In this paper we focus on human-robot interaction particularities that occur during programming by demonstration. Understanding what makes the interaction rewarding and keeps the user engaged helps optimize the robot's learning. Two user studies are presented. The first one validates facially-displayed expressions on the iCub robot. The best recognized displays are then used in a second study, along with other ways of providing feedback during teaching a manipulation task to a robot. We determine the preferred and more effective way of providing feedback in relation to the robot's tactile sensing, in order to improve the teaching interaction and to keep the users engaged throughout the interaction.