Grammar-Forced Translation of Natural Language to Temporal Logic using LLMs

2025-12-18

International Conference on Machine Learning

AI Summary

To address three key bottlenecks in natural language (NL) to temporal logic (TL) translation—imprecise atomic proposition (AP) extraction, challenging coreference resolution, and poor few-shot generalization—this paper proposes the first syntax-constrained, two-stage collaborative optimization framework. In the *lifting* stage, a theoretically grounded constrained decoding mechanism compresses the solution space to enhance learning efficiency. In the *translation* stage, TL syntactic structure guidance, AP extraction constraints, and domain-adaptive prompting are jointly integrated. Crucially, our method enables end-to-end joint optimization of lifting and translation—previously unachieved. Evaluated on CW, GLTL, and Navi benchmarks, it achieves an average 5.49% improvement in end-to-end accuracy and a 14.06% gain in cross-domain generalization accuracy.

Technology Category

• Natural Language Processing: Language Grounding & Multi-modal NLP
• Planning, Routing, and Scheduling: Planning with Language Models
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Translating natural language (NL) into a formal language such as temporal logic (TL) is integral for human communication with robots and autonomous systems. State-of-the-art approaches decompose the task into a lifting of atomic propositions (APs) phase and a translation phase. However, existing methods struggle with accurate lifting, the existence of co-references, and learning from limited data. In this paper, we propose a framework for NL to TL translation called Grammar Forced Translation (GraFT). The framework is based on the observation that previous work solves both the lifting and translation steps by letting a language model iteratively predict tokens from its full vocabulary. In contrast, GraFT reduces the complexity of both tasks by restricting the set of valid output tokens from the full vocabulary to only a handful in each step. The solution space reduction is obtained by exploiting the unique properties of each problem. We also provide a theoretical justification for why the solution space reduction leads to more efficient learning. We evaluate the effectiveness of GraFT using the CW, GLTL, and Navi benchmarks. Compared with state-of-the-art translation approaches, it can be observed that GraFT the end-to-end translation accuracy by 5.49% and out-of-domain translation accuracy by 14.06% on average.

Problem

• Improves NL to temporal logic translation accuracy
• Addresses challenges in atomic proposition lifting
• Enhances learning efficiency with reduced solution space

Innovation

• Grammar Forced Translation restricts valid output tokens
• Solution space reduction improves learning efficiency
• Framework enhances translation accuracy across benchmarks