Hyōki yure: keeping spellings consistent across a long Japanese document

Open a long Japanese document written by more than one person, or by one person over several weeks, and you will almost always find the same word written two ways. できる on page 3, 出来る on page 12. サーバー in the overview, サーバ in the setup steps. 申し込み in the body text, 申込み on the form. Japanese calls this 表記ゆれ (hyōki yure), literally "wobble in the written form".

Each spelling on its own is usually correct. That is what makes hyōki yure different from a typo: no dictionary will mark 出来る as wrong. The problem is the mix. Readers notice it and take it for carelessness. A translator wonders whether two spellings mean two different things. And a search for サーバー skips every page that says サーバ, which in a manual, a help site or a contract is reason enough on its own to fix it.

What follows is a routine in six steps, from choosing a reference to keeping the style sheet alive after the document ships.

Where the wobble comes from

Japanese gives writers more legitimate choices per word than most languages do, and each kind of choice produces its own kind of wobble.

• Kanji or kana: 出来る / できる, 下さい / ください, 事 / こと, 様々 / さまざま, 全て / すべて.
• Okurigana: the 1973 cabinet notice on okurigana (送り仮名の付け方) sets a main rule and allows variants, so 行う and 行なう, or 申し込み and 申込み, can both be defended. The piece on okurigana rules has the detail.
• Loanword spelling: サーバ / サーバー, インターフェース / インタフェース, ウィンドウ / ウインドウ, バイオリン / ヴァイオリン.
• Character width: ３ and 3, ＡＢＣ and ABC, and half-width katakana pasted in from older systems.
• The IME itself: it learns from what you type and offers last time's choice first, so whichever form you picked once tends to spread through everything you write afterwards.

Add several writers, material pasted from older documents and a few rounds of edits, and a fifty-page document can easily carry a dozen wobbling words.

Step 1: Choose a reference before you choose spellings

Decide what you are following before you decide any single word. The usual order of authority looks like this:

1. The client's or publisher's own style guide, if there is one. It wins over everything below it.
2. Earlier documents in the same series, or an existing glossary. Readers have already seen those spellings, and consistency with them counts.
3. The public standards: the Jōyō Kanji table (常用漢字表) for which kanji are in general use, the okurigana notice, and the 1991 cabinet notice on loanword spelling (外来語の表記). For official writing, the 2022 guidance (公用文作成の考え方) builds on all three.
4. A published press or house style handbook, if the text is journalistic or written for a general audience.

Without a reference, every choice turns into a debate. With one, most choices are already made and you only argue about the gaps.

Step 2: Harvest the variants

Before changing anything, collect. Read a representative stretch, such as the first ten pages and one section from the middle, and write down every word you see in two forms. Then search the whole document for the usual suspects:

• kana and kanji pairs: できる / 出来る, ください / 下さい, いただく / 頂く, こと / 事, とき / 時, もの / 物, ほか / 他, など / 等
• okurigana pairs: 行う / 行なう, 申し込み / 申込み / 申込, 取り扱い / 取扱い, 問い合わせ / 問合せ
• loanwords that end in a long vowel: サーバ(ー), ユーザ(ー), コンピュータ(ー), フォルダ(ー), プリンタ(ー)
• loanwords with variant small kana: ウィ / ウイ, ヴァ / バ
• compound loanwords with and without the middle dot: ユーザーインターフェース / ユーザー・インターフェース
• digits and Latin letters in two widths

Note how many hits each form gets. When nobody has decided yet, the majority form is often the cheapest to keep: fewer edits, and fewer chances to break something. The checker on the front page also flags some common pairs, such as サーバ / サーバー and 下さい / ください, so pasting in one section at a time makes a quick first pass.

Step 3: Decide, and write down why

Now make one decision per item and record it in a style sheet, often called a 表記統一表 (hyōki tōitsu hyō) or a list of 用字用語 (yōji yōgo, "characters and terms"). The "why" column matters more than it looks. Later on, or when someone else takes over the checking, a bare "use できる" invites a well-meaning colleague to undo it. "Use できる for the potential verb; 出来 stays in compounds such as 出来事" settles it.

