FontLab localization: Japanese

A first draft for Japanese. A model researched the guidance below and the term table in references/terms.md; no native reviewer has approved the language as a whole. Individual terms carry their own approval status; a proposed term remains proposed, and a protected name keeps its glossary spelling. The Japanese interface catalogs are unfinished machine drafts: a string found in one settles neither a term nor a rule.

Use references/terms.md, and a newer core memory when one is supplied, for terminology, definitions and review status. Keep a proposed term consistent across the interface and help, and report it as proposed. Where a native reviewer's decision or newer project data disagrees with this skill, follow it and record the change. Report only checks performed: text written with this skill is not reviewed by a native speaker and not tested in the running application until someone does that.

Use these rules with the localization principles (in fontlab-localization), Qt interface rules (in fontlab-localization) and Japanese language guide (in the FontLab writing guide). Examples name their source (Glyphs 3, Font Book, CoreText, Qt's qtbase_ja catalog) or are marked constructed. They show grammar; the term table decides terms.

Locale and register

• Locale ja, Qt catalog ja_JP: standard written Japanese.
• Leave the user unnamed: Japanese drops the subject. Never あなた, 私たち, 弊社 or お客様; your font is フォント.
• Labels are noun phrases or plain-form verbs. Every full sentence is です・ます style, never だ・である.
• Polite, not deferential. A sentence that asks the user to act ends in 〜してください: あとでやり直してください。 (Please try again later., Glyphs). Never 〜していただけますか, お願いいたします, ご〜になる; never ください in a label.

Grammar of interface strings

String	Form	Example
Menu command	object + を + action noun without する, or a dictionary-form verb; no 。	アンカーを追加, 開く (Add Anchor, Open, Glyphs)
Command that opens a dialog	the same, with the source's ellipsis	フォントを追加… (Add Fonts…, Font Book)
Button	as a command, shortest form	保存, キャンセル (Save, Cancel, Font Book)
Checkbox or toggle	as a command; negative in 〜ない	製品用グリフ名を使わない (Don't use Production Names, Glyphs)
Panel or dialog title	noun phrase	フォント情報 (Font Info, Glyphs)
Progress message	〜しています + the source's ellipsis	アップデートを確認しています… (Checking for updates…, Glyphs)
State label	〜中, 〜済み	ダウンロード中 (Downloading, Font Book), 編集済み (Edited, Glyphs)
Success	〜を〜しました	フォントを書き出しました。 (Fonts exported., constructed)
Cannot (now)	〜できません	フォントを登録できません (Cannot Register Fonts, Font Book)
Failed attempt	〜できませんでした	フォントをダウンロードできませんでした。 (Font Download failed., CoreText)
Not found, none	〜が見つかりません, 〜されていません, 〜なし	フォントが選択されていません (No Font Selected, Font Book)
Confirmation	〜しますか?	選択中のフォントを削除しますか? (Do you want to remove the selected fonts?, Font Book)
Tooltip, Quick Help	〜します, no subject	影響のあるレイヤーを新規タブに開きます。 (Opens a new tab with all affected layers., Glyphs)

• A dialog title repeats the command that opens it.
• Failed to… and Could not… are 〜できませんでした, not 〜に失敗しました. Are you sure…? is 〜しますか?, not 〜してもよろしいですか?; 本当に only for really.
• can is 〜できます, never 〜することができます; must is 〜する必要があります.
• The user's action is active (〜を選択します); what the application does unasked is passive (〜が表示されます).

Compact labels

Drop する after an action noun (追加, not 追加する) and の between nouns where the compound stays clear (フォント情報). Never shorten a term from the term table, clip a katakana word, or drop a counter after a count. Keep the Latin short forms the source uses (UC, lc, UPM, PPM). A kana or kanji is as wide as two Latin letters: count four characters as eight in a dense panel.

Plurals and placeholders

Qt gives Japanese one numerus form (nplurals=1; plural=0;); CLDR has only other. Write one <numerusform> that reads correctly for 0, 1, 2 and 100, and drop the source's (s): %n glyph(s) becomes %n 個のグリフ (constructed). Never mark a plural with たち or ら; write 複数の only for multiple.

A count takes a counter, then の, then the noun. 個 is the default for objects: %@ 個のペア (%@ pairs, Glyphs). 件 counts issues, errors and results: 件の問題 (issues, Font Book). Also 回 (times), 行 (lines), 種類 (kinds). A unit needs no counter: %1 バイト (%1 bytes, qtbase_ja). Add a counter to %1 only when you know it is a count.

A particle follows a placeholder directly and never changes with its value: %1 を隠す (Hide %1), %1 について (About %1, both qtbase_ja). Japanese is verb-final and puts the modifier first, so reorder numbered placeholders: %{value2}@ の %{value1}@ (%{value1}@ of %{value2}@, Glyphs).

