adiletcodex

+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.avro filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.lz4 filter=lfs diff=lfs merge=lfs -text
+*.mds filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+# Audio files - uncompressed
+*.pcm filter=lfs diff=lfs merge=lfs -text
+*.sam filter=lfs diff=lfs merge=lfs -text
+*.raw filter=lfs diff=lfs merge=lfs -text
+# Audio files - compressed
+*.aac filter=lfs diff=lfs merge=lfs -text
+*.flac filter=lfs diff=lfs merge=lfs -text
+*.mp3 filter=lfs diff=lfs merge=lfs -text
+*.ogg filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text
+# Image files - uncompressed
+*.bmp filter=lfs diff=lfs merge=lfs -text
+*.gif filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.tiff filter=lfs diff=lfs merge=lfs -text
+# Image files - compressed
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+*.webp filter=lfs diff=lfs merge=lfs -text
+# Video files - compressed
+*.mp4 filter=lfs diff=lfs merge=lfs -text
+*.webm filter=lfs diff=lfs merge=lfs -text

CODEBOOK.md ADDED Viewed

	@@ -0,0 +1,101 @@

+# AdiletCodex - Codebook
+One record = one article of one document in one language. Files: `adiletcodex.jsonl`
+(nested fields kept) and `adiletcodex.csv` (flat; nested fields omitted).
+## Identity & language
+| Field | Type | Description |
+|---|---|---|
+| `doc_id` | string | adilet NGR of the document (same id across languages), e.g. `K1400000226` |
+| `doc_type` | string | `code` or `law` |
+| `slug` | string\|null | short mnemonic for codes (e.g. `criminal`), null for laws |
+| `lang` | string | `rus` (Russian), `kaz` (Kazakh), `eng` (English) |
+| `title` | string | document title in this language |
+| `url` | string | canonical adilet URL |
+## Document metadata (from the adilet requisites page)
+The adilet document "passport" is published in **Russian only**. In v2.0 several passport fields
+are **localized** to the row's `lang`; each keeps an `*_ru` companion with the original Russian
+value for provenance. Russian (`rus`) rows keep the Russian value in the field itself. Two
+localization methods are used, and they are **not** equivalent in authority:
+- **curated term map** - hand-reviewed official translations (`act_form`, `adopting_organ`,
+  `database_section`);
+- **reference metadata (machine-translated)** - translated by local Qwen3.5-122B-A10B (thinking
+  disabled) for convenience (filtering / search), **not** authoritative, with the original
+  Russian kept in the `*_ru` companion (`legal_sphere`, `publication_reference`); the Kazakh
+  output uses correct Kazakh legal terminology. The substantive text (`article_text`, `title`,
+  etc.) is the official adilet translation, not machine translation.
+Fields still marked **Russian-by-source** are not translated at all.
+| Field | Type | Language | Description |
+|---|---|---|---|
+| `status` | string | per-language | e.g. `upd` (updated / in force). Use this for validity / in-force state. |
+| `act_form` | string | **localized (curated)** | form of act, localized to `lang`: EN `Code`/`Law`/`Constitution`/`Constitutional Law`; KK `Кодекс`/`Заң`/`Конституция`/`Конституциялық заң`; RU `Кодекс`/`Закон`/`Конституция`/`Конституционный закон` (RU may include amendment-noise values). |
+| `act_form_ru` | string | Russian source | original Russian `act_form` from the adilet card, kept on every row for provenance (v2.0). |
+| `legal_sphere` | string | **reference metadata (machine-translated)** | subject area / legal-relations sphere; local Qwen3.5-122B-A10B MT (254 unique values), convenience only. |
+| `legal_sphere_ru` | string | Russian source | original Russian `legal_sphere`, kept on every row for provenance (v2.0). |
+| `database_section` | string | **localized (curated)** | adilet database section (thematic classifier), 35 fixed sections; curated EN/KK map. |
+| `database_section_ru` | string | Russian source | original Russian `database_section`, kept on every row for provenance (v2.0). |
+| `adopting_organ` | string\|null | **localized (curated)** | body that adopted the act (Parliament / President, with former-name notes); curated EN/KK map. |
+| `adopting_organ_ru` | string\|null | Russian source | original Russian `adopting_organ`, kept on every row for provenance (v2.0). |
+| `state_registry_number` | string | - | number in the State Register of NLAs |
