Markdown for LLMs

Languages: labels, forms, and editions

The source Markdown for this article. Copy it into your assistant or download it as a text file.

# Languages: labels, forms, and editions

Law is written in words, and the corpus speaks many languages. This page
covers how declarations carry human-readable text — labels with mandatory
status, grammatical forms, translation overlays, and multilingual source
editions — without letting prose leak into semantics.

No commands are needed to follow this page. Each claim carries a status
in words with a letter mark; the full legend is on the
[topic index](/corpus/#how-this-topic-marks-confidence).

## Labels

Any declaration with a body can carry multilingual labels — **available
in the public release** (S). Each label names a language tag, a
mandatory status, and the text. The status is one of three values:

- `official` — the wording of the source itself.
- `unofficial` — project wording with no source authority.
- `translation` — a rendering of an official label into another language.

For example, a parcel entity with its English source wording and a
Kazakh translation:

```law
entity Parcel {
  label en official "parcel";
  label kk translation "жер телімі";
}
```

Labels stay outside the semantic and artifact hashes, so rewording prose
never changes what the rules mean; they stay inside the node content
hash, so label edits are still tracked (**S**). Parameter labels and case
entity labels follow the same pattern, each indexed under its own stable
identity (**S**).

## Grammatical forms

Some languages need more than a bare string to render fluent sentences.
Declarations may attach a closed per-family dictionary of grammatical
forms: five cases in two numbers for Russian-family tags, six cases plus
possessive forms for Kazakh-family tags, plural and possessive for
English-family tags, and an empty dictionary for all other tags (**S**).
An unknown slot is a warning, and a missing form degrades to the plain
label text — the renderer never invents morphology.

## Translation overlays

Per-language JSON files may sit beside a package manifest to supply
labels without touching the package sources. Reading overlays is
**available in the public release** (S); authoring them **requires a
source checkout** (I). Overlays are never compiled into the package; readers
pick them up by repeatable flags and resource references, and the
full-profile checks verify that overlays match the package they annotate
(**G**). Observed in the corpus: up to seven overlay languages beside a
single constitutional text package.

## Verbalization packs and review

One pack per language turns labels, forms, and templates into sentences;
fourteen languages ship today (**S**). Language is a review axis rather
than a property: reaching the top review rung needs full coverage in one
language, not partial coverage in many (**S**).

Coverage of labels and verbalization totality is held by dedicated checks
— a **full-profile CI check** (G). The label check compares each
package with a recorded minimum and refuses any drop below it; the
minimum is kept in `corpus/ratchets/label-ratchet.json`. Authoring
tooling itself stays English-only; canon data is never translated by the
pipeline (**T**).

## Multilingual editions

One revision of a source in two languages is two editions, linked by a
declared `align` block rather than by assumed meaning identity (**S**).
Four separate questions about that link have four separate answers,
kept in sync with the alignment notes and with the
[bridges page](/corpus/vocabulary/bridges/#edition-alignment):

- Grammar: the declaration is supported and compiles in real packages
  (**S**).
- Queries: no structural query reads align rows back yet, so tooling
  cannot list them (**S**).
- Corpus use: broad — over two thousand blocks across dozens of source
  files (**S**).
- Semantic effect: none — the block stays document metadata and never
  becomes rule content (**S**).

Each edition carries its own article-level fragments with pinned text
and status. Observed in the corpus: a constitution package with
parallel Russian and Kazakh editions and per-article fragments in both
languages, and historic code packages with classical Arabic fragments
(**S**).

## Where to go next

- The companion page on [contracts](/corpus/vocabulary/contracts/) covers predicate templates, which
  render through labels.
- The companion page on [bridges](/corpus/vocabulary/bridges/) covers edition-level alignment.
- For construct-level detail, see [the Vocabulary construct reference](/constructs/vocabulary/).
- For querying label rows structurally, see [LawQL](/lawql/).