Furigana are the small kana printed beside Japanese kanji to show how they are read, and software can now add them to any text. Ordinary words mostly survive the process. New baby names often do not. We ran the 21 names in the published 2025 top tens of Meiji Yasuda, a Japanese life insurer, through a dictionary from 2007. It matched the list's top reading on 4 and left a character unread on 2. On 15 it printed a complete reading the list does not have, and eleven of the spellings are not in it at all. In a 2010 NINJAL benchmark, by contrast, MeCab with the same dictionary misread 819 of 67,556 words across news, fiction and blogs, counting kana words as well as kanji ones.
A 2007 dictionary misreads few words in news, fiction and blogs
Japanese runs without spaces, so a program called a tokenizer first cuts the text into words, then takes each word's reading from a dictionary. The dictionary here is IPADIC, version 2.7.0-20070801, which the NINJAL paper tested under MeCab 0.98. Fiction was hardest at 379 of 22,620 words, or 1.68% on our arithmetic; newspaper text came in at 214 of 23,358, or 0.92%. The denominator is every word, particles included, and kana words count as correct even when unknown. The rate among kanji words, the ones furigana exist for, is probably higher, though the paper gives no split to check that.
Japan's 2010 Jōyō kanji table offers a second test. Its appendix, the 付表, lists words whose readings cannot be built from their characters, such as 今日 きょう: 123 written forms by our parse. Typed into the neutral sentence これは〇〇です。 on our own page's code, 108 came back with the table's reading. Many of the rest are other real words, as 五月 ごがつ is where the table gives さつき.
The same dictionary pronounces most top baby names its own way
Meiji Yasuda's lists draw on its own policy records as at September 2025, covering 12,505 children. We read the 2025 edition on 1 November 2026 and set each name in the sentence この子の名前は〇〇です。. The sort below is ours.
| Result | Names, of 21 | Example (printed, then listed) |
|---|---|---|
| List's top reading printed | 4 | 湊 みなと |
| Listed reading in the dictionary, another chosen | 3 | 蓮 はちす, レン |
| Spelling in the dictionary, listed reading absent | 3 | 詩 し, ウタ |
| Spelling absent from the dictionary | 11 | 陽翔 ひしょう, ハルト |
This is a stress test on a few of the newest names, and no accuracy rate. The list records one reading per child, so a reading off it may still be genuine: 翠 みどり has its own given-name entry in the dictionary.
Age explains over half of the table, since a 2007 word list cannot hold a 2025 coinage. It does not explain the rest. Ogihara got to the deeper finding first: his 2021 paper counted 18 readings for 大翔 over 15 years of the same company's lists, so a newer dictionary would still have to pick one.
Japan's family register sounds out each name, with a test for new entries
Since 26 May 2025 Japan's family register has carried the reading of a name for nationals, under Act 48 of 2023. People already on it got theirs by notification or, a year later, by default. The Ministry of Justice says the notification window ended on 25 May 2026; our guide to name registries covers that procedure.
A new entry faces Article 13(2) of the Family Register Act: the reading must be one 「氏名として用いられる文字の読み方として一般に認められている」, generally accepted as a reading of characters used in names, in our translation. A separate transitional route let people already registered keep a different reading they could show was in use. Where the municipality cannot confirm a reading, Rule 30-3(2) has it ask for a written explanation citing dictionaries, newspapers, magazines or books.
The ministry's FAQ gives examples: 太郎 read as ジョージ, 高 as ヒクシ. Its own verb is hedged, 「認められないものと考えられます」. Neither the Act nor the FAQ uses the term kira-kira, and we found no published list of accepted readings. We say nothing about whether any name would pass. The register is also no public lookup, so software cannot consult it.
Official tables supply a child's characters and leave the reading unsaid
The Jōyō table's 2,136 characters are a 「目安」, a guide for general use. Its preface says it does not cover proper nouns, prefecture names and the like aside, and it gives no name readings. Yet Article 50 requires 「常用平易な文字」 in a child's name, and Rule 60 lists the table's characters first among those. So the law takes the table's characters and leaves the readings to Article 13. Old and new character forms and a name's passport spelling are separate matters.
Software reads 辛い one way and needs a list to voice the doubt
辛い is からい, spicy, or つらい, hard to bear. Type このカレーは辛いです。, "this curry is spicy", and the tokenizer returns つらい, because it chooses by word costs and curry means nothing to it. So our furigana generator keeps a list of doubtful forms, 277 at commit 9e9e9ba96, where every page figure here was run. It still prints つらい, then underlines the word and offers からい. IPADIC has 10,977 written forms with two or more readings, so most go unflagged.
Counters present a different problem: sound changes that the tokenizer does not apply. The Jōyō table notes it in the rows for 羽 and 把: 一羽（わ），三羽（ば），六羽（ぱ）. All 40 pairings of the digits 1 to 10 with 本, 匹, 杯, 分 now match the regular pattern, which is ours, spot-checked, and no official table.
Until 1 November 2026 the page did less than it said. Its list held 22 forms; the counter's ruby never changed, so all 21 pairings that take a sound change came out wrong; 令和, the current era name, printed りょうわ; and the 15 names above showed no doubt at all. Now 令和 prints れいわ by a hand entry, nine of those names are underlined as built from parts, and six still pass silently, among them 蓮 はちす. A missing underline means unchecked: 二行目 still comes out as にこうめ.
Browsers read furigana from an element named after a British type size
The W3C's 2001 Recommendation says the name "originated from the name of the 5.5 pt font size in British printing". Today's markup comes from the HTML Standard: <ruby lang="ja">漢字<rp>（</rp><rt>かんじ</rt><rp>）</rp></ruby>. The rt element holds the reading, and rp elements "are used for legacy user agents that do not support ruby at all".
In the W3C's note on Japanese layout, 総ルビ puts ruby on every ideographic character and パラルビ only on hard ones. Meiji Yasuda's next list will replace these 21 names, and we did not read the register's schedules of permitted katakana and symbols, which are images in the official text.
View author profile → · Editorial policy
