Search papers, labs, and topics across Lattice.
This paper introduces a novel method for translating natural language (NL) task descriptions into Signal Temporal Logic (STL) formulas by explicitly preserving structural ambiguities during the translation process. They achieve this through a three-stage pipeline using Combinatory Categorial Grammar (CCG) for ambiguity-preserving parsing, template-based semantic composition, and canonicalization with score aggregation. The method generates a set of scored STL candidates, representing multiple plausible interpretations of ambiguous NL instructions, improving the reliability of NL-to-STL translation.
Instead of forcing a single interpretation, this work embraces the inherent ambiguity of natural language to generate multiple plausible STL formulas from a single NL task description.
Signal Temporal Logic (STL) is widely used to specify timed and safety-critical tasks for cyber-physical systems, but writing STL formulas directly is difficult for non-expert users. Natural language (NL) provides a convenient interface, yet its inherent structural ambiguity makes one-to-one translation into STL unreliable. In this paper, we propose an \textit{ambiguity-preserving} method for translating NL task descriptions into STL candidate formulas. The key idea is to retain multiple plausible syntactic analyses instead of forcing a single interpretation at the parsing stage. To this end, we develop a three-stage pipeline based on Combinatory Categorial Grammar (CCG): ambiguity-preserving $n$-best parsing, STL-oriented template-based semantic composition, and canonicalization with score aggregation. The proposed method outputs a deduplicated set of STL candidates with plausibility scores, thereby explicitly representing multiple possible formal interpretations of an ambiguous instruction. In contrast to existing one-best NL-to-logic translation methods, the proposed approach is designed to preserve attachment and scope ambiguity. Case studies on representative task descriptions demonstrate that the method generates multiple STL candidates for genuinely ambiguous inputs while collapsing unambiguous or canonically equivalent derivations to a single STL formula.
