wordcraft/docs/localization-parity.md

Localization parity with Microsoft Word

Last reviewed: 2026-10-10 · Last updated: 2026-10-13 · Change: trivial (Czech interface catalog completed, 2,447 entries; previously trivial: interface audit runs in CI again: size labels translated, 20 strings in every catalog, Read Aloud error race fixed, #1128; previously trivial: Review › Language reaches every language again — Set Proofing Language… and Set As Default, #860; previously minor: Italian proofing: it-IT spelling from the installed Hunspell dictionary, accent-first suggestions, conservative grammar with Italian messages; previously minor: Portuguese proofing; previously trivial: Turkish interface catalog completed, 2,325 entries, #774; previously minor: Hungarian interface added; previously trivial: sorting and table numbers follow the text's language, #836, #837; previously minor: Polish interface; Polish proofing — spelling, grammar, hyphenation — and a proofing language per run; previously minor: Turkish interface added; previously major: first version; catalogs measured from crates/ui-egui/src/i18n, Word's languages from its bundle listing) · Target: Microsoft Word (Microsoft 365) for Mac 16.113.4

Interface languages, script support and proofing per language. How translation works (catalog format, adding a language, clean-room rule): crates/ui-egui/src/i18n/mod.rs.

Dimension: ~29% (estimated; 28% before Italian proofing, 27% before Portuguese proofing, 25% before Polish and per-run proofing languages), 70–115 h to full for the twelve key languages plus ~45 h for the other ~17 languages Word ships, plus native-speaker review (human).

What Word ships (measured by listing the bundle)

• 30 interface languages (31 .lproj folders incl. Base): ar, cs, da, de, el, en, en-GB, es, es-MX, fi, fr, fr-CA, he, hu, id, it, ja, ko, nl, no, pl, pt (BR), pt-PT, ru, sk, sv, th, tr, zh-CN, zh-TW. Arabic and Hebrew get a mirrored interface. No Hindi or Vietnamese interface on Mac.
• 120 proofing tools (SharedSupport/Proofing Tools): spellers for ~50 languages, grammar checkers for ~20, hyphenators for ~16, thesauri for ~20.

What WordCraft ships (measured)

• 20 interface languages: English plus zh-hans, zh-hant, ja, uk, es, pt-br, Serbian sr/sr-latn (#250), Estonian et, Turkish tr, Russian ru, Italian it (#594), German de (#574), Polish pl, Czech cs, Hungarian hu, French fr, Bulgarian bg and Vietnamese vi (#1434). Catalogs have 2,057–2,454 entries each; Polish has every key of every other catalog, including the built-in style names and the equation gallery, except the citation style names only Hungarian and Bulgarian translate so far (a test checks it). A test (every_tab_and_command_is_translated) enforces that every ribbon tab, command label and ribbon location is translated in every language. pl, pl-PL and pl_PL.UTF-8 system locales pick Polish automatically.
• Of the 675 English strings a scan extracts from tl!(…) calls and command labels/locations, each catalog covers 661 (98%). Some newer dialog and pane strings (Equation tab tooltips, Zotero, Read Aloud, Backstage greetings) aren't wrapped in tl! yet, so they show in English in every language: the true share of visible strings is ~93–95% (estimated).
• Pending PRs: Norwegian Bokmål (#208) and Nynorsk (#210), Hebrew with RTL (#20).
• Proofing: English, Polish, Portuguese and Italian, chosen per run from its language (w:lang, resolved through the styles; Review › Language and the status bar set it: a short list, plus Set Proofing Language… for any of 29 languages and Set As Default for the document). Languages without proofing tools get the styles; Review › Language and the status bar set it). Languages without proofing tools get no spelling or grammar marks (as in Word without the language's tools) instead of being flagged as English. Detail in Proofing per language.
• Document scripts: shaping via HarfRust (HarfBuzz port) for every script; bidi (UAX #9) and Arabic shaping tested (#207); IME on all desktop platforms (winit fixes pending, #155–#164).

The twelve key languages

Status: full = every user-visible string translated and the script renders and edits correctly; partial; menus only; none. Hours are to full, Opus 5.5 agent wall-clock, excluding the shared items below.

