Designing Social Robots for Inclusive Child Wellbeing Assessment: Insights from Communities Supporting Developmental Language Disorder and Forced Migration

Technical Summary: Designing Social Robots for Inclusive Child Wellbeing Assessment

Problem Statement
Assessing the wellbeing and mental health of children is inherently complex, a challenge exacerbated for populations facing communication barriers. Specifically, children with Developmental Language Disorder (DLD) and children with forced migration backgrounds often struggle with traditional self-report questionnaires due to language comprehension deficits, verbal expression difficulties, and the cognitive demands of abstract items. While social robots have emerged as promising tools for multimodal interaction (combining speech, non-verbal behavior, and social engagement) to support wellbeing assessment, existing research has primarily focused on feasibility and data elicitation in controlled settings with typically developing children. There is a critical gap in understanding how to design robot-mediated assessment interactions that are inclusive, appropriate, and ethically acceptable for children with diverse communication needs and lived experiences.

Methodology
This study employed a qualitative, participatory design approach within the broader MICRO project to investigate design considerations for children aged 8–11 years. The research involved two distinct focus groups:

1. DLD Focus Groups: Two sessions involving seven participants (two parents of children with DLD and five professionals such as teachers and counselors) from the UK.
2. Forced Migration Focus Groups: Two sessions involving seven practitioners (psychologists, nurses, health promoters) from Sweden working with children from forced migration backgrounds.

Design Probes:
Prior to the focus groups, the research team developed a set of five candidate child–robot interaction activities based on interdisciplinary expertise in robotics, HRI, affective computing, and child psychiatry. These activities targeted four domains: language/communication, social/non-verbal interaction, and emotional understanding/expression. The activities included:

• Getting to Know You: Low-pressure introductions.
• Structured Storytelling/Picture Description: Eliciting language and social-emotional content.
• Gesture Game: Observing non-verbal behavior (mimicry, imitation).
• Turn-Taking Game: Assessing social interaction and cooperation.
• Emotion Expression and Understanding: Discussing and identifying emotions.
• Story-based Self-disclosure Task: (Specific to the forced migration group) A narrative-based reflection using third-person storytelling to elicit self-relevant judgments without direct verbal questioning.

Procedure:
The activities served as design probes. Facilitators guided participants through discussions regarding the appropriateness, inclusiveness, and ethical implications of these interactions. The protocol covered robot roles, interaction dynamics, individual differences, child agency, and specific ethical concerns (e.g., data privacy, attachment, bias). Data was analyzed using reflexive thematic analysis (Braun & Clarke), involving independent coding by researchers followed by collaborative synthesis to identify overarching themes and population-specific nuances.

Key Results
Thematic analysis of the focus group discussions yielded four overarching themes, with specific sub-themes and population-specific contributions:

1. The Role and Capability of the Robot Should Be Established:

   • Playful Tool: Robots should be framed as playful, low-pressure partners to reduce anxiety and shame, particularly for sensitive topics.
   • Managed Expectations: Clear boundaries are essential to prevent over-attachment and to clarify that the robot is not a human or a pet. Transparency regarding data usage (who sees the data and why) is critical, especially for forced migration families concerned about immigration status.
   • Supplementary, Not Replacement: Robots must be positioned as tools to complement, not replace, human clinicians. They lack the empathy and judgment required for sensitive disclosures.

2. Interactional Nuances Are Fundamental to the Outcome:

   • Getting to Know Each Other: A "warming up" phase is crucial. For DLD children, this may require organic, individualized icebreakers; for forced migration contexts, an adult introduction helps bridge the transition.
   • Interaction Dynamics: Engagement relies on reciprocity. Children need to feel they can initiate interaction, not just respond. The robot's affective behavior (facial expressions, tone) must be consistently positive to avoid misinterpretation of negative cues.
   • Preparedness for the Unexpected: Protocols must exist for handling distress, trauma triggers, or safeguarding disclosures. A human support person must be present to interpret non-verbal cues and intervene.

3. The Interaction Should Be Designed with Individual Differences in Mind:

   • Layered Adaptability: A "one-size-fits-all" approach is insufficient. Content, pacing, and complexity must be adjustable based on the child's interests, developmental stage, and past experiences (e.g., avoiding military themes for children with war trauma).
   • Multifaceted Communication: Reliance on verbal communication is limiting. Multimodal supports (images, emojis, gestures, music) and adult facilitation are necessary to accommodate diverse communication preferences and abilities.

4. Child Agency Should Be Cultivated:

   • Active Agency: Children should be enabled to make decisions, guide the robot, or "save the day," shifting the power dynamic from adult-led to child-led.
   • Avoiding Failure: The design must eliminate the perception of "right or wrong" answers to reduce performance pressure and social desirability bias, which is particularly relevant for children who may feel the need to protect their families or themselves from judgment.

Population-Specific Insights:

• DLD Context: Emphasis was placed on the need for reduced verbal demands, clear affective cues (to prevent misreading of robot expressions), and the role of adults in facilitating interaction and safeguarding.
• Forced Migration Context: Critical concerns included data privacy and trust in institutional systems, the need for cultural and linguistic representation in materials, and the risk of re-traumatization through specific triggers (e.g., police/military imagery). There was also a noted risk that children might feel devalued if a robot replaces human care, potentially reinforcing feelings of marginalization.

Significance and Contributions
The paper contributes a set of ethical and inclusive design recommendations for robot-mediated wellbeing assessment, moving beyond technical feasibility to address the socio-technical and ethical dimensions of the interaction. The authors argue that inclusive assessment requires:

1. Human Oversight: Maintaining human control over high-stakes interpretation and safeguarding disclosures.
2. Community Co-design: Validating systems with the target populations (DLD and forced migration) rather than relying on typically developing or WEIRD (Western, Educated, Industrialized, Rich, Democratic) datasets.
3. Bias Mitigation: Rigorously testing robot components (LLMs, classifiers) for intersectional bias regarding disability, age, nationality, and religion before deployment.
4. Content Constraints: Using curated, reviewed content banks to avoid culturally loaded or trauma-triggering material.
5. Epistemic Justice: Framing the robot transparently to avoid false authority and ensuring the child is recognized as the credible author of their own experience.

The study concludes that robot-mediated assessment should be viewed as a relational and contextual process. While the proposed activities serve as design probes, the authors note that future work is required to empirically test these activities in practice and to explore how these design principles apply to other populations with diverse communication needs, such as autistic children or those using AAC.