Automaton-Based Representations of Task Knowledge from Generative Language Models
This work addresses the gap between natural-language descriptions and formal automata for tasks like everyday activities and specialized domains, though it appears incremental in combining existing GLM and verification techniques.
The paper tackles the problem of obtaining high-level task knowledge for automaton-based representations in sequential decision-making by proposing GLM2FSA, an algorithm that constructs finite state automata from natural-language task descriptions using generative language models, enabling formal verification and iterative refinement.
Automaton-based representations of task knowledge play an important role in control and planning for sequential decision-making problems. However, obtaining the high-level task knowledge required to build such automata is often difficult. Meanwhile, large-scale generative language models (GLMs) can automatically generate relevant task knowledge. However, the textual outputs from GLMs cannot be formally verified or used for sequential decision-making. We propose a novel algorithm named GLM2FSA, which constructs a finite state automaton (FSA) encoding high-level task knowledge from a brief natural-language description of the task goal. GLM2FSA first sends queries to a GLM to extract task knowledge in textual form, and then it builds an FSA to represent this text-based knowledge. The proposed algorithm thus fills the gap between natural-language task descriptions and automaton-based representations, and the constructed FSA can be formally verified against user-defined specifications. We accordingly propose a method to iteratively refine the queries to the GLM based on the outcomes, e.g., counter-examples, from verification. We demonstrate GLM2FSA's ability to build and refine automaton-based representations of everyday tasks (e.g., crossing a road), and also of tasks that require highly-specialized knowledge (e.g., executing secure multi-party computation).