Data-Secure Natural Language Interfaces to Databases via Large Language Models

Abstract

Natural Language Interfaces to Databases (NLIDB) aim to allow users to query databases using natural language, eliminating the need for proficiency in formal query languages like SQL. Recent advancements in Large Language Models (LLMs) have shown promise in connecting natural language and structured query generation. This paper proposes a novel approach utilizing an LLM system to translate natural language queries into optimized SQL statements while ensuring data security by not exposing database contents to external models. The system leverages the database schema. The challenges of working with morphologically rich languages like Turkish are also discussed, along with how the proposed approach addresses these issues. © 2025 Elsevier B.V., All rights reserved.