Natural Language Access to Domain-Specific Metadata: A Reusable Framework for LLM Query Generation

The paper presents NLKGQ, a reusable framework for querying domain-specific archive metadata with natural language. A domain-agnostic harness uses an LLM to translate user questions into SPARQL and execute them against a knowledge graph defined by an OWL ontology. The demonstration uses metadata from a large neuroimaging research archive and evaluates multiple LLMs and ontology representations. The best configurations reportedly reach 100% accuracy on expert-developed competence and regression question sets without fine-tuning, retrieval augmentation, or multi-agent orchestration. Ablations across eight ontology representations identify readable entity names and semantic annotations as the strongest accuracy factors, while OWL-based structure outperforms an auto-generated SQL backend.