Law & Language: How Rules Learn to Mean

Savigny and the historical school

Friedrich Carl von Savigny, writing at the turn of the nineteenth century, rejected the idea that law was a timeless, deductive system that a legislature could simply invent. In On the Vocation of Our Age for Legislation and Jurisprudence (1814) he argued that law grows organically out of a people's life — its Volksgeist, its spirit — the same way language does. A language cannot be designed by committee and decreed into being; it lives in speakers, changes with usage, and dies when no one speaks it.

Law, Savigny held, is the same: it is not a closed formal system but a living practice carried by jurists and citizens. The insight that matters for us is not romantic nationalism but the analogy itself — rules, like words, acquire meaning through use, drift with time, and resist being fixed by flat.

Hart: rules about rules, and open texture

H. L. A. Hart's The Concept of Law (1961) gives the analogy a precise shape. Law, for Hart, is a union of two kinds of rule:

• Primary rules impose obligations — do not launder money, do not mislead investors, do not misreport.
• Secondary rules are rules about rules — how rules are recognised (a constitution, a statute book), how they are changed (legislation), and how they are adjudicated (courts).

Hart's crucial claim for systems design is open texture: natural-language terms always have a "penumbra" of uncertainty. Words like "suspicious", "reasonable", "beneficial", and "risk" have a settled core and a hazy fringe. At the fringe, judges exercise discretion. Open texture is not a defect of language; it is what lets law adapt to cases the legislator never imagined. But it is exactly the property that makes natural-language rules hard to execute mechanically.

The translation problem in compliance

Every AML and compliance rule faces the same journey: legislative sentence → regulator guidance → internal policy → executable predicate (a Boolean filter, a threshold, a machine-learning score). Each step of translation is a lossy compression of meaning.

• "Suspicious transaction" becomes a pattern rule about structuring around reporting thresholds.
• "Beneficial ownership" becomes a lookup across a corporate register with a cutoff at a percentage ownership stake.
• "Reasonable steps" becomes a documented due-diligence checklist.

The gap between what the words mean and what the predicate tests is where compliance risk actually lives. A model that exactly follows the predicate can still be entirely unfaithful to the rule — and a rule faithfully applied can still be wrong in open-textured cases. Understanding this gap is the difference between a compliance function that mechanically checks boxes and one that reasons about whether the boxes mean what the law says.

Semantic drift and the linguistic arms race

Words change meaning as the world changes, and law must chase them. "Money" once meant notes and coin; it now includes crypto-assets, stablecoins, and tokenised instruments. "Identity" once meant a paper document; it now means a verified credential in an ecosystem. Regulators respond through guidance, interpretive notes, and amendment — a linguistic arms race in which the regulated community and the regulator continuously renegotiate what the words mean.

Two modern movements respond directly to this. The plain-language movement tries to shrink open texture by drafting rules a citizen can read. The Rules-as-Code / legislation-as-code experiments (in New Zealand, Canada, and elsewhere) try to draft law in machine-executable form from the start, so that the translation loss described above shrinks to zero.

What a schema has to do with a statute

From a data-engineering perspective, the analogy is almost literal. A schema is a legislature for a database: it fixes the shape of the words (fields, types, allowed values) and the rules of change (versioning, migration, breaking-change review). A data contract between two systems is a treaty about meaning. The same open texture Hart found in statutes appears in schemas whenever a field is vaguely defined or a contract is silent on semantics. The discipline of governing a database's vocabulary is, in miniature, the discipline of governing a legal system's vocabulary. That is why this essay belongs in the law–language–data story: the problems of meaning are shared, and so are the tools.