Conversational Control with Ontologies for Large Language Models: A Lightweight Framework for Constrained Generation

Researchers developed a lightweight framework that uses ontological definitions to provide modular and explainable control over Large Language Model outputs in conversational systems. The method fine-tunes LLMs to generate content according to specific constraints like English proficiency level and content polarity, consistently outperforming pre-trained baselines across seven state-of-the-art models.

• →New framework enables controllable generation in conversational AI by using ontological definitions as constraints.
• →Method successfully controls two key aspects: English proficiency level and content polarity in conversations.
• →Testing on seven open-weight conversational LLMs showed consistent improvements over pre-trained baselines.
• →Framework is model-agnostic, lightweight, and interpretable, making it extensible to new domains.
• →Approach enhances alignment with strategy instructions while maintaining explainability in AI systems.