Language	Code	UI strings translated	Dialogs / tooltips / help	Script support	Native review	Status	Hours to full
English	en	source (100%)	✅	Latin ✅	yes	full	0
Simplified Chinese (Mandarin)	zh-hans	929 entries; 98% of extracted, ~95% of visible	most; newer panes English	CJK ✅, IME ✅; Windows UI font fixed (#248); interface font chain covers every catalog character, installed CJK font after the embedded ones when it doesn't, PingFang found on macOS 12 (#486); font list by Chinese names (宋体, 微软雅黑, #505); no vertical text	no	partial	3–5
Spanish	es	~930; 98% / ~95%	most	Latin ✅	no	partial	1–2
Hindi	hi	0	❌	Devanagari shaping via HarfRust, untested; no hyphenation	no	none	4–6
Arabic	ar	0	❌	shaping, joining, lam-alef, bidi ✅ (#207); no kashida; no mirrored UI	no	none	6–8
French	fr	2,033 entries	most; newer dialogs partly English	Latin ✅	no	partial	1–2
Portuguese	pt-br (pt-PT none)	~930; 98% / ~95% (pt-BR)	most	Latin ✅	no	partial	2–4 (pt-BR finish + pt-PT)
Indonesian	id	0	❌	Latin ✅	no	none	3–4
Japanese	ja	~930; 98% / ~95%	most	CJK ✅, IME ✅; no vertical text, ruby or document grid	no	partial	3–5
German	de	2,021 entries; same key set as zh-hans (#574); 98% of extracted, ~95% of visible	most	Latin ✅	no	partial	1–2
Korean	ko	0	❌	Hangul ✅; macOS Korean IME fixes pending (#163, #164)	no	none	3–5
Vietnamese	vi	2,448 entries; every key of the reference catalogs (#1434); 98% of extracted, ~95% of visible	most	Latin ✅	no	partial	0–1 (review)

Other shipped languages: 14 — Traditional Chinese (zh-hant), Ukrainian (uk), Serbian Cyrillic and Latin (sr, sr-latn, #250), Estonian (et), Turkish (tr), Russian (ru), Italian (it, #594), Polish (pl), Czech (cs), Hungarian (hu), French (fr), Bulgarian (bg) and Vietnamese (vi, #1434), all partial. Estonian uses the existing fonts, follows et-EE system locales, and is available in Options or through ui.language with {"value":"et"}. Its 1,307 entries cover the existing interface catalog. Estonian spelling, grammar, dates and document templates are not added by the interface catalog. Turkish works the same way (tr-TR and tr-CY system locales, {"value":"tr"}) with all 2,325 entries, the same keys as the Russian catalog. Command search folds the Turkish dotted and dotless I (italik finds İtalik, YAZI TİPİ finds Yazı Tipi) while English labels and ASCII command ids keep their usual matching. Turkish proofing is not included; terminology and checks: localization-tr.md. Bulgarian (bg-BG system locales, {"value":"bg"}) has 2,419 entries: every key of the reference catalog, the equation gallery's labels and everything the interface audit finds (the audit checks Bulgarian as it does Russian). Counts use count-neutral wording (Думи: 21). Bulgarian text is hyphenated by the Bulgarian syllable rules (#826); Bulgarian spelling and grammar are not included. Hungarian follows hu-HU system locales and {"value":"hu"}, with a complete catalog of 2,419 entries: every key of the Russian catalog, built-in style names shown translated, plus the citation style names. Hungarian proofing is not included; terminology and conventions: localization-hu.md.

Czech follows cs-CZ system locales and {"value":"cs"}, with 2,447 entries: every key of the Russian and Spanish catalogs, including built-in style names and numbered families, the equation gallery, watermarks, paper sizes and preview tiles. The Czech coverage test enforces both reference catalogs and retains the Chinese catalog coverage; counts keep the existing count-neutral wording (Počet slov: 21). Czech proofing is not included.

Polish is the only one besides English with proofing tools, and a Polish interface writes new documents in Polish (pl-PL).

Interface audit (runs in CI). untranslated_interface_text lays the app out in a pseudo-language (every tab, File page, dialog, pane, the right-click menu, and a hover and a click on every ribbon control) and fails on visible text that skipped the catalog; then Russian, Hungarian, Turkish and Bulgarian must have every string it looked up. Proper names (citation styles such as APA or GOST, LaTeX, Unicode) and bad parameters: errors, which name a command's JSON parameters for scripts and MCP, stay English. Size labels (Envelopes, Labels, Layout › Size) are translated patterns ({name} {height} × {width} in) with the region's decimal separator; paper names other than envelopes (Letter, Legal, A4, B5 (JIS)) stay as they are (#1128).

