Japanese Address & Company Name Normalizer

Turn messy Japanese company names and addresses into clean, joinable fields.

If you have ever tried to match ㈱アイウエオ against 株式会社アイウエオ, or to join 東京都渋谷区上原３丁目６番１２号 against 東京都渋谷区上原3-6-12, you know the problem: the same entity is written a dozen different ways, and a plain string comparison finds none of them.

This Actor splits both into structured parts so the join actually works.

No network requests. It is pure computation — nothing is scraped, no third-party site is called, no API key is needed. That means it cannot break when some website changes its HTML, and your data never leaves the run.

What it does

Company names

Input	normalized	corp_form	corp_position	body
(株)アイウエオ	株式会社アイウエオ	株式会社	prefix	アイウエオ
アイウエオ㈱	アイウエオ株式会社	株式会社	suffix	アイウエオ
カキク(有)	カキク有限会社	有限会社	suffix	カキク
ＮＰＯ法人タチツ	特定非営利活動法人タチツ	特定非営利活動法人	prefix	タチツ
株式会社髙島屋	株式会社高島屋	株式会社	prefix	高島屋
キマリテラス	キマリテラス	null	null	キマリテラス

• 15 legal forms are recognised, including the parenthesised and single-character variants (株式会社 / (株) / （株） / ㈱), 有限会社, 合同会社, 合資会社, 合名会社, 一般社団法人, 一般財団法人, 公益社団法人, 公益財団法人, 特定非営利活動法人 (NPO法人), 医療法人, 学校法人, 社会福祉法人, 宗教法人, 独立行政法人.
• Prefix and suffix position is preserved. 株式会社アイウエオ and アイウエオ株式会社 can be two different companies, so they are never collapsed into one value.
• Old-form kanji common in company names are folded to their modern forms (髙→高, 﨑→崎, 澤→沢, 邊/邉→辺, and 7 more — 12 characters in all).
• Full-width alphanumerics become half-width (ＡＢＣ → ABC); katakana is left full-width.

Addresses

Input	prefecture	city	town	block	building
東京都渋谷区上原３丁目６番１２号	東京都	渋谷区	上原	3-6-12	null
東京都渋谷区上原3-6-12	東京都	渋谷区	上原	3-6-12	null
神奈川県横浜市西区みなとみらい2-3-1	神奈川県	横浜市西区	みなとみらい	2-3-1	null
愛知県名古屋市中村区名駅1-1-4 JRゲートタワー	愛知県	名古屋市中村区	名駅	1-1-4	JRゲートタワー
渋谷区上原3-6-12	null	渋谷区	上原	3-6-12	null

• 丁目 / 番 / 号 and 1-2-3 collapse to the same block form, so the two spellings join.
• Designated cities keep 市 + 区 together as one unit (横浜市西区, not 横浜市).
• Building names, floors and room numbers are kept in building. They are not discarded — an address without them is not a deliverable address.
• Kanji numerals are converted only where they mean a chōme/banchi (丸の内一丁目9番1号 → 丸の内 + 1-9-1). Place names keep their kanji: 四谷 stays 四谷, not 4谷; 九段北, 八重洲, 三条通, 北一条西 are likewise untouched.

What it deliberately does NOT do

Read this before you buy — these are design decisions, not missing features.

• It never guesses. Given 渋谷区上原3-6-12 with no prefecture, it returns prefecture: null. Filling in 東京都 looks helpful until you meet a same-named ward in another prefecture and silently corrupt the join. Fields that could not be determined are null, never an empty string, so you can tell "not present" from "was blank".
• No validation, no geocoding, no postal codes. It does not check that an address exists, does not return latitude/longitude, and does not look up or verify 郵便番号. It reformats what you give it.
• No corporate registry lookup. It does not fetch 法人番号 or confirm a company exists.
• Kanji numerals up to 99. Enough for chōme/banchi; larger kanji numerals are left alone.
• Old-form kanji folding is a curated list, not an exhaustive JIS conversion table.

Input

{
  "records": [
    { "id": "1", "company": "(株)アイウエオ", "address": "東京都渋谷区上原３丁目６番１２号" },
    { "id": "2", "company": "カキク㈲" },
    { "id": "3", "address": "愛知県名古屋市中村区名駅1-1-4 JRゲートタワー" },
    "キマリテラス"
  ]
}

• company and address are both optional; supply either or both.
• id is echoed back untouched so you can join the results to your source rows.
• A plain string is treated as a company name.
• Rows with neither company nor address are skipped — and not charged.
• Up to 100,000 records per run.

Output

One dataset item per input record:

{
  "id": "1",
  "company": {
    "input": "(株)アイウエオ",
    "normalized": "株式会社アイウエオ",
    "corp_form": "株式会社",
    "corp_position": "prefix",
    "body": "アイウエオ"
  },
  "address": {
    "input": "東京都渋谷区上原３丁目６番１２号",
    "prefecture": "東京都",
    "city": "渋谷区",
    "town": "上原",
    "block": "3-6-12",
    "building": null,
    "normalized": "東京都渋谷区上原3-6-12"
  }
}

input always carries the original string back, so the output is a complete audit trail on its own.

Pricing

Pay per event: $0.50 per 1,000 records normalized.

You are charged only for records that produced a result. Empty input, malformed rows, and rows with neither company nor address cost nothing.

Typical uses

• Deduplicating a CRM or lead list where the same company was entered several ways.
• Joining two Japanese datasets that were typed by different people.
• Cleaning form submissions before they reach a database.
• Preparing an address column for a mail merge without losing the building name.