Large language models are increasingly used as formalizers: given a reasoning problem in natural language, they produce a formal specification that a symbolic solver can then solve, with guarantees relative to that specification. But those guarantees only hold if the specification itself is correct, and LLM-generated formalizations can contain errors.
A new paper on arXiv proposes repairing such LLM-generated specifications using deterministic diagnosis. The abstract describes this as a way to fix flawed formal output, rather than simply accepting or regenerating it. The method appears to target the gap between natural-language intent and the formal artifact the solver receives.
The abstract is truncated, so the specific diagnostic and repair mechanisms are not fully described in the source. Still, the stated direction is clear: making the formalization pipeline more robust by detecting and correcting specification errors in a deterministic, non-random way.