Team research project · June 2026
SentenceBot: adaptive sentence practice with a social robot
An Alpha Mini prototype combining tangible topic selection, adaptive difficulty and language-model feedback.
Conceptual illustration; not an experimental result.
Individual contribution
My role
I led the adaptive difficulty engine, QR integration and the robot’s celebration gesture. We shared the wider integration work, with equal overall contribution across the team.
Contribution statement in the reportDesign question
How can a social robot support structured sentence-production practice with a simple interface and feedback adapted to the learner? SentenceBot was designed as a supplementary tool for children with Developmental Language Disorder.
Implementation
The Alpha Mini robot reads physical QR topic cards using OpenCV. A sentence-production game combines speech input, language-model validation, spoken feedback and gestures. Three adaptive rounds change the difficulty according to responses. The implementation supports a local Gemma 3 model through Ollama and an optional Gemini backend.
Prototype and contribution
The group report records equal contributions and describes an implemented system demonstrated on the physical robot, with a terminal mode for local testing. The work integrates dialogue, perception and feedback within one interaction loop.
Limitations
Formal clinical evaluation with children diagnosed with DLD remains future work. Speech-recognition errors can incorrectly reject otherwise valid sentences, and response timeouts can interrupt the interaction. The prototype demonstrates system integration; it does not establish therapeutic efficacy.
Related research
Fairness & explainability