Proofing per language

Language	Spelling	Suggestions	Grammar	Hyphenation
English (en, any region; text with no language)	Moby word list (public domain, ~160k words) with inflection rules	edit distance	repeated word, a/an, sentence capital, space before punctuation, double space	Moby dictionary, our Liang patterns, heuristic
Polish (pl)	The Hunspell pl_PL dictionary installed on the system (OS package, Homebrew, LibreOffice, the folder set in File › Options › Proofing › Dictionaries folder, or WORDCRAFT_DICTIONARIES; not bundled, so no Polish spelling marks without it; the same pane shows which file each language found), read at runtime by a Hunspell-style affix engine (crates/proof/src/affix.rs); names capitalized, compounds part by part, Kennedy'ego	missing diacritics first (zolw → żółw), the dictionary's confusion table (żeka → rzeka, chuśtawka → huśtawka), missing space (napewno → na pewno), one typo, then a stem scan within two typos	Polish messages: repeated word, space before punctuation, double space, sentence capital (knows np., r., initials, ordinals, dialogue dashes), missing comma before że/który/ponieważ/gdyż/aby/żeby/jeśli/jeżeli/gdyby/lecz (not after mimo, chyba, tak, prepositions…), we/ze (we wtorek, ze szkoły), doubled punctuation	TeX Polish patterns (Kołodziejska, Jackowski, Ryćko; MIT option) with their 20 exception words; a Polish syllable heuristic for letters outside the alphabet
Portuguese (pt, any region)	The Hunspell pt_BR dictionary installed on the system — VERO and its derivatives (same places as Polish; not bundled, LGPL/MPL), read at runtime by the same affix engine, which reads its FLAG UTF-8, 106 flags, continuation classes and FORBIDDENWORD spellings; names capitalized, compounds part by part (guarda-roupa, ex-marido)	missing accents first (coraçao → coração, saude → saúde), the dictionary's confusion table (seje → seja, excessão → exceção), missing space (nãoé → não é), one typo, then an accent-folding stem scan	Portuguese messages: repeated word (the emphatic doubles are allowed), space before punctuation, double space, sentence capital (knows etc., pág., Dr., initials, ordinals, dialogue dashes), doubled punctuation; no comma rule — que opens a clause only sometimes	TeX Portuguese patterns (Rezende, Almeida, Araujo, Benevides; BSD 3-clause) with their 2 exception words
Italian (it, any region)	The Hunspell it_IT dictionary installed on the system (same places as Polish and Portuguese; not bundled — the upstream file is GPL-licensed), read at runtime by the same affix engine; elided forms check the word after the apostrophe (l'amico, un'idea, c'è), hyphenated compounds part by part	missing accents first (caffe → caffè, perche → perché), the dictionary's confusion table, missing space (perme → per me), one typo, then an accent-folding stem scan	Italian messages: repeated word (emphatic doubles like piano piano are allowed), space before punctuation, double space, sentence capital (knows ecc., dott., prof., initials, ordinals), doubled punctuation; no comma rule	TeX Italian patterns (Beccari; MIT option), no exception words
Turkish (tr, any region)	The Hunspell tr_TR dictionary installed on the system (same places as Polish; not bundled), read at runtime; proper nouns capitalized, apostrophe suffixes (Ankara'da, WordCraft'ta) and hyphenated compounds	missing diacritics / ASCII restoration first (turkce → Türkçe, isik → ışık), common confusion table (yalnış → yanlış, herşey → her şey), detached suffixes (bende → ben de, geldinmi → geldi mi), one typo	Turkish messages: repeated word (idiomatic reduplications allowed: yavaş yavaş), space before punctuation, double space, sentence capital (knows vb., vs., prof., dr.), no comma before/after conjunctions (ve, veya, de), doubled punctuation	Turkish syllable rules (TDK rules); no external pattern file or word list needed
Any other	— (no marks)	—	—	the English patterns, as before

Measured with the SJP.PL dictionary: every one of the ~4.5 million forms an independent expansion of its affix file produces is accepted, and 200,000 one-edit mutations that aren't words are all rejected; a check takes ~1 µs, a suggestion list 1–35 ms (release build). Not yet: other languages' dictionaries, a thesaurus for Polish, Word's style and clarity refinements, Polish-specific AutoCorrect.

Italian (it, #594) follows it, it-IT and it-CH system locales. Its catalog has every key of the others plus 51 strings that are wrapped in tl! but were in no catalog yet (File › Info properties, the File › Home greeting, list levels, Equation tab gallery and tooltips). Ribbon tabs, groups and command names use the names Italian Word users know (feature names, as AGENTS.md allows); all other text is written from the meaning of the English. No Italian proofing yet.

