Enhancing Children's Self-Reporting in Chatbot Diaries through Rhyming Style

Paper Title

Enhancing Children's Self-Reporting in Chatbot Diaries through Rhyming Style

Publication Info

• Topic area: Child-computer interaction and conversational agent design for self-reporting.
• Keywords: Chatbot, self-reporting, children, rhyming style, phonological scaffolding, engagement, voice-based diaries, conversational agents, sleep diaries, LLM-powered systems.

Background and Problem

• Problem / challenge: Existing self-reporting tools for children, such as surveys and diaries, are often perceived as tedious, leading to disengagement and low-quality responses. Capability-adapted chatbots improve accessibility but fail to sustain engagement and elicit rich responses due to adult-centered phrasing and lack of expressive conversational styles.
• Significance: Improving children’s self-reporting is crucial for research, education, and healthcare, as it captures experiences like emotions and behaviors that cannot be directly observed. Enhancing engagement and response quality can lead to better data collection and insights into child development and well-being.
• Motivation and related work: Prior research has focused on simplifying vocabulary and tailoring prompts to children’s comprehension, but these approaches overlook phonological features like rhythm and rhyme, which are known to support attention, enjoyment, and memory. Rhyme has not been explored as a conversational style in self-reporting tools, leaving a gap in leveraging phonological scaffolding for improved interaction.

Solution

• Proposed approach: A rhyming conversational style integrated into a voice-based sleep diary chatbot powered by GPT-3.5, designed to enhance engagement and response quality in children’s self-reporting tasks.
• Novelty:
  1. Introducing rhyme as a phonological scaffolding strategy to complement semantic adaptation in child-agent interaction.
  2. Empirical evidence showing rhyme improves response quality across descriptive and non-descriptive questions.
  3. Development of age-sensitive conversational profiles and design rationales informed by co-design with children.
• Procedure and key techniques:
  • Conducted a co-design workshop with 35 children to identify preferences for chatbot conversational styles.
  • Developed a rhyming-style chatbot using three design rationales: short and lightweight rhymes, playful yet meaningful prompts, and bedtime-appropriate rhythms.
  • Compared rhyming and prose styles in a within-subjects study with 40 children aged 8–12, using structured sleep diary questions and engagement measures like the Giggle Gauge and Again-Again Table.
  • Implemented the system using GPT-3.5 for rhyming generation, Google Cloud APIs for speech-to-text and text-to-speech, and prosody adjustments for bedtime-friendly tone.

Results

• Concrete findings:
  • Rhyming style elicited higher response quality (RQI: M=2.88) compared to prose (RQI: M=2.25), with significant improvement across descriptive and non-descriptive questions (β = 0.63, p = 0.03).
  • Engagement ratings were high for both styles, but rhyming showed a slight advantage, with 61.53% of children preferring it overall.
  • Younger children (75%) preferred rhyming more than older children (50%), though the relative advantage of rhyme for response quality was consistent across age groups.
• Advantage over baselines:
  • Rhyming prompts improved response richness and engagement compared to prose, addressing limitations of capability-adapted chatbots that rely solely on semantic simplification.
• Experiments / evaluation:
  • 1 × 2 within-subjects design with counterbalanced order.
  • Quantitative measures: Response Quality Index (RQI), Giggle Gauge, Again-Again Table.
  • Qualitative interviews to explore children’s perceptions and preferences.
• Limitations and future work:
  • Conducted in a lab-based, short-term setting, limiting ecological validity and long-term adherence insights.
  • Future work includes longitudinal home deployment, adaptive conversational designs (e.g., mixed styles, user-controlled switching), and exploration of contextual factors like bedtime routines and family dynamics.

Summary

This study introduces rhyme as a conversational style in voice-based sleep diaries for children, demonstrating its ability to improve response quality and engagement compared to prose. Rhyming prompts function as phonological scaffolding, reducing cognitive load and enhancing verbal expression across descriptive and non-descriptive questions. Preferences varied by age, with younger children favoring rhyme more strongly, but its benefits for response quality were consistent across groups. While the findings highlight the potential of conversational style as a cognitive and affective design lever, further research is needed to evaluate its long-term efficacy and adaptability in real-world contexts. This work contributes to child-agent interaction by broadening the design space to include phonological cues alongside semantic adaptation.