary to make these automated systems understand what they are actually doing during execution and why they are doing that. In other words, these systems do not need to understand the underlying semantics of the plans they execute and how these plans have been generated. However, there are a significant number of key application domains, such as disaster relief mission support or project planning, where plans are supposed to be executed by a human user who is not necessarily a planning expert, an application expert, or both. In addition, for real-world applications, the plans and the plan generation are often complex. In order to unlock a part of the application potential of the AI planning technology, it is necessary that the user trusts the technology (Glass et al. 2008, pp. 12-18). Increasing trust in AI planning systems requires the design and implementation of user-friendly interfaces and the development of plan explanation methods that allow for taking into consideration the human user’s queries related to some components of the plan about their meaning and relevance for the plan and giving back the appropriate information that answers these queries.

The verbal communication by speech constitutes the most natural form of communication for humans. By means of natural language dialogs in this work, we focus on the explanation of plans that are generated by a refinement-based planning system. Contrary to most approaches presented in the literature that try to provide explanations when backtracking occurs in failure situations during search, we assume in this work that the plans for which explanations are looked for are consistent. We present a domain-independent approach to enabling verbal human queries and producing verbal plan explanations.