Challenges of a Real-World HRI Study with Non-Native English Speakers

Using a Personalised Socially Assistive Robot for Cardiac Rehabilitation: A Long-Term Case Study

“…In fact, the patient verbally thanked the robot in three of these cases, which did not happen towards the other non-personalised behaviours. The patient negatively reacted to the misidentifications in user recognition, which may further support the social agent perspective, however, he very positively responded to the corresponding UTAUT question as well as to the other personalisation aspects, which suggests that personalisation mitigates the negative user experience [29].…”

Coffee With a Hint of Data: Towards Using Data-Driven Approaches in Personalised Long-Term Interactions

“…While there are available corpora for restaurant bookings ( Henderson et al, 2014 ; Bordes et al, 2017 ; Joshi et al, 2017 ) or travel bookings ( Hemphill et al, 1990 ; Bennett and Rudnicky, 2002 ; El Asri et al, 2017 ) based on Wizard-of-Oz human-machine interactions (i.e., a robot is tele-operated without knowledge of the user) or simulated datasets ( Serban et al, 2018 ), there was no publicly available corpus on the barista or personalised barista dialogues with customers at the time of conducting this work (January 2019) or the barista robot study (August 2019) ( Irfan et al, 2020a ). In October 2019, Taskmaster 3 ( Byrne et al, 2019 ) was released, which contains conversations with a personal digital assistant through Wizard-of-Oz or by “self-dialog” (i.e., crowdsourced workers imagined having a dialogue with a personal digital assistant and wrote the interaction for both sides).…”

“…Moreover, personalisation can increase task efficiency and awareness of the situational context of the conversation ( Neururer et al, 2018 ; Kocaballi et al, 2019 ). In addition, personalisation can facilitate user engagement and responsiveness in long-term HRI after the novelty effect wears off ( Dautenhahn, 2004 ; Bickmore and Picard, 2005 ; Kanda et al, 2010 ; Leite et al, 2013 ; Irfan et al, 2019 ), and it can overcome negative user experiences ( Irfan et al, 2020a ).…”

“…Most solutions in HRI either rely on touch screens or tele-operated robots (e.g., Kanda et al, 2010 , Lee et al, 2012 , Leite et al, 2017 , Glas et al, 2017 , Kanda et al, 2010 ; Lee et al, 2012 ; Leite et al, 2017 ; Glas et al, 2017 ) to bypass these issues, or use rule-based methods in structured transaction-oriented interactions by matching user responses to predefined templates (e.g., Kanda et al, 2007 , Churamani et al, 2017 , Zheng et al, 2019 , Kanda et al, 2007 ; Churamani et al, 2017 ; Zheng et al, 2019 ). However, rule-based approaches are inflexible to the variations in the user responses and are often experienced as time consuming and frustrating ( Williams et al, 2018 ; Bartneck et al, 2019 ; Irfan et al, 2020a ). Moreover, automatic speech recognition errors may arise from various accents, quietly speaking users and pronunciation errors of non-native speakers, which could decrease the robustness of rule-based approaches ( Irfan et al, 2020a ).…”

“…As context, we use task-oriented interactions between a customer and a robot barista in a coffee shop. In order to evaluate the state-of-the-art data-driven dialogue architectures that were strong baselines in other applications of personalisation ( Joshi et al, 2017 ; Zhang et al, 2018 ), and create a set of rules for a rule-based dialogue manager for generic and personalised barista robots ( Irfan et al, 2020a ), two simulated text-based Barista Datasets were created. 1) The Barista Dataset with generic interactions of a customer with a barista, and 2) the Personalised Barista Dataset with personalised long-term interactions, where the barista would recognise the users and learn and recall their preferences.…”