A sample style sheet

Here is the kind of sheet a general business or technical document might use. It records one set of house choices, not the only correct set.

Fifteen rows is plenty for most documents. A sheet that grows to two hundred rows stops being read, so keep only the words that actually wobble in this text.

Step 4: Search, do not replace all

Replace all is how hyōki yure turns into real errors. Japanese words have no spaces around them, so a search string matches inside other words:

• Replacing 事 with こと turns 事業 into こと業 and 食事 into 食こと.
• Replacing 時 with とき turns 時間 into とき間 and 同時に into 同ときに.
• Replacing 下さい with ください everywhere also changes the main verb, which your sheet may want to keep in kanji.
• Replacing サーバ with サーバー turns every correct サーバー into サーバーー.

The last one is the classic, and it is the reason to search with a pattern, not a plain string. Step through the matches one at a time, or use a pattern narrow enough that every hit is a real target.

Search patterns that help

Most text editors, and some word processors, accept regular expressions in the search box. The patterns below are written for that kind of search. Tools differ in the details, so check your editor's help and try each pattern on a copy first.

サーバ(?!ー)        サーバ not followed by ー (needs lookahead support)
ユーザ(?!ー)        the same for ユーザ
出来[るたなまれ]     出来る and its forms, but not 出来事
行な[わいうえおっ]   行なう and its forms
して下さい          the auxiliary only, not the main verb
して頂[きくけい]     the auxiliary いただく written in kanji
[０-９]             full-width digits
[Ａ-Ｚａ-ｚ]        full-width Latin letters

If your tool has no lookahead, search for サーバ and step through the hits by hand. It is slower but just as safe. For width problems, including half-width katakana, the full-width and half-width cleanup guide has more patterns and a sensible order to apply them in.

Step 5: Check the places search forgets

Body text gets searched. Plenty else does not. After the main pass, go through:

• Headings and the table of contents, especially a generated contents page that has to be refreshed after the edits.
• Tables, captions, footnotes, comments and text boxes, which some tools skip unless you tell them not to.
• Text inside images and diagrams, which search cannot see at all.
• Alt text, page titles and metadata on web pages.
• Interface strings in a manual. If the software says ユーザ, the manual should quote the screen exactly, even when the sheet prefers ユーザー. Note that exception on the sheet so nobody "fixes" it later.
• Headers, footers, the cover page and the file names themselves.

Step 6: Keep the sheet alive

A style sheet only works if everyone writing or checking the document uses it. Keep it next to the document, in the same folder or as the first tab of a shared file. Give one person the job of adding to it, and add items as you find them, not at the end. Send it to translators and to anyone writing new sections.

If the writers type with an IME, they can register the preferred forms in its user dictionary, which stops some of the wobble at the source. When a new decision replaces an old one, mark the change on the sheet and search the whole document for the old form, not just the section in front of you.

When to leave a variation alone

Not every pair of spellings is a wobble. Before you unify, check for these:

• Different meanings. 始め (the start of an action) and 初め (first in time), 早い (early) and 速い (fast), 聞く (hear, ask) and 聴く (listen with attention), 表す (express) and 現す (reveal, appear) are separate words that share a reading. Merging them introduces errors. A style sheet can list them as "both, by meaning".
• Kanji kept for a concrete sense. Many house styles, including the conventions of official writing, use とき in kana for "in the case that" but keep 時 for a point in time, and use もの for an abstract "thing" but 物 for a physical object. That is one rule applied consistently, not an inconsistency.
• Proper nouns. Company names, product names and titles of works keep their own spelling, ヴ and all.
• Quotations and screen text. Quote what the source actually says.
• Deliberate effect. Fiction, poetry and advertising sometimes choose kana or kanji for tone. If the choice is clearly intentional, ask before you touch it.

The checker can point at likely pairs, but the decisions belong to you and the sheet. A good sheet is short, gives a reason on every line, and travels with the document to whoever checks it next.