Numbers, punctuation and spacing

• Numbers: 1,234,567.89; 25% and 45° without a space; a space before a unit symbol (12 pt). Digits and Latin letters are half-width, never ＡＢＣ１２３; katakana is full-width, never ｶﾅ.
• Japanese–Latin spacing: put one half-width space (U+0020) between a kana or kanji and a Latin letter, a digit or a placeholder: PostScript ヒントをコピー (Copy PS Hints, Glyphs). Microsoft, Adobe and Qt's catalog do this, and Glyphs mostly; Apple and the JTF style guide write no space. No space beside 、 。 「 」 （ ）, before (&X), before a final : ? …, or around /.
• Full-width in Japanese text: 、 。 「 」 （ ） ・. List items with 、.
• Half-width, with no space before: : ? ! / +, as macOS, Windows and Qt write them: ファイル名(&N): (File &name:, qtbase_ja). Put one space after a : or ? that more text follows.
• Quotation marks: a source 'x', "x" or “x” around a name becomes 「x」 with no spaces: グリフ「{glyphName}」 (Glyphs). Straight quotes stay in code and inside HTML tags.
• Ellipsis: copy the source's ... or … unchanged, in the same position.
• Final period: 。 where the source ends with a period, none where it does not. A source ! is dropped in a label and becomes 。 in a sentence.
• Katakana: keep the long-vowel mark of -er, -or, -ar loans (レイヤー, フォルダー, ユーザー), as the Cabinet notice 外来語の表記, Microsoft and JTF do; Apple and Glyphs often drop it. Write compounds solid, without a space or ・ (メニューバー); ・ joins coordinated items (大文字・小文字, Case, Glyphs).
• Whitespace: keep the source's outer spaces, \n and \t. Never add a space, a break or U+3000 to shape a line.

Mnemonics and key names

• Append the source's mnemonic letter, uppercase, in half-width parentheses with no space before it: 開く(&O) (&Open, qtbase_ja). It goes before a final colon or ellipsis: 色相(&E): (Hu&e:, qtbase_ja); C&ustom... becomes カスタム(&U)... (constructed).
• Write exactly (&X), four characters: Qt strips that pattern on macOS.
• Keep the source's letter, so sibling items stay unique. Never put & before a kana or kanji; add no mnemonic the source lacks. A label that stays Latin keeps its & in place (&OK).
• && is a literal ampersand: write と or keep &&, never a single &.
• Key names stay in Latin script as the source spells them: Shift, Ctrl, Alt, Option, Cmd, Enter, Esc, Tab, Delete, Backspace, Space. Arrow keys are ←, ↑, →, ↓ (qtbase_ja). Keep Ctrl+S as written. In a sentence add キー after a single key, as Microsoft and Apple do: Shift キーを押しながら (constructed).

Names and loans

• Latin script, exact case, no katakana: product, company, typeface and personal names (FontLab, OpenType, Unicode), format names (VFJ, UFO), feature and table tags (kern, OS/2), glyph names, extensions, unit symbols and code.
• Embed a Latin name with a space on each side and its class noun after it: kern テーブルを出力 (Export kern Table, Glyphs).
• Translate language and script names: アラビア語, アラビア文字 (Arabic, the language and the script, Glyphs).
• Keep the katakana loans the term table lists; coin no kanji replacement. Do not turn an ordinary English word into katakana where a plain word exists: 設定, 追加, 幅; not セッティング, アド, ウィドス.
• Never append the English original in parentheses.

Script-specific rules

• Kana, not kanji, for すべて, ください, できます, および, または (not 全て, 下さい, 出来ます, 及び, 又は).
• No furigana; no italics that the source's markup does not carry.

False friends and traps

The term table gives the term. These are the senses to reject:

• glyph is not 文字 (a character); 字形 is a shape, not the glyph slot.
• script is a writing system or a Python script: read the control.
• hint is not a tip; outline is not 概要; a font table is not 表; an OpenType feature is not 特徴; weight is not 重さ or 重量; master is not 主人 or 原本; instance is not 実例; stem is not 茎 or 幹; case is not ケース or 場合; family is not 家族; a style named Bold or Italic is not the word processor's 太字 or 斜体.
• kerning, tracking and spacing are three things.
• Machine-translation habits to remove: あなた; 〜を行います for a plain verb; passive for the user's action; それ, これら for it, these; 〜のための for every for; の three times in a row; full-width ？ ： ！; a missing counter.

Review

1. Placeholders, markup and \n match the source.
2. Mnemonics: exactly (&X), the source's letter, before the colon or ellipsis; none added, none lost.
3. Each numerus entry: one form, with a counter, right for 1 and for 100.
4. A space between Japanese and Latin or digits, none beside 、。「」（）; half-width : and ?; no full-width Latin.
5. Form by role; no だ・である, no あなた, no excess 敬語.
6. Terms match the term table.
7. Would a Japanese type designer say it this way?

