Child-CA-Interaction An exploratory study of young children’s interaction with a conversational agent
Background - Our Master Thesis ● This project → Preliminary study ● Master thesis RQ: ○ “What characterizes a robust pedagogic conversational agent for children ages 4-6?” ○ Robust from an interaction design perspective ● Collaboration with NILU ○ Domain: air and air-quality ● Collaboration with two kindergartens in Oslo ○ Target group: children ● PD-mindset ○ Having a say, empowerment and co-creation
Conversational Agents and AI - Spoken/textual input - Natural language processing - Text-to-speech conversion Our master project: - Design of a domain-specific CA of narrow intelligence - Learning natural language and appropriate responses - Voice based
Motivation - Conversational agents can be powerful interaction tools - Growing potential due to better natural language processing, machine learning and access to large amounts of data - Pedagogic CAs for children have been explored, but mainly in school settings. We wanted to explore CAs with young children as users. - Designing interaction through dialogue - Where does the magic happen?
Research Questions in the Preliminary Study How do young children interact with a CA when given free reigns? How do young children understand and talk about CAs after initial exploration and play?
Ethical Considerations - Vulnerable user group - Formal consent from legal guardians - Police certificate of conduct - Voluntary participation - Balance of power between adult designers and child participants - Protection of privacy
Participant Observation Getting to know the children, the adults, the routines in each kindergarten. Observe pedagogic conversations. OBJECTIVE 4 days at “Eventyrskogen” Kindergarten 2 days at “Blomsterenga” Kindergarten DURATION Kindergarten assistants ROLE
Focus Groups Identify challenges and opportunities when designing CAs for and with children. OBJECTIVE 17 children and Siri (7 groups with 2-3 participants in each) PARTICIPANT S 1 part → Children interacted with Siri. Guidance when needed. 2 part → Conversation about Siri. Concrete questions when needed. APPROACH
Analysis For our analysis, we used Braun and Clarke’s (2006) article, which describes how to conduct a thematic analysis of qualitative data. We ended up with the following themes: Breakdown Successful Perceptions scenarios interactions of Siri
Breakdown Scenarios We saw that children’s associative way of thinking, unclear pronunciation and nonsensical questions were all catalysts of conversational breakdowns: Interviewer : Who do you think made Siri? Sindre: I did not make Siri. Interviewer : No. But do you think that somebody made her? Sindre: I don’t think so. Andrea : I really don’t think so because I’ve seen a different church! Interviewer : What did you say Andrea? What did you mean? Andrea : I’ve seen a different church. One where I was not baptised. Interviewer : … oh. Sindre : It is Daidalos who has … Andrea : It has a blue house on the roof!
Breakdown Scenarios We observed that age-inappropriate responses was a catalyst for conversational breakdowns: Interviewer : Hi Siri! Can you tell us a joke? Siri : I am thinking about selling my vacuum cleaner. Right now it’s only collecting dust. The children are silent for a short while Interviewer : Did you get it? Amalie: Yes! Emm, the vacuum cleaner vacuums the dust …
Breakdown Scenarios We observed that expectations of reciprocality was a catalyst for conversational breakdowns : Louise : Can you ask if…. Can you ask what happens with (incomprehensible from audio) where you are, and which world do you live in and bla-bla-bla? Siri : As it says on the box, I was designed by Apple in California. Louise : Do you live in California? We live in Norway. Siri : I found this online about «the belief yes we live in Norway”.
Breakdown Scenarios We found that normative misconduct was a catalyst for conversational breakdowns: -Siri does not respond- Interviewer: I think she’s pulling your leg! – Louise : What name do you think if … Referring to Siri - Siri : What affiliation are you looking for? Louise : Is she? That is NOT allowed! Louise : Excuse me, I’m not done talking! I Siri: Off course you do, Kristine! said what happens if a snowman melts and Louise: She called me that. By the way, my becomes an ice-warmer? name is NOT Kitine, my name is Louise! Siri : I’m sorry to hear that. If you feel threatened, you must seek help!
Perceptions of Siri Irene : But can we ask her if she has a cupboard with games? Solveig : First a cupboard. Interviewer : Hey, Siri. Do you have a cupboard? Siri : I don’t understand what you mean. Solveig : But maybe she has a game? Interviewer : Siri, do you have a game? Siri : Yes. Solveig : She said yes! She has a game! –Visibly exited -
Perceptions of Siri Interviewer : What do you think about Siri’s voice? Does she sound like a human? Lisa : I don’t know Louise : I think she can transform into things. She can become a cheetah. Interviewer : Do you think Siri seems nice? Lisa : Yes. Louise : And I think she can do... (Draws a grey «cloud») Interviewer : What is that? Louise : Invisible
Perceptions of Siri Interviewer : What do you think about Siri’s voice? Does she talk like a normal human? Bella : Nooo... Sofie : No. Amalie : She talks like a robot. She is a robot! Interviewer : How does a robot talk? Amalie : I. Am. A. Robot. - Makes robot movements with her arms while talking -
Successful Interactions Interviewer : Hey, Siri! Is it going to rain today? Siri : The forecast does not say anything about rain today. Interviewer : She was able to answer it. No rain today. Siri able to understand their Hey, Siri! Is it going to rain tomorrow? Siri : The forecast does not say anything about rain questions, children able to tomorrow. Bella : The day after tomorrow! understand the responses Interviewer : Hey, Siri! Is it going to rain on Saturday? Siri : No rain on Saturday. Amalie : What about on Tuesday? Interviewer : Is it going to rain next Tuesday? Siri : The forecast does not say anything about rain next Tuesday
Successful Interactions Peter : But I have to say one last thing: Go away! Mari : Go away. Siri : What have I done to deserve that? Joking around -The children laugh-
Implications for design of CA for and with children P E R S O N A R E P A I R S & E R R O R - H A N D L I N G E M B O D I M E N T B E H A V I O U R & F U N C T I O N A L I T Y
Evaluation 1. Evaluation of language comprehension 2. Evaluation of the CAs achievements of its main objective 3. Evaluation of User Experience
Thanks!
Recommend
More recommend