+| `npa_registration_number` | string | - | act's own number, e.g. `226-v` |
+| `adoption_date` | string | - | dd.mm.yyyy |
+| `change_date` | string\|null | - | date of last amendment (null if never amended) |
+| `publication_date` | string\|null | - | official publication date |
+| `publication_reference` | string | **reference metadata (machine-translated)** | official publication source(s); local Qwen3.5-122B-A10B MT (333 unique values) with a digit/date/number-preservation check (Russian kept on mismatch), convenience only. |
+| `publication_reference_ru` | string | Russian source | original Russian `publication_reference`, kept on every row for provenance (v2.0). |
+| `adoption_place` | string | **Russian-by-source** | e.g. `г. Астана` |
+| `region_action` | string | **Russian-by-source** | territorial scope |
+| `citation` | string | per-language | human-readable source citation of the act (e.g. `Code of the Republic of Kazakhstan dated 20 June, 2003 № 442`); `doc_date` / `doc_number` are parsed from it. |
+| `doc_date`, `doc_number` | string\|null | - | adoption date (ISO `yyyy-mm-dd`) / number parsed from `citation` |
+> **Removed in v2.0:** `legal_force`. It duplicated `act_form` in 323/343 documents and was
+> mislabeled - it was never a validity indicator. Use `status` for validity and `act_form` for
+> the type of act. See `../CHANGELOG.md`.
+## Structural hierarchy & article
+| Field | Type | Description |
+|---|---|---|
+| `part` | string\|null | top level (e.g. `ОБЩАЯ ЧАСТЬ` / `SPECIAL PART` / `Жалпы бөлік`) |
+| `section` | string\|null | Раздел / N-бөлім / Section |
+| `chapter` | string\|null | Глава / N-тарау / Chapter |
+| `article_id` | string\|null | internal HTML anchor id |
+| `article_no` | string\|null | article number (e.g. `99`, inserts like `50-1`) |
+| `article_title` | string | full article heading |
+| `article_text` | string | full article body (concatenated points) |
+| `paragraphs` | list | numbered points: `{id, number, text, links}` (JSONL only) |
+| `notes` | list | amendment/editorial footnotes: `{text, links}` (JSONL only) |
+| `links` | list | article-level hyperlinks (JSONL only) |
+## Notes
+- Languages align by (`doc_id`, `article_no`). A document keeps the same `doc_id` across
+  languages.
+- Gaps in `article_no` within a document correspond to repealed articles.
+- English exists for most but not all documents (adilet does not translate everything):
+  296 of 343 documents are present in all three languages, 37 in Kazakh+Russian, 5 in Kazakh
+  only, 3 in Russian only, 2 in Kazakh+English.
+- `adopting_organ` / `adopting_organ_ru` are null for 288 records (documents whose adilet card
+  lists no adopting body); `publication_reference` / `publication_reference_ru` are empty for
+  127 records (documents with no publication source recorded).
+## Known limitations
+- **Same-number articles: fixed in v2.0.** The v1.1 pre-release audit found 19 (`doc_id`,
+  `lang`, `article_no`) groups with more than one record, from three causes: (a) a displayed
+  number that recurs across structural parts; (b) entry-into-force / amendment clauses inside
+  "Final and transitional provisions" (or enactment) articles that the parser emitted as
+  separate records (their `article_title` was a clause fragment, e.g. `article 38;`); and (c)
+  compound inserted numbers (e.g. `Article 15-24`) whose `article_no` had been truncated to the
+  base number (`15`). All 19 were repaired in v2.0 without dropping any wording: (c) was fixed
+  by restoring the full `article_no` (confirmed against the aligned Kazakh/Russian rows); (b)
+  and the stale same-language duplicates were merged back into the correct parent article and
+  the empty fragment removed. After the fix `article_no` is unique within every (`doc_id`,
+  `lang`) group - 0 duplicate keys remain. See `../CHANGELOG.md` and
+  `../localization/fix_duplicate_articles.py`.
+- **Reference-metadata fields are machine translation.** `legal_sphere` and
+  `publication_reference` (and only these) are translated by a local LLM for
+  filtering/search convenience and are not authoritative; the original Russian is always kept in
+  the `*_ru` companion column. The substantive article content (`title`, `status`,
+  `article_title`, `article_text`) is the official adilet.zan.kz translation.

