Beyond Prompt-Following: Empowering Social Agency in Autistic Children through Narrative Framing
Min Zhou, Xinheng Song, Xiaolan Peng, Mingxuan Jin, Soumya C. Barathi · 2026 · Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems (CHI EA ’26) · doi:10.1145/3772363.3798353
Summary
NarraSocial is a Unity-based virtual environment for autistic children aged 6–10 that reframes social-skills interventions away from the prompt-and-response loop characteristic of Applied Behavior Analysis (ABA) and most VR-based social skills training. The authors observed therapist-led social-skills classes at two Beijing autism intervention centers and saw that interactions were treated as discrete instructional tasks (e.g., "select the appropriate response") that positioned children as prompt-followers. NarraSocial instead places the child in a low-poly first-person amusement-park environment as a character named Xiaoke spending a day at the park, encountering everyday social moments such as saying goodbye to a parent, greeting a friend, apologizing after a collision, and meeting unfamiliar adults. Each scene presents branching choices framed inside the story ("Dad is waving goodbye to you. How would you choose to say goodbye?") with non-punitive, color-coded feedback: green choices advance the narrative, red choices produce reflective feedback rather than failure. The within-subjects pilot study compared this narrative framing against an instruction-driven baseline that used identical assets, animations, and choices but decontextualized the prompts ("The character is waving. Select the appropriate response"). Ten children with prior clinical autism diagnoses, accompanied by caregivers and observed by two rehabilitation therapists, played both conditions in counterbalanced AB/BA order. Data came from real-time observation and 15-minute post-session semi-structured interviews with caregivers and therapists, analyzed via Reflexive Thematic Analysis.
Key findings
Two qualitative patterns emerged. First, in the instruction-driven baseline children behaved as prompt-followers, often waiting for explicit guidance before acting and showing disengagement signals such as asking to stop or seeking a caregiver, while in the narrative condition a subset showed exploratory behavior including "anticipatory scanning" of the environment to find interaction points before any prompt; one caregiver said their child "kept walking around the game map, curious about everything." Second, in the narrative condition children sometimes produced contextually situated social gestures (e.g., spontaneously waving back during a goodbye scene) that did not appear in the baseline, where the same input was treated as a functional step. Therapists framed the system as potentially complementary to institutional ABA sessions, useful as at-home review rather than a replacement. The authors are explicit and disciplined about scope: they do not claim long-term skill acquisition, generalization to real life, or clinical efficacy. They acknowledge alternative explanations including imitation, playfulness, and novelty effects, and they situate "social agency" as an interactional orientation in the moment, drawn from interactive-narrative theory (Murray's procedural agency, Tanenbaum's commitment-to-meaning), not as a stable trait.
Relevance
For accessibility designers working on autism-focused tools, the central contribution is a clean reframing: interaction framing is a structural design choice on par with mechanics or content, and the same assets can produce very different relationships between an autistic child and a system depending on whether the system treats them as a responder or an interpreter. The non-punitive color-coded feedback design, where "wrong" answers prompt reflection rather than failure, directly answers the Bottema-Beutel critique that social-skills training can stigmatize autistic ways of being. The Spiel et al. critique of prior autism technology research (cited as [27]) that autistic children are rarely positioned as agents is taken seriously here. Practical limitations should temper any deployment: sample size is 10, the condition difference is identified through caregiver and therapist accounts rather than measured behavior, the intervention is text-heavy in its current form, and findings come from a single open-ended scenario. The authors flag that observed gestures may reflect novelty rather than orientation change, and they make no claims about transfer to real-world social situations. The design implications, contextual narrative scaffolding and meaning-driven feedback over correctness markers, are the most transferable artifact for practitioners building inclusive social technology.
Tags: autism · autistic children · social agency · narrative design · virtual environment · serious games · neurodiversity · interaction design · social skills