In this paper we reflect on the use of questionnaires as an evaluation tool in child-robot interaction research. We provide a case study containing eight user studies. While doing these user studies we ran into two major challenges: violations of the constructs used in questionnaires and a ceiling effect in the responses of the children. These issues are caused by a combination of factors such as, but not limited to, misinterpretations of questions, response biases, and the novelty effect. A first lesson learned is that a proper design of a questionnaire, and how questions are asked and answered, is essential. In this paper we discuss two questionnaire methods we have been developing that potentially could circumvent some of the issues. A second lesson learned is that user studies could benefit if they reflect the long-term nature of the child-robot interaction.
Social robots are entering the private and public domain where they engage in social interactions with nontechnical users. This requires robots to be socially interactive and intelligent, including the ability to display appropriate social behaviour. Progress has been made in emotion modelling.However, research into behaviour style is less thorough; no comprehensive, validated model exists of non-verbal behaviours to express style in human-robot interactions. Based on a literature survey, we created a model of non-verbal behaviour to express high/low warmth and competence—two dimensions that contribute to teaching style. In a perception study, we evaluated this model applied to a NAO robot giving a lecture at primary schools and a diabetes camp in the Netherlands. For this, we developed, based on expert ratings, an instrument measuring perceived warmth, competence, dominance and affiliation. We show that even subtle manipulations of robot behaviour influence children’s perceptions of the robot’s level of warmth and competence.
Educational technology needs a model of learning goals to support motivation, learning gain, tailoring of the learning process, and sharing of the personal goals between different types of users (i.e., learner and educator) and the system. This paper proposes a tree-based learning goal structuring to facilitate personal goal setting to shape and monitor the learning process. We developed a goal ontology and created a user interface representing this knowledge-base for the self-management education for children with Type 1 Diabetes Mellitus. Subsequently, a co-operative evaluation was conducted with healthcare professionals to refine and validate the ontology and its representation. Presentation of a concrete prototype proved to support professionals’ contribution to the design process. The resulting tree-based goal structure enables three important tasks: ability assessment, goal setting and progress monitoring. Visualization should be clarified by icon placement and clustering of goals with the same difficulty and topic. Bloom’s taxonomy for learning objectives should be applied to improve completeness and clarity of goal content.
This paper describes ongoing work carried out in the European project PAL which will support children in their diabetes self-management as well as assist health professionals and parents involved in the diabetes regimen of the child. Here, we will focus on the construction of the PAL ontology which has been assembled from several independently developed sub-ontologies and which are brought together by a set of hand-written interface axioms, expressed in OWL. We will describe in detail how the triple model of RDF has been extended towards transaction time in order to represent time-varying data. Examples of queries and rules involving temporal information will be presented as well. The approach is currently been in use in diabetes camps.
The PAL project is developing an embodied conversational agent (robot and its avatar), and applications for child-agent activities that help children from 8 to 14 years old to acquire the required knowledge, skills, and attitude for adequate diabetes self-management. Formal and informal caregivers can use the PAL system to enhance their supportive role for this self-management learning process. We are developing a common ontology (i) to support normative behavior in a flexible way, (ii) to establish mutual understanding in the human-agent system, (iii) to integrate and utilize knowledge from the application and scientific domains, and (iv) to produce sensible human-agent dialogues. The common ontology is constructed by relating and integrating partly existing separate ontologies that are specific to certain contexts or domains. This paper presents the general vision, approach, and state of the art.
The PAL project is developing: (1) an embodied conversational agent (robot and its avatar); (2) applications for child-agent activities that help children from 8 to 14 years old to acquire the required knowledge, skills and attitude for adequate diabetes self-management; and (3) dashboards for caregivers to enhance their supportive role for this self-management learning process. A common ontology is constructed to support normative behavior in a flexible way, to establish mutual understanding in the human-agent system, to integrate and utilize knowledge from the application and scientific domains, and to produce sensible human- agent dialogues. This paper presents the general vision, approach, and state of the art.
Social Robots are increasingly applied in healthcare and education. Pedagogical Agents (PAs) are being developed to adapt to the users knowledge, and efforts are made in strategic action selection: what action is appropriate given the context and user preference. However, the issues of how these actions can be appropriately communicated receives less attention. In this paper we propose the development of an adaptive pedagogical interaction style for a robot. We discuss the role of style in human-human interaction and the lack thereof in human-robot interaction. While human educators heavily rely on their ability to identify and respond accordingly to social signals in a fluent and natural way, robots cannot adapt their style of interaction effectively. By adapting the pedagogical interaction style of a robot to the learner and context we expect to be able to create rich and fruitful personalized educational interactions and ultimately facilitate social bonding between the learner and robot. In this position paper we present our view as a starting point for the management of this interaction style. As a basis for the proposition, pedagogic and motivational theories are used.
We study turn-taking behaviour in non-cooperative dialogue for the development of believable characters in a serious game for conversational skill learning in the police interview context. We describe a perception study to see how participants perceive a suspect’s interpersonal stance, rapport, face, and deception when the turn-taking of the subject varies. We influence the perception of the suspect’s stance by altering the timing of the start of speech with respect to the ending of the interlocutor’s speech. The results of the study contribute to the development of an embodied conversational agent capable of natural humansystem conversation with appropriate turn-taking behaviour.
A serious game for learning the social skills required for effective police interviewing is a challenging idea. Building artificial conversational characters that play the role of a suspect in a police interrogation game requires computational models of police interviews as well as of the internal psychological mechanisms that determine the behaviour of suspects in this special type of dialogues. Leary’s interactional circumplex is used in police interview training as a theoretical framework to understand how suspects take stance during an interview and how this is related to the stance and the strategy that the interviewer takes. Interactional stance is a fuzzy notion. The question that we consider here is whether different observers of police interviews agree on the type of stance that suspect and policemen take and express in a face-to-face interview. We analyzed police interviews and report about a stance annotation exercise. We conclude that although inter- annotator agreement on stance labeling on the level of speech segments is low, a majority voting “meta-annotator” is able to reveal the important dynamics in stance taking in a police interview. Then we explore the relation between the stance taken by the suspect and turn-taking behaviour, overlaps, interruptions, pauses and silences. Our findings contribute to building computational models of non-player characters that allow more natural turn-taking behaviour in serious games instead of the one-at-a-time regime in interview training games.