Papers
Keeping the Conversation on Track: Using Large Language Models for Interest Based Topic Management in Child Robot Interactions
Author: Anita Vrins //
The conversational abilities of social robots can benefit from large language models (LLMs) to manage topics. Pre-structured and partially pre-scripted dialogue ensures safety, particularly in educational settings with children. However, using a fully pre-scripted approach does not allow for taking the child’s interests into account nor can it handle child-initiated topic changes. These are key for engagement in Child-robot interaction (CRI), which affects the children’s experience and willingness for long-term interaction. This research aims to enhance CRI beyond fixed dialogue trees by programming the robot with LLM-powered conversation topic management capabilities. The robot will be able to identify the level of interest of the children in real time based on linguistic features of their conversation with the robot, propose relevant continuation topics based on the detected interest level, and briefly engage in mini dialogues using personalized LLM prompting.