A Hierarchical Hybrid-Intelligence Architecture with Consensus, Debate, and Reflection for High-Fidelity NL-to-PPTL Conversion
Published in 2025 32nd Asia-Pacific Software Engineering Conference (APSEC), 2025
This paper proposes a hierarchical hybrid-intelligence multi-agent architecture combining consensus, debate, and reflection mechanisms for high-fidelity translation of natural language requirements into Projected Propositional Temporal Logic (PPTL). The work addresses challenges in formal specification of software requirements using Large Language Models and multi-agent systems.
Keywords: Adaptation models; Accuracy; Translation; Large language models; Knowledge based systems; Reflection; Safety; Logic; Synthetic data; Multi-agent systems; Formal Methods; Temporal Logic; Natural Language Processing
Recommended citation: T. Zhang, J. Lyu and Y. Cai, "A Hierarchical Hybrid-Intelligence Architecture with Consensus, Debate, and Reflection for High-Fidelity NL-to-PPTL Conversion," 2025 32nd Asia-Pacific Software Engineering Conference (APSEC), Macau, China, 2025, pp. 69-80, doi: 10.1109/APSEC66846.2025.00018.
Download Paper