Japanese Text Analyzer – Tokenizer, Furigana, Romaji & Keywords

Turn Japanese text into structured data: morphological tokens with part of speech, kana readings, furigana, romaji, keywords and text statistics. Paste texts or enrich any existing Apify dataset (Google Maps reviews, YouTube comments, product titles, company names…) in one click.

Runs fully offline inside the Actor (kuromoji.js + IPADIC dictionary). No API key, no external AI service, no data sent anywhere. About 100 texts of 800 characters per second.

What you get for each text

Field	Example (input: 東京都で美味しいラーメンを食べました。)
normalizedText	Full-width/half-width unified (NFKC), invisible characters removed
readingHiragana	とうきょうとでおいしいらーめんをたべました。
readingKatakana	トウキョウトデオイシイラーメンヲタベマシタ。
furigana	東京[とうきょう]都[と]で美味[おい]しいラーメンを食[た]べました。 (or <ruby> HTML)
romaji	toukyouto de oishii raamen o tabemashita.
keywords	[{"keyword":"ラーメン","count":1},{"keyword":"東京都","count":1}] – nouns and compound nouns by frequency
stats	characters by script (kanji / hiragana / katakana / latin / digits), kanji ratio, sentences, average sentence length, token count, isJapanese
tokens (optional)	surface, part of speech in Japanese and English, base form (lemma), reading, pronunciation, conjugation

Use cases

• Review & comment analysis – run a Google Maps / Tabelog-style review or YouTube comment scraper, then point this Actor at its dataset to get keywords and lemmas for word clouds, topic counts and sentiment pipelines.
• Romaji / kana for names and product titles – generate readings and romaji for Japanese names, shop names or product lists (e.g. to build search indexes or English-friendly CSVs).
• Furigana for learning content – annotate kanji for Japanese learners, subtitles or easy-reading pages (plain 漢字[かんじ] or <ruby> HTML).
• Search & RAG preprocessing – Japanese has no spaces; tokens and base forms make keyword search, n-gram indexes and LLM chunking work much better.
• Text quality checks – kanji ratio and sentence length show how hard a text is to read.

Input

Analyze texts directly:

{ "texts": ["東京都で美味しいラーメンを食べました。", "鈴木一郎"] }

Enrich another run's dataset:

{ "datasetId": "YOUR_DATASET_ID", "textField": "text", "idField": "reviewId", "includeTokens": false }

Dot notation works for nested fields ("review.text"). Turn on Keep all original fields to get the source item plus the analysis in one row.

Output example

{
  "index": 2,
  "text": "鈴木一郎",
  "readingHiragana": "すずきいちろう",
  "furigana": "鈴木[すずき]一郎[いちろう]",
  "romaji": "suzuki ichirou",
  "keywords": [{ "keyword": "鈴木一郎", "count": 1 }],
  "stats": { "charCount": 4, "kanjiRatio": 1, "sentenceCount": 1, "isJapanese": true }
}

Accuracy notes (honest)

• Readings come from the IPADIC dictionary (2007). Common words are very accurate; rare personal names, new slang and brand names may get a wrong or missing reading (unknown words keep their original characters).
• Romaji is word-separated wapuro style (toukyou, raamen), with particles は/を/へ written as wa/o/e. It is not strict Hepburn with macrons.
• Keywords are frequency-based (no AI); they work best on texts of a few sentences or longer.

Pricing

Pay per event: $0.0005 per analyzed text ($0.50 per 1,000 texts) plus a tiny start fee. Empty texts are skipped and not charged. Use Max texts to cap the cost.

Licenses

kuromoji.js (Apache-2.0), WanaKana (MIT) and the mecab-ipadic dictionary (NAIST license, free for commercial use; notice in THIRD_PARTY_NOTICES.md). Your texts are processed only inside your own Actor run.

Support

Missing a feature (user dictionary, Hepburn with macrons, sentence splitting output)? Open an issue on the Issues tab.