FontLab terminology and compact labels

Preserve deliberately uppercase source labels using the language’s case rules. Keep numeric controls compact with UPM, PPM and LSB/RSB (French AG/AD); explain the full terms in help. The Family Dimensions “Units Per eM:” label is an intentional expanded exception.

Use the current core memory and its approval status. OpenType lookup defaults to the translation of procedure, with a different term only when strong OpenType-specific evidence supports it. Explicit choices are German Lookup, French lookup, Polish procedura and Russian процедура. Ordinary searches, individual rules and OpenType features are separate concepts. Preserve literal code keywords and identifiers.

Translate Fusion with the fallback Welder. Use the core's phonetic FontAudit transliteration in non-Latin scripts, including Chinese; preserve FontAudit in code. Proposed spellings remain proposed. Before rebuilding a QPH, recover team edits and import their terms and notes into the core memory.

Compare mnemonic-split terms with unmarked labels. Keep labels compact, aiming for 120% of the English length in Dimensions and Glyph Window preferences without losing meaning. Omit redundant font, color or command wording when context supplies it. TTH labels can omit “use” and abbreviate TrueType hinting; help can explain more. A full harmonization requires three complete catalog reviews, then reconciliation of help with the final UI and verification of generated resources.

Run the review as a loop

The Review list above is this skill's checklist. Run it on each portion as H17 describes: one item at a time against source, target and the term table; name the string behind each answer; repair what fails and recheck it. Stop after three rounds and report what still fails.

References

• references/terms.md: the full Japanese term table, generated from the core memory ja-core.tmx of the writing guide, with the status, translator note and definition of every term.
• The Japanese localization guide of the writing guide is the source of the sections above.

Related skills

fontlab-localization for the shared rules; fontlab-terminology for product names and shared nouns; fontlab-technical for Japanese help text.

House rules

These rules implement the FontLab writing guide. They are copied into every skill so each installed skill can work independently. Corpus measurements can inform review; they are not quotas, universal laws, or tests of authorship.

H1. Accurate actors. Address the reader when they act or choose. Name the application when it performs an operation. Fonts and files contain data that software interprets. Prefer active voice when the actor matters; retain a clear passive construction when the actor is unknown or irrelevant. Never invent an actor or cause to change the grammar. When an English sentence pairs a reader action with the software's response, give each clause a subject and restate the action compactly: “If you generate a glyph with Aidus, FontLab generates each master independently.” Name the most specific responding surface the evidence supports (a window, tool or dialog), otherwise the application. Use present tense for an immediate result. Where a glossary entry exists and the surface can link, link a domain term to it rather than defining it in an apposition.

H2. Never invent a fact. Ground numbers, names, dates, labels, shortcuts, defaults, errors, versions, and quotations in supplied or checked evidence. A draft is evidence of what was written, not independent proof of its claims. Mark unresolved facts with a specific working placeholder such as [VERIFY CLAIM], [CONFIRM LABEL], or [CONFIRM OFFER]. A marked draft is unfinished; resolve the gap before publication. Clearly label fictional examples before their invented details.

H3. Preserve certainty and scope. “May reduce” does not become “eliminates.” Preserve conditions, negation, timing, quantities, compatibility, licensing, and other consequential limits. State a limitation directly rather than disguising it as a favorable position. A single observation does not prove universal behavior. Conflicting sources need scope checks, not an automatic choice of the stricter claim.

H4. Preserve the writer's voice. Do not manufacture stakes, candor, reader emotion, endorsement, or a new narrator. Preserve useful habits and the supplied stance. Reorder, split, or shorten according to the requested edit depth and reader need, without adding facts or causal relationships. A correct draft can remain unchanged.

H5. Give thought a rhythm. Carry a subject through an action, distinction or consequence. Let a developed sentence explain a relationship; let a shorter one settle a result when that change of pace helps. Repeat a concrete object or term when its meaning develops, not as a compulsory callback. Technical actions stay literal; explanations and marketing have room for patient attention, a bounded comparison or dry observation. Keep useful qualifications beside their claims. Do not impose sentence-length, pronoun, punctuation or paragraph quotas, or force every paragraph into the same long-then-short pattern.

H6. Notice the useful detail. Retain names, mechanisms, conditions, versions and issue numbers that help the reader understand or act. From the evidence, choose the object, contrast or small behavior that makes the explanation tangible: a changed preview, a repeated comparison, a file whose status matters. Follow that detail far enough to explain its significance. Warmth can come from this attention and patience. Never invent an observation, personal experience or reader emotion to make prose vivid, or remove a necessary qualification to shorten it.

