In this chapter, we systematically examine the diverse metrics and methodologies for evaluating interactive NLP (iNLP) systems across four distinct interactive entities. Section 10.1 addresses Human-in-the-loop interactions, dissecting general metrics that assess alignment with human preferences and task-specific metrics tailored to particular applications. We then shift to KB-in-the-loop evaluations in Sect. 10.2, where we discuss metrics for knowledge acquisition and knowledge-enhanced generation, pivotal for enhancing NLG with reliable knowledge. Section 10.3 explores Model/Tool-in-the-loop interactions, delving into chain-of-thought capabilities, tool-use abilities, and collaborative behaviors. Finally, Sect. 10.4 considers Environment-in-the-loop interactions, highlighting the development of embodied task platforms and corresponding metrics that gauge the model’s efficacy in dynamic, real-world scenarios. Each section offers insights into the current state-of-the-art evaluation practices and identifies areas where further innovation is necessary, providing a comprehensive overview of the multifaceted process of iNLP system evaluation.