Hybrid AI System Bridges Natural Language and Answer Set Programming

A new hybrid AI system aims to simplify the interaction between humans and complex code by translating unconstrained English into Answer Set Programming (ASP) programs. ASP, a declarative programming paradigm, is a powerful tool for describing and solving combinatorial problems, but its syntax can be challenging for non-programmers. Researchers at the University of Dayton and the US Air Force have developed a system that combines Large Language Models (LLMs) and Abstract Meaning Representation (AMR) graphs to generate ASP rules, facts, and constraints. This approach minimizes the role of the LLM to straightforward tasks like simplifying sentences and identifying keywords, while AMR graphs are parsed from the simplified language to systematically generate ASP constraints. The system was successfully tested on example logic puzzles, demonstrating its ability to create runnable ASP programs that represent the original problem. Unlike methods that rely entirely on LLMs, this hybrid approach seeks to create a lighter-weight and more explainable system. By parsing AMR graphs and extracting relevant concepts, the system guarantees correct syntax given the proper inputs. Future work will focus on expanding the system's capabilities, improving consistency, and incorporating more complex logical concepts.