LICENSE ADDED Viewed

	@@ -0,0 +1,20 @@

+AdiletCodex - License
+======================
+Dataset compilation (this dataset: the selection, parsing, article segmentation,
+cross-language alignment, and metadata):
+  Creative Commons Attribution 4.0 International (CC BY 4.0)
+  Full text: https://creativecommons.org/licenses/by/4.0/legalcode
+You are free to share and adapt the material for any purpose, provided you give
+appropriate credit (cite the dataset, see README_DATASET.md) and indicate if changes
+were made.
+Underlying legal texts:
+The article texts are official normative legal acts of the Republic of Kazakhstan,
+which are public. They were obtained from the official Information and Legal System
+adilet.zan.kz. Please retain attribution to adilet.zan.kz. This is an unofficial
+research copy; the authoritative, in-force text is the one published on adilet.zan.kz.
+The dataset is provided "as is", without warranty of any kind.

README.md ADDED Viewed

	@@ -0,0 +1,133 @@

+---
+license: cc-by-4.0
+language:
+- kk
+- ru
+- en
+multilinguality: multilingual
+pretty_name: 'AdiletCodex: Trilingual (KK/RU/EN) Article-Level Corpus of Kazakh Legal Codes and Laws'
+tags:
+- legal
+- law
+- legal-nlp
+- kazakhstan
+- kz
+- kazakh
+- russian
+- english
+- legislation
+- statute
+- codes
+- laws
+- adilet
+- article-level
+- trilingual
+- multilingual
+- low-resource
+- central-asia
+- legal-information-retrieval
+task_categories:
+- text-retrieval
+- text-classification
+- text-generation
+size_categories:
+- 10K<n<100K
+---
+# AdiletCodex: A Trilingual (Kazakh-Russian-English) Article-Level Corpus of the Codes and Laws of Kazakhstan
+A full-text, **article-level**, **trilingual (Kazakh + Russian + English)** corpus of the
+currently in-force codes and substantive laws of the Republic of Kazakhstan, parsed from the
+official legal database [adilet.zan.kz](https://adilet.zan.kz).
+- **65,807** article records across **343** documents (**24 codes + 319 laws**, incl. the Constitution)
+- **296** documents present in all three languages (RU 22,112 / KK 22,098 / EN 21,597)
+- Adoption years **1991-2026**
+- Each record: full article text + structural hierarchy (part/section/chapter/article) +
+  numbered points + amendment notes + cross-reference links + rich document metadata
+  (act form, legal sphere, adopting body, registration numbers, dates, status)
+To our knowledge this is the **first article-structured, trilingual, full-text** corpus of
+Kazakh legislation.
+## Why this corpus
+Kazakh legal NLP has lacked a **structured, full-text** foundation: existing resources are
+either document-level metadata, Russian-only, or downstream QA sets built on top of statutes
+they do not themselves release. AdiletCodex fills that gap with the **article** as the unit of
+record, the **three official language versions** side by side, and the **structural +
+amendment + metadata** context needed for legal information retrieval, cross-lingual and RAG
+grounding, translation-fidelity study, and temporal ("law-as-of-date") modeling.
+## Related datasets and how AdiletCodex differs
+- **KazakhLawCorpus** (Tleubayeva, *Data* MDPI 2026; 215,889 records) - document-level
+  **metadata only**, Kazakh only, no article text. AdiletCodex is complementary and strictly
+  more granular: full text, article-level, trilingual.
+- **justicedao/ipfs_kazakhstan_laws_ir** (HF, ~399k rows) - article-level but **Russian only**,
+  undocumented, no hierarchy / amendments / status / KK / EN. AdiletCodex adds the two other
+  official languages, the structural hierarchy, amendment notes, status, metadata, and docs.
+- **olzhasAl/adilet-legal-qa-kz** (HF, 13,159 QA pairs, RU+KK) and other Kazakh legal-QA / LoRA
+  sets - **downstream tasks**, not a source corpus. AdiletCodex provides the underlying statute
+  text those tasks are grounded in.
+- **Uzbek legal corpus** (Tomaris AI) - a methodological twin for a neighbouring jurisdiction