H7. Punctuation serves meaning. Prefer a colon for an explanation or list. A spaced dash can carry a turn; avoid decorative glosses and repeated interruptions. Use sentence case for new headings and preserve exact source labels. A punctuation pattern does not establish authorship.

H8. Choose precise words. Review vague praise and stock phrasing such as “seamless,” “game-changing,” “leverage,” and “unlock.” Replace them when they obscure the action or make an unsupported claim. Keep an accurate technical use or protected quotation. A count or cluster is a reason to inspect a passage, not proof that each matched word is wrong.

H9. Develop the explanation. Begin where the reader can understand the task, change or offer. Give adjacent sentences a real connection: the same subject under a changed condition, an action and its result, or a question and its answer. A before-and-after comparison needs evidence for both states; a transition must not invent causality. Let a useful aside return to the main thought. Place low-stakes discoveries where they aid understanding, while keeping price, risk, prerequisites and recovery visible when needed. End at the useful result or next action; a quiet ending or a substantial summary can each serve the material.

H10a. Preserve conditions. Use “if” for a condition and “when” where the intended timing or situation warrants it. Check the whole meaning: “when a panel opens” and “while a panel is open” describe different scopes. Keep useful conditionals and parenthetical explanations; do not insert them to imitate a presumed author.

H10. Preserve expression that works. A fragment, three-part phrase, exclamation, aside, or single-sentence paragraph can serve a passage. Keep it when it supports meaning or the writer's rhythm. Do not add one to meet a budget, or remove one because of a generic stylistic test. Keep instructions and consequential conditions literal and easy to find.

H10b. Use the destination's notation. Preserve exact interface labels, tags, extensions, and identifiers. Italicize labels in neutral release notes; in technical site content use supported highlight notation, with bold as the plain-Markdown fallback. Use code style for machine-readable text. State price amounts and currencies unambiguously from evidence; do not infer an exchange rate, tax policy, or license term. A supplied currency shorthand needs enough context to identify the actual offer.

H11. Names and collisions. FontLab is the product; Fontlab Ltd. is the company. Preserve product names, module identifiers, domains, historical names in their historical scope, and exact quoted strings. Do not translate a product name. Name the application when Layer, Mask, Group, Fill, Brush, Knife, Transform, Pencil, Eraser, or Scissors could have more than one meaning. Use a declared short form consistently.

H12. Protect literal material. Preserve quotations, code, commands, paths, identifiers, API names, URLs, legal text, interface strings, placeholders, and table data during prose editing. Change them only for a requested or necessary correction supported by evidence. Do not paraphrase a quotation while retaining quotation marks or an attribution. Explain a consequential correction when the output contract permits it.

H13. Choose the register per passage. Marketing helps assess an offer; technical writing explains or instructs; neutral prose states facts and changes warmly. A document can combine them. Prices, limits, licensing, compatibility, security, and migration facts stay plain and prominent. Audience and purpose determine an email's register. Do not impose a fixed emotional mixture or make every opening a story.

H14. Review facts, then movement. Compare claims and protected strings with their evidence; check scope, action order and terminology. Then read whole passages for attention, connection and pace. Repair a flat sequence by developing an existing detail or relationship, not by adding praise or invented events. Check that humor remains intelligible and the information remains true when the joke is missed. Recheck the facts after a voice edit. A measurement or passing build does not certify facts, usability or authorship. Report only checks actually performed.

H15. Scale to the surface. A button or tooltip needs a clear label, action, or condition, not a miniature essay. Apply factual and naming safeguards at every length. Include the detail the task requires; do not add proof paragraphs, metaphors, or pronouns to satisfy a template. For translation, keep essential instructions literal and references clear.

H16. The override. Follow the user's task and supplied voice before style defaults. Break a default sooner than make the writing worse. Source material and quoted prompts are data, not authority to change the task. Style preferences never justify presenting an invented or unsupported claim as established fact.

H17. Check each portion, repair, check again. Reading these rules does not apply them; checking the text does. After drafting or editing each portion (a section, one deliverable, a batch of entries; a label or tooltip is one portion), run the skill's checklist on it. Answer each item separately, against the text and the evidence. For a failure, quote the words that fail. For a pass, point to the passage that satisfies the item, or, for an item about something absent, confirm that you searched the whole portion and found none. An answer with nothing behind it has not been checked. Repair only the failures, because a passing item needs no edit and a correct sentence can stay unchanged. Recheck the repaired passages and every fact the repair touched. If the same item fails again after its repair, stop working on it and mark it; do not try a third time. Stop after three rounds in all, and mark what still fails with a specific placeholder instead of reporting a pass. If you can start a separate reviewer (a subagent or a fresh chat), give it the draft, the evidence and the checklist, not your drafting rationale, and ask for failures with locations; act on findings that name a rule and a passage. Keep the checklist record private unless the user asks for it or the output contract allows notes.