egov-laws-jp

The statutes, cabinet orders, ministerial ordinances and rules of Japan that e-Gov 法令検索 distributes as its bulk download, one row per law revision, with the original XML and a plain-text extraction side by side.

CC BY 4.0. The text of a statute is not a copyrightable work under Article 13 of the Copyright Act, and what the Digital Agency adds (titles' readings, categories, the revision history) is published under PDL1.0, which is compatible with CC BY 4.0.

What is here

One table, laws, 10,679 rows.

rows (law revisions)	10,679
distinct laws (law_id)	8,997
revisions not yet in force (is_unenforced)	1,676
characters of text	573,048,356
characters of xml	2,476,643,432

By law_type:

law_type	revisions	laws
府省令 (ministerial ordinance)	4,452	4,098
法律 (act)	3,225	2,101
政令 (cabinet order)	2,518	2,324
規則 (rule)	416	407
勅令 (imperial order)	67	66
憲法 (constitution)	1	1

Columns:

column
revision_id	{law_id}_{enforcement date}_{amending law id}, unique per row
law_id	the law's id, shared by all of its revisions
law_type	憲法 / 法律 / 政令 / 勅令 / 府省令 / 規則
law_num, law_title, law_title_kana, old_law_title	as in the publisher's list. old_law_title is null for most rows
promulgation_date, amendment_promulgation_date, enforcement_date	the publisher's own text, in the era calendar with kanji numerals
*_iso	the same three dates as YYYY-MM-DD
amendment_law_title, amendment_law_num	the law that produced this revision. Null for a law's first revision
enforcement_note	the publisher's remark on the enforcement date
is_unenforced	true when the revision is promulgated but not yet in force
source_url	the page for this revision on laws.e-gov.go.jp
xml	the revision's XML, exactly as distributed
text	every text node of the XML, one per line, with ruby readings removed

The thing to know before training on it

Every revision is a full copy of the law. A law amended thirty times appears thirty times, and the copies differ in a handful of articles. The most revised law here has 33 revisions. If the table is used as a corpus as it stands, the large laws dominate it.

Keeping one revision per law (the latest one in force) leaves 8,997 rows, 163 million characters of text against 573 million for the whole table:

-- DuckDB, straight off the Parquet
WITH r AS (
  SELECT *, row_number() OVER (PARTITION BY law_id
                               ORDER BY enforcement_date_iso DESC) AS rn
  FROM read_parquet('parquet/laws.parquet')
  WHERE NOT is_unenforced
)
SELECT law_id, law_title, text FROM r WHERE rn = 1;

Every law has at least one revision in force, so this keeps all 8,997 laws.

The history only reaches back to 2015-08-01 for acts, orders and imperial orders, and to 2016-10-01 for ministerial ordinances and rules, because that is where e-Gov started keeping revisions. Before those dates only the version in force on that day exists.

What text is and is not

It is the text nodes of the XML joined with newlines. Article and paragraph boundaries, and the boundary between the main provisions and the supplementary provisions, are lines and nothing more. Read xml when the structure matters.

Two things are in it that a reader may not want:

• Tables are split cell by cell, so a schedule appears as a column of word fragments. A few laws are mostly schedules (the customs tariff, the income tax tables) and their text is mostly such fragments.
• Formulas are split the same way.

The 27,808 ruby readings are removed from text and kept in xml.

Images are not here. <Fig src="./pict/..."> refers to a picture that the bulk download does not contain, so 42,233 figures are references to nothing.

Reading it

import pyarrow.parquet as pq

# the rows are heavy (the XML of the largest law is 17 MB); read the columns you need
t = pq.read_table("parquet/laws.parquet", columns=["revision_id", "law_title", "text"])

What is not here

• Repealed and lapsed laws. The bulk download carries only laws in force or coming into force. The API lists 559 more, which are not in this table.
• Laws before e-Gov's start dates above, and any revision before them.
• 告示, 訓令, 通達, 条例, 規程. e-Gov does not carry them.
• Images and attachments.

Reproducing it

git clone https://github.com/yuiseki/egov-laws-jp
cd egov-laws-jp
uv sync
# data/raw/all_xml.zip from https://laws.e-gov.go.jp/bulkdownload?file_section=1&only_xml_flag=true
uv run scripts/02_export_parquet.py

data/provenance.yaml records the md5 and size of the archive this table was made from.

Licence

CC BY 4.0.

出典：e-Gov法令検索（デジタル庁）
https://laws.e-gov.go.jp/
（2026年9月16日時点のデータを加工）

The publisher applies PDL1.0, 公共データ利用規約（第1.0版）, whose clause 1.7(3) states that it is compatible with CC BY 4.0.

PDL1.0 asks that a modification be declared apart from the attribution. The modifications are listed in LICENSE beside this file, in Japanese as the terms ask. None of them changes a value in xml.