TRR 318-2 - Project C07: Co-construction-following large language models for explaining

C07 studies the adjustment of LLM training processes and knowledge integration to enable the successful co-construction of explanations. Scaffolding and monitoring are operationalized via a novel fine-tuning process that teaches LLMs to explain dialogically while modeling dialog states. A novel co-construction-aware RAG approach exploits these dialog states to update and provide relevant domain knowledge and context factors to the LLM and thus guarantee the context awareness of explanations.