Sorting and numbers in the text's language

Sorts follow the alphabet of the text's language (w:lang, else the document default): Table › Sort, Home › Sort, the index and its letter headings, the bibliography and the Table of Authorities. Letters with diacritics sort with their base letter (é with e) unless the language makes them letters of their own: Polish ą ć ę ł ń ó ś ź ż, Czech/Slovak č ch ř š ž (and Slovak ä ô), Hungarian cs dz dzs gy ly ny sz ty zs ö ü, Swedish/Finnish å ä ö and Danish/Norwegian æ ø å after z, Spanish ñ, Turkish, Croatian/Bosnian/Serbian Latin, Romanian, Estonian, Latvian, Lithuanian, Slovenian and Icelandic (crates/doc/src/collate.rs, a hand-written per-language alphabet table; full CLDR collation is not used). Table sort and table formulas read numbers with the language's decimal and group separators (1 250,50 in Polish, 1.250,50 in German, 1,250.50 in English) and write formula results with its decimal separator (crates/doc/src/locale_number.rs). Interface lists (styles, fonts) still sort by code point.

Region conventions (#838)

As in Word, a new blank document takes its paper from the system region, not the interface language: US Letter in the territories CLDR lists for it (US, Canada, Mexico, the Philippines, Chile, Colombia, Venezuela and a few Central American ones), A4 everywhere else. Lengths are shown in inches in the US (and Liberia, Myanmar) and centimetres elsewhere, written with the region's decimal separator (2,54 cm in Poland, Germany, France…); fields take either separator and any unit (1 in in a centimetre field). An explicit File › Options › Advanced › Measurement units choice still wins; its Automatic entry follows the region again.

Where the region comes from:

• File › Options › Advanced › Region (saved; agents: file.options {"region": "pl-PL"}, auto returns to the system's);
• else WORDCRAFT_REGION or LC_ALL;
• else, on Unix, each convention from its own locale: numbers from LC_NUMERIC (else LANG), paper from LC_PAPER, measurement from LC_MEASUREMENT;
• else the platform's preferred locale: the macOS preferred locale (en-PL), the browser's language on the web, and on Windows the home location (Settings › Time & language › Region) with the display language. Word follows Windows' regional format instead, so an en-US format set by someone living in Germany gives A4 and a decimal comma here; reading the regional format needs Win32 calls that are unsafe in Rust, which WordCraft forbids.

Runs without preferences (WORDCRAFT_NO_PREFS) stay en-US unless WORDCRAFT_REGION is set, so agents and screenshots give the same result on every machine. The decimal separator belongs to the language's own number format (CLDR): a decimal-comma language uses a comma everywhere except where CLDR says otherwise (de-CH), and English, whose en-XX locales follow the territory, uses a comma in decimal-comma territories (en-DE, en-PL). Other decimal-point languages keep the point in any territory (ja-PL is 2.54 cm), as CLDR's fallback to the language gives.

Templates (Letter, Resume, Report) keep their own size. Not yet: Word's "Default paper size" preference of its own, and Word's locale-specific Normal margins (2,5 cm in some regions; WordCraft keeps 2,54 cm).

Shared work (not in the per-language hours)

Item	Hours	Notes
Wrap the remaining English literals in tl! and translate them in every catalog	4–6	Makes the six partial languages full except script items
Mirrored (right-to-left) interface for Arabic and Hebrew	15–25	Ribbon, panes, dialogs, rulers
Vertical text, ruby, document grid (Chinese, Japanese, Korean)	counted in layout-parity.md
Proofing dictionaries, grammar and hyphenation for the 11 non-English key languages	20–40	Open dictionaries only (licence check per dictionary); issues #25, #40, #100. Per-run languages, the Hunspell-style affix engine and the TeX pattern reader landed with Polish, so a language with an open Hunspell dictionary and hyphenation patterns is mostly data plus grammar rules
Native-speaker review of every catalog	human	None reviewed yet

