TRANSoM is a collaborative effort among university and industry researchers aimed at producing an intelligent tutoring system for training pilots of remotely operated vehicles (ROVs). ROVs are unmanned, tethered, underwater vehicles used in a range of applications such as inspection, search and salvage, and mine countermeasures. Pilots have to learn to maneuver the ROV, keeping track of its tether and its surroundings, using little more than a video camera and sonar. To minimize workload while a trainee is practicing, the primary mode of feedback during a mission is verbal. The design of this verbal coaching was modeled closely on techniques observed in human instructors. We report on several of the issues faced in implementing this coaching paradigm that make it effective yet unintrusive.