Syntax Is Easy, Semantics Is Hard: Evaluating LLMs for LTL Translation
2026-04-08 • Logic in Computer Science
Logic in Computer ScienceArtificial Intelligence
AI summaryⓘ
The authors studied how well Large Language Models (LLMs) can turn English sentences into Linear Temporal Logic (LTL) formulas, which are used to describe software and security rules. They found that LLMs are better at copying the structure (syntax) than understanding the meaning (semantics) of these formulas. More detailed instructions helped the models, and treating the problem like writing Python code made their translations much better. The paper also talks about the challenges of fairly measuring this ability and suggests directions for future research.
Linear Temporal LogicLTLLarge Language Modelsnatural language processingformal verificationsecurity policiessyntaxsemanticscode completion
Authors
Priscilla Kyei Danso, Mohammad Saqib Hasan, Niranjan Balasubramanian, Omar Chowdhury
Abstract
Propositional Linear Temporal Logic (LTL) is a popular formalism for specifying desirable requirements and security and privacy policies for software, networks, and systems. Yet expressing such requirements and policies in LTL remains challenging because of its intricate semantics. Since many security and privacy analysis tools require LTL formulas as input, this difficulty places them out of reach for many developers and analysts. Large Language Models (LLMs) could broaden access to such tools by translating natural language fragments into LTL formulas. This paper evaluates that premise by assessing how effectively several representative LLMs translate assertive English sentences into LTL formulas. Using both human-generated and synthetic ground-truth data, we evaluate effectiveness along syntactic and semantic dimensions. The results reveal three findings: (1) in line with prior findings, LLMs perform better on syntactic aspects of LTL than on semantic ones; (2) they generally benefit from more detailed prompts; and (3) reformulating the task as a Python code-completion problem substantially improves overall performance. We also discuss challenges in conducting a fair evaluation on this task and conclude with recommendations for future work.