DSVA: An Iterative Refinement Multi-Agent Framework for NL-to-MTL Translation
Published in IEEE Access, 2026
This paper proposes DSVA, an iterative refinement multi-agent framework for translating natural language (NL) requirements into Metric Temporal Logic (MTL). The framework leverages retrieval-augmented generation (RAG) and large language models (LLMs) within a multi-agent system to achieve high-fidelity formal specification, addressing key challenges in requirements engineering for real-time systems.
Keywords: Translation; Logic; Semantics; Retrieval augmented generation; Manuals; Iterative methods; Large language models; Measurement; Syntactics; Real-time systems; Formal specification; large language models; metric temporal logic; multi-agent system; requirements engineering
Recommended citation: J. Lyu, Y. Wu and Y. Cai, "DSVA: An Iterative Refinement Multi-Agent Framework for NL-to-MTL Translation," IEEE Access, vol. 14, pp. 28457-28477, 2026, doi: 10.1109/ACCESS.2026.3665806.
Download Paper