+  (Uzbek-Latin only, no descriptor paper); AdiletCodex is the Kazakh, trilingual counterpart.
+- **RusLawOD** (HF irlspbru/RusLawOD, 304k texts) - Russian national statutory corpus; a
+  language / scope comparator, not Kazakh.
+## Files
+- `adiletcodex.jsonl` - one article per line (keeps nested paragraphs/notes/links)
+- `adiletcodex.csv` - flat columns
+## Load
+```python
+from datasets import load_dataset
+ds = load_dataset("json", data_files="adiletcodex.jsonl", split="train")
+```
+## Schema
+`doc_id, doc_type, slug, lang, title, status, act_form, act_form_ru, legal_sphere, legal_sphere_ru,
+adopting_organ, adopting_organ_ru, database_section, database_section_ru, state_registry_number,
+npa_registration_number, adoption_date, change_date, publication_date, publication_reference,
+publication_reference_ru, adoption_place, region_action, citation, doc_date, doc_number, part,
+section, chapter, article_id, article_no, article_title, article_text, paragraphs, notes, links, url`
+(37 fields total; `paragraphs`, `notes`, `links` are nested and JSONL-only, so the CSV carries
+the 34 scalar fields. `citation` is the human-readable source citation, with `doc_date` /
+`doc_number` parsed from it.)
+**Language of metadata fields.** Only `title`, `status`, `article_title`, `article_text` are
+truly per-language and are the **official** adilet translations. The document-"passport" fields
+come from adilet's Russian-only document card; in v2.0 several are **localized** to the row's
+language, each with an `*_ru` companion preserving the Russian source (Russian rows keep the
+Russian value):
+- **Curated term map** (hand-reviewed, not machine translation): `act_form`, `adopting_organ`,
+  `database_section`.
+- **Machine-translated (local Qwen3.5-122B-A10B, thinking disabled) reference metadata** -
+  convenience only, not authoritative, with the Russian source kept in the `*_ru` companion:
+  `legal_sphere`, `publication_reference`. The Kazakh output uses correct Kazakh legal
+  terminology (the earlier Qwen2.5-32B Kazakh was replaced). A number-preservation check
+  protects all digits/dates/article numbers in `publication_reference` (Russian original kept
+  on any mismatch).
+- **Still Russian-by-source:** `region_action`, `adoption_place`.
+Use `status` for validity/in-force state. Reproducibility artifacts (term + translation maps,
+QA-failure log, scripts) are in `localization/`.
+> **Changed in v2.0 (major):** `act_form`, `adopting_organ`, `database_section` are localized per
+> `lang` via a curated map, and `legal_sphere`, `publication_reference` via machine translation
+> (all with `*_ru` provenance companions); the redundant/mislabeled `legal_force` field was
+> removed (use `status` for in-force state). Separately, 18 parser-artifact records (split-off
+> entry-into-force / amendment fragments, a quoted old-code article, truncated compound article
+> numbers, and stale same-language duplicates) were repaired without dropping any wording, so the
+> record count is 65,807 (was 65,825 in v1.0) with 0 duplicate `(doc_id, lang, article_no)` keys.
+## License & source
+Compilation: CC-BY-4.0. Underlying texts are official public acts of the Republic of
+Kazakhstan (source: adilet.zan.kz). Unofficial research copy; verify in-force text on adilet.
+## Citation
+Mukhsimbayev, B., Pak, A., & Kuralbayev, A. (2026). *AdiletCodex: A Trilingual (Kazakh-Russian-English)
+Article-Level Corpus of the Codes and Laws of Kazakhstan* (v2.0) [Data set]. Zenodo.
+https://doi.org/10.5281/zenodo.22812625 (concept DOI, always resolves to the latest version).
+Latest is v2.0 (version DOI 10.5281/zenodo.23260689, published 2026-10-09); v1.0 version DOI:
+10.5281/zenodo.22812626.

adiletcodex.csv.gz ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:07b163eb48cbfb8f345c6b77f8d3b1073cf3407a0293cc0379f36022e6a58c4d
+size 36317305

adiletcodex.jsonl.gz ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:1573b2a7be5ec9ac7d584f69bf3418f5d1da22936331ddbbb3a8ccacc322d44b
+size 52383106