Miscommunication and repair in Dutch task-oriented human-robot interaction

The communicative skills of robots and voice agents tend to be evaluated critically, while - or maybe because – we expect them to be as skilled as we are. One of the important resources in social interaction for maintaining communication is the mechanism of repair (Albert & De Ruiter, 2018; Schegloff, Jefferson, Sacks, 1977). EMCA studies of the interaction between various conversational agents and robots and humans have already examined instances of interactional trouble with voice agents and its resolution (e.g., Porcheron et al, 2018). Based on the extensive body of work on repair in human interaction, we were interested in understanding the various kinds of interactional trouble and repair occurring in human-robot interaction. Our data consist of 37 videorecordings of elderly people completing a lengthy health survey at the hospital with a Pepper robot. The participants were asked to answer the robot’s questions after an instruction from a nurse and a few try-out questions from Pepper (e.g. “how did you get here today?”). We identified all instances of observable miscommunication and repair in the data, finding that miscommunication was abundant, with some types of repair being much more common than others. These seemed mainly related to “hearing trouble” on the side of the robot, leading to participants’ repeats and rephrasals. Interestingly, interactional trouble appeared to escalate, with multiple repair initiations, sometimes leading to “second best answers”. This seems to be a strategy for participants to avoid repair in the service of progressivity. Overall, miscommunication and repair in these task-oriented human-robot interactions require a lot of adaptation from the human (cf. Pelikan & Broth, 2016).