With advances in neural network-based computation, socially assistive robots have been endowed with the ability to provide natural conversation to users. However, the lack of transparency in the computation models results in unexpected robot behaviors and feedback, which may cause users to lose their trust in the robot. Theory of mind (ToM) in cooperative tasks has been considered as a key factor in understanding the relationship between user acceptance and the explainability of robot behaviors. Therefore, we develop a dialog system using previously collected data from a robot-mediated cooperative communication task data to simulate natural language smart feedback. The system is designed based on the mechanism of ToM and validated with a simulation test. Based on the result, we believe the designed dialog system bears the feasibility of simulating ToM and can be used as a research tool for further studying the importance of simulating ToM in human-robot communication.