The long vowel mark in katakana loanwords: サーバ or サーバー?

Put a Japanese software manual next to a daily newspaper and you will find the same word spelled two ways. The manual says サーバ, the paper says サーバー. You will see コンピュータ and コンピューター, ユーザ and ユーザー, ブラウザ and ブラウザー, sometimes on the same screen. Neither spelling is a typo. They come from two conventions that have lived side by side for decades, and the person checking the text does not need to pick a winner for the whole language. The job is smaller and more useful than that: make sure one document follows one convention.

This piece covers what the mark is, where the two habits come from, how to choose between them, and how to find the stray spellings once you have chosen.

What the mark does

ー is the 長音符 (chōonpu), also called 長音記号 or, informally, 音引き (onbiki). It lengthens the vowel before it: カー (kā, "car"), ルール (rūru, "rule"), メール (mēru, "email"). In loanwords from English it often stands for the vowel sound at the end of words spelled with -er, -or or -ar: server, monitor, elevator. That final position is where the wobble lives. Nobody argues about the mark in the middle of ルール. The argument is about the last one in サーバー.

In vertical text the mark turns to run downward, so it follows the direction of the writing. Keep that in mind, because it gives you a quick way to spot impostors, which we will come to.

Where the two spellings come from

The engineering habit: JIS Z 8301

JIS Z 8301 is the Japanese Industrial Standard that sets out how JIS standards themselves are drafted and laid out. For a long time it carried a rule for technical terms: if the word runs to three morae or more, drop the final long vowel mark; if it is two morae or fewer, keep it. That rule gives コンピュータ, プリンタ and エレベータ, while a short word such as カー keeps its mark.

Standards, engineering specifications and equipment manuals followed JIS practice, and much of the early computer industry wrote the same way. That is why the short forms still feel natural in IT documents, and why many engineers read サーバー as slightly off. If a client's style guide cites JIS Z 8301, ask which edition they follow and what the rule says there. Standards get revised, and a style sheet should quote the rule it applies rather than lean on a standard's name.

The general rule: 外来語の表記 (1991)

The cabinet notice 外来語の表記 (gairaigo no hyōki, "how to write loanwords") of 1991 is the official guidance on writing loanwords in general text. In outline, it writes long vowels with ー as a rule, and for English endings such as -er, -or and -ar it gives the long form, like コンピューター and エレベーター, as the basic spelling. It also allows the final mark to be left off where established custom does so. Both spellings therefore fit inside it, but it leans clearly toward the long one.

Newspaper and broadcast style generally follows that lean, and so do most general publishers. A reader who mostly reads the news will find サーバー normal and サーバ clipped.

The software shift in 2008

In 2008 Microsoft announced that its Japanese products and documentation would move to the long forms for words of this kind, aligning with the general rules in the cabinet notice. The change was widely reported because it touched words that millions of people saw every day on screen. Since then the long forms have become common in consumer software, while engineering documents, standards-heavy fields and many in-house manuals kept the short ones. The result is the mixed landscape we have now: an interface string from one source, help text from another, and both spellings in the same window.

Which one should your document use?

There is no single right answer, but there usually is a right answer for a given document. This table covers the common cases.

Two notes on that last row. First, a product called "Example Server" in Japanese might be written by its maker either way, and you copy the maker. Second, the name exception can leave a document with both spellings on purpose: サーバー in your running text, サーバ inside a product name. That is fine, and worth a line on the style sheet so the next reviewer does not "fix" it.

Other katakana wobbles are a separate choice

The JIS habit is about the final mark. Words also vary in the middle, and those variations belong on the same style sheet but follow their own logic. インターフェース, インタフェース and インターフェイス are all in use: one drops a long vowel mid-word, another spells the end with イ instead of ー. Small kana vary too, as in ウィンドウ and ウインドウ. A document that settles サーバー versus サーバ and then lets three spellings of interface through has fixed half the problem. When you build the word list for a project, list each word in full, not just its ending. The launch piece on keeping spellings consistent across a long document goes through building that list.

How mixed spellings get in

Most of the mixing is mechanical, not a writer changing their mind.

1. Typing. With romaji input, "sa-ba-" gives サーバー and "sa-ba" gives サーバ. Whether the writer hit the hyphen key at the end decides the spelling, and that habit differs from one writer to the next.
2. Copy and paste. A paragraph lifted from vendor documentation brings its convention with it.
3. Several writers. Each brings their own habit, and a shared document ends up with all of them.
4. Translation memory. Segments translated years apart for different clients come back in the old spelling.
5. Compounds. サーバー管理 and サーバ管理, ユーザー名 and ユーザ名: a writer who is careful with the bare word often forgets it inside a compound.

The checker on the front page flags pairs such as サーバ and サーバー when both appear in one text; the legend of issue types explains how it decides and why it sometimes flags a mix that is deliberate.

Impostor characters

Here is the part that ordinary spellcheck misses entirely. Several characters look like ー in many fonts and are not the long vowel mark at all.

• The long vowel mark itself is U+30FC, KATAKANA-HIRAGANA PROLONGED SOUND MARK. This is the one you want.
• The kanji 一 (ichi, "one"), U+4E00. In some fonts it is nearly the same width and weight.
• The full-width hyphen-minus －, U+FF0D.
• The half-width long vowel mark ｰ, U+FF70, which arrives with half-width katakana.
• Various dash characters, such as the horizontal bar at U+2015, which can look identical at body text size.

These cause real trouble. Search for サーバー and you will not find a version spelled with the kanji for one. A screen reader may read that version as "sā ichi ba ichi". Sorting and indexing break. The half-width form also pulls in a separate cleanup job, covered in the piece on full-width and half-width characters.

Two ways to find them:

• Switch the text to vertical layout if your editor allows it. The real mark turns to run down the column. The kanji 一 stays horizontal, and most dash-like characters behave differently enough to stand out.
• Search with a regular expression. [ァ-ヴ][一－] finds a full-width katakana followed by the kanji 一 or a full-width hyphen-minus. It also matches real words such as ページ一覧 (pēji ichiran, "page list") or データ一式 ("a full set of data"), so read each hit instead of replacing all.

Searching for the variants

Once you have chosen a convention, finding the strays is mostly a matter of searching word by word. If your editor supports lookahead in regular expressions, サーバ(?!ー) finds サーバ where no mark follows, so you can catch the short form without stopping on every correct サーバー. Reverse it for documents that use the short form: サーバー is enough on its own.

For a long document, a quicker route is to list every katakana word that ends in ー or could end in it, sort the list, and look for near pairs. Most editors and spreadsheet programs can do this with a little effort, and the pairs jump out once they sit next to each other.

Questions that come up

Is one of the spellings wrong?

No. Both are established, and the cabinet notice itself allows the short form where custom uses it. The error a proofreader should flag is mixing them without a reason.

Should a translator change the client's spelling?

Follow the client's glossary or style guide if there is one. If there is none and the source material is mixed, pick the convention that suits the audience, apply it throughout, and say so in a short note with the delivery.

Does the mark appear in hiragana?

In standard writing, long vowels in native words are written with vowel kana: おかあさん (okāsan), おとうさん (otōsan). In casual writing, manga and social media posts, ー shows up in hiragana for effect, as in すごーい (sugōi, "wow"). In formal text, treat ー after hiragana as something to query.

What about words from languages other than English?

The final mark question mostly concerns English -er, -or and -ar endings. Words borrowed from other languages tend to have settled spellings of their own, so check a current dictionary rather than applying the engineering rule by analogy.