Editorial: Failures and Repairs in Human-Robot Communication

Frank Förster ^* Patrick Holthaus

• University of Hertfordshire, Hatfield, United Kingdom

This research topic arose on the back of the WTF workshop series (Förster et al., 2022(Förster et al., , 2023a) ) that brought 2 together an interdisciplinary group of researchers ranging from roboticists and computational linguists to 3 conversation analysts and cognitive scientists to openly and frankly discuss failures of (robotic) speech 4 interfaces they experienced when deploying these in their studies. Some of the issues discussed in the 5 workshops are elaborated in the contributed articles below, more pointers can be found in the workshop 6 summary article by Förster et al. (2023b).This research topic contributes towards two main objectives: Firstly, we provide a platform for reporting 8 commonly occurring communicative failures in human-robot interaction (HRI). Secondly, this topic aims Galbraith (2024) investigate how virtual assistants deal with the interactionally highly relevant and 28 frequent 'huh?', an other-initiated, and likely universal repair marker (cf. Dingemanse et al., 2015). They further investigate what repair strategies these assistants utilise when encountering unintelligible speech, 30 and how native speakers judge these different repair strategies. In their study, two different virtual assistants, 31 Google Assistant and Apple's Siri, are compared across two different languages (English and Spanish).Galbraith finds that neither assistant actively produces 'huh?' but rather employs more specific repair 33 strategies when confronted with unintelligible speech. The assistants frequently have trouble dealing with a 34 'huh?' produced by human users, and some of the repair strategies employed by the two assistants were 35 rated negatively by human judges. While these insights were gained by interacting with virtual assistants, 36 we expect some of these to apply to SDS more generally (cf. Lopez et al., 2022).