deontic

A constraint language for obligation, permission and prohibition — the things a system must do, may do, and must not do.

Status: alpha. The grammar is implemented and reproduces the whole conformance corpus: deontic.parse gives the AST, deontic.load reads a dictionary and deontic.resolve checks a sentence against it. There is no evaluator in this package, by design; the corpus's evaluate/ cases are for implementations that add one.

Why a language rather than a library

Deontic statements are usually buried in imperative checks, where the rule and the enforcement of the rule are the same code. Writing them down as data makes them reviewable by the people who own the policy rather than only by the people who own the codebase, and lets more than one evaluator agree on what a rule means.

Design

docs/design.md states what the language is meant to be: the closed-class vocabulary, the shapes, filters and paths, attestation, the storage form, the interpretation rules, the outcome model, and how it relates to Attempto Controlled English and to deontic logic.

Conformance

conformance/ is the normative test data: source text plus the result every implementation must agree on, kept as data rather than Python tests so that an implementation in another language can consume it too. An implementation is conformant when it reproduces the corpus. It has three parts — parse/ (source to AST, no lexicon needed), reject/ (source to error) and evaluate/ (sentences, a world and a date to outcomes) — over one invented fixture lexicon, and its README records the language decisions the cases fix.

Use

import deontic

ast = deontic.parse("Every $exhibit$ tagged fragile must @be located in@ a $gallery$ tagged storage.")
# {'shape': 'obligation', 'subject': {...}, 'predicate': {...}}

lexicon = deontic.load("conformance/lexicon.json")
deontic.resolve(ast, lexicon)   # raises a ResolutionError with the corpus's error code otherwise

Sentences are authored as English and stored with markers on the typed spans ($Term$, @verb@, #90 days#, <parameter>); editors write the markers, people do not. A verb's meaning is a pattern in the dictionary or, failing that, code under the contract in docs/hooks.md.

Install

pip install deontic

Stdlib only, no runtime dependencies — a grammar meant to be implemented by other people should not put a parser generator in everyone's dependency tree.

License

MIT.