TalBot: Developing an LLM-Based Robot-Assisted Learning Application for Children with Language Vulnerabilities

Children with language vulnerabilities need extra
support in their language development, and game-based learning
is often used as a part of the interventions. Speech language
therapists, educators and parents form a team around the
child, working together to support language learning. However,
resources such as time, guidance and competence are often
scarce, leaving room for alternative solutions, like robot-assisted
language learning (RALL). We introduce TalBot, a large language
model (LLM)-powered robot application, aiming to lead and
play the language game Alias with a small group of children
with language vulnerabilities. Our application is designed to
lead the game, give adaptive responses, manage turn-taking
and engage the players by providing emotionally congruent
verbal and non-verbal responses. By constraining the context
and using an LLM, we believe that the effectiveness of automatic
speech recognition (ASR) and management of turn-taking can
be improved. In general, we suggest LLMs enable robots to
better support children with language vulnerabilities — and seem
especially suited to an area such as this where variable input is of
importance. We hope other researchers will also further explore
the use of LLMs in RALL, especially applications designed for
children with language vulnerabilities.