MORA: Designing an AI-Mediated Role-Playing System for Child-Caregiver Speech Sound Practice
TL;DR
MORA, an AI-mediated role-playing story-based system in which target-bearing words recur as characters, locations, and props, eliciting productions under two response constraints while retaining recordings for clinician review is presented.
Abstract
Treating speech sound disorder (SSD) requires repeated practice across linguistic complexity levels. However, in the home setting, caregivers lack the expertise to judge children's productions to sustain intensive practice. Using a formative study with six speech-language pathologists (SLPs) and an expert walkthrough with seven SLPs, we present MORA, an AI-mediated role-playing story-based system in which target-bearing words recur as characters, locations, and props, eliciting productions under two response constraints while retaining recordings for clinician review. Because ASR artifact, more than child error, dominated phoneme-distance scores of disordered recordings at low error levels, the dashboard abstains rather than classifies there. A lab-based user study with 10 child-caregiver dyads (five reporting SSD or therapy history) surfaced three tensions: connected speech and comprehensible tasks, a responsive signal and the authority it accrues, and child autonomy and the coordination it requires. We discuss design implications for revisable configuration, scope-bounded feedback, and coordination support.
