Position: a path to legal coherence through language models

Law, administration, and jurisprudence are riddled with inconsistencies. However, there may be a technological remedy (we do not say cure). In this article, we i) quantitatively illustrate incoherence in immigration courts and frame coherence in the context of law; ii) provide evidence that large language models (LLMs) can accurately compile local information into natural data structures that enable coherence-driven inference (CDI); iii) outline how efficient computation of globally coherent substructures enables useful forms of machine cognition; and iv) discuss how the technology we describe can most plausibly be developed and deployed with an eye towards legal and governmental applications.