Revision history

Date	Change	Summary
2026-10-12	minor	File › Options › Proofing › Dictionaries folder (#1063): a saved folder of Hunspell dictionaries searched first, so an app started from Finder or the Dock finds them without WORDCRAFT_DICTIONARIES; the pane lists the dictionary file each language found — Polish, Portuguese, Italian, Croatian, Turkish (file.options dictionariesFolder)
2026-10-12	minor	Czech interface catalog completed: 386 entries added (2,447 total), including built-in style names, numbered families, equation gallery, watermarks, paper sizes and preview tiles; coverage test requires every Russian and Spanish key and retains Chinese coverage
2026-10-12	minor	Italian proofing: it-IT spelling from the installed Hunspell dictionary (system it_IT, never bundled — upstream is GPL), accent-first suggestions, conservative grammar with Italian messages, TeX Italian hyphenation patterns (MIT option); Review › Language already offered Italian; an it interface writes it-IT documents
2026-10-13	minor	Interface audit back in CI (#1128): Envelopes, Labels and Layout › Size labels go through translated patterns with the region's decimal separator (the old fixed-format paper rows, dead since #838, removed); 20 strings added to all 18 catalogs (status-bar messages, Reading Mode toolbar, Go To, Search's Suggested); citation style names and bad parameters: errors exempt; the Read Aloud player no longer draws a speech error before the frame has polled it, the audit's rare failure under load
2026-10-11	minor	Turkish interface catalog completed (2,325 entries, the same keys as Russian): equation gallery, built-in style names, watermarks, paper sizes and preview tiles; the interface audit checks Turkish too (#774)
2026-10-12	trivial	Region conventions: File › Options › Advanced › Region (saved) and an Automatic units entry; Unix LC_PAPER/LC_MEASUREMENT/LC_NUMERIC each decide their own convention; the web reads the browser's locale; WORDCRAFT_NO_PREFS runs stay en-US; where the decimal separator comes from (#838 review)
2026-10-12	minor	New blank documents follow the system region: A4 outside the Letter territories, centimetres outside the US, the region's decimal separator in length fields; explicit Measurement units choice wins (#838)
2026-10-11	minor	Czech (cs) interface catalog added
2026-10-13	trivial	Review › Language and the status bar menu open Set Proofing Language… (29 languages, Set As Default) and Language Preferences…; French or German text shows by name (#860)
2026-10-12	minor	Portuguese proofing: pt-BR spelling from the installed VERO dictionary (the affix engine grew FLAG UTF-8, 128 flag bits, continuation classes and FORBIDDENWORD), accent-first suggestions, conservative grammar with Portuguese messages, TeX Portuguese hyphenation patterns (BSD 3-clause); Review › Language offers Português (Brasil); a pt-br interface writes pt-BR documents; dimension ~27% → ~28%
2026-10-11	minor	Polish interface (13 languages) and Polish proofing: SJP.PL spelling with suggestions, Polish grammar rules, TeX Polish hyphenation; proofing language per run; dimension ~25% → ~27%
2026-10-11	minor	German interface catalog (#574): 2,021 entries covering the full zh-hans key set; locale, saved-preference and bundled-font tests; no German proofing
2026-10-11	minor	Italian interface catalog (#594), locale selection, saved preference and font coverage checks; count labels put the noun first so they read right at 1
2026-10-12	minor	Language-aware sorting (tables, paragraphs, index, bibliography, Table of Authorities) and table numbers/formulas in the text's language (#836, #837)
2026-10-11	minor	Font dialog sets the Asian text font (w:eastAsia) and the Latin text font separately; the Asian list holds the installed Chinese, Japanese and Korean fonts by their interface-language names (#505)
2026-10-11	minor	French interface catalog (2,033 entries, @pechnikov2026-star), fr-* system locales, saved preference and font coverage checks; no French proofing
2026-10-11	minor	Font list shows installed fonts by their names in the interface language (宋体 for SimSun in Chinese, Japanese and Korean names likewise) and finds them by either name; a Chinese, Japanese or Korean font is stored as w:eastAsia too (#505)
2026-10-11	minor	Chinese/Japanese interface fonts: a test checks every catalog character has a glyph; an installed CJK font follows the embedded faces whenever they miss one; macOS font asset folders searched for PingFang (#486)
2026-10-10	minor	Russian interface (#300; 1,524 entries, built-in style names shown translated; covers everything the interface audit finds)
2026-10-12	minor	Bulgarian interface catalog (2,419 entries), locale selection, saved preference, audit coverage and font checks
2026-10-11	minor	Hungarian interface catalog (2,419 entries, built-in style names shown translated; covers everything the interface audit finds), locale selection, saved preference and font coverage checks
2026-10-11	minor	Turkish interface catalog (1,307 entries), locale selection, saved preference, Turkish casing in command search and font coverage checks; catalog sizes re-counted
2026-10-11	trivial	English spelling (#520): our supplement assets/spelling/en-extra.txt adds ~220 common modern words the Moby list lacks ("desktop", "download", "email"…)
2026-10-10	minor	Estonian interface catalog, locale selection, saved preference and font coverage checks; count labels put the noun first so they read right at 1
2026-10-10	minor	Serbian (Cyrillic and Latin) interface merged (#250)
2026-10-10	major	First version: twelve-language table, catalog coverage measured, Word's 30 UI languages and 120 proofing tools listed