t1k:translate
| Field | Value |
|---|---|
| Module | t1k-extended |
| Version | 2.21.0 |
| Effort | medium |
| Tools | — |
Keywords: code-switch, diacritics, dịch, dịch thuật, en-vi, english to vietnamese, loanword, localization, localize, machine translation tell, natural translation, not word by word, register, render vietnamese pdf, sounds like google translate, tiếng việt, transcreation, translate, translation, vi-en, vietnamese, vietnamese to english
How to invoke
Section titled “How to invoke”/t1k:translate<source-file-or-text> [--to vi|en|<lang>] [--mode doc|contract|marketing|ui|email] [--render]Translate — Read as Natively Written, Never Word-by-Word
Section titled “Translate — Read as Natively Written, Never Word-by-Word”Machine/AI translation defaults to swapping words in place. The result is grammatical but foreign — it “sounds like Google Translate.” This skill makes a translation read as if a native writer wrote it from scratch, while keeping every fact, number, and defined term exact.
Flagship pair: English ↔ Vietnamese. The method is general; the deep guides are VI-first.
When to use
Section titled “When to use”Activate before translating any document, email, UI string, or marketing copy — and when a draft translation “sounds translated.” Especially any English↔Vietnamese work.
The one rule
Section titled “The one rule”Translate the MEANING, not the words. Ask “how would a native writer say this idea from scratch?” — not “what is the word for each source word?” Restructure freely: split/merge sentences, reorder information, change voice. Naturalness wins except where precision must (see Domain modes).
Process — run every time, in order
Section titled “Process — run every time, in order”- Comprehend the whole first. Read the entire source. Fix in mind: literal meaning, intent (what the reader should do/feel), audience, register, domain. Never translate a sentence in isolation before you understand the whole.
- Flag the hard bits. Idioms, cultural refs, puns, defined/legal terms, numbers, proper nouns, jargon, ambiguous pronouns/tense. Decide per item: keep literal (precision) vs re-express (natural).
- Pick the register FIRST. Choose formality and — for Vietnamese — the pronoun/address pair before drafting. This is a decision, not a default. (VI has no neutral “you”.)
- Draft thinking natively. Compose in the target language as a native would. Reorder, restructure, change active/passive — do not mirror source syntax.
- Back-check for drift. Nothing added, dropped, or reversed; tone preserved; every number, date, amount, and defined term intact and consistent.
- Read aloud for naturalness. Any phrase a native reader would flag as “translated” → rewrite.
- QA pass. Run the checklist in
references/en-vi-linguistics.md§ QA.
Domain modes — how much freedom
Section titled “Domain modes — how much freedom”| Mode | Freedom | Rule |
|---|---|---|
contract / legal | Low — precision wins | Defined terms exact + consistent, clause structure preserved, never “improve” wording. Numbers/caps/law verbatim. |
doc / technical | Medium | Natural prose, but terms/numbers exact and consistent 1:1. |
marketing / transcreation | High — recreate the effect | Re-invent taglines/CTAs/puns for the same emotion; discard literal words. |
ui / microcopy | Medium + length limits | Fit the control; keep terse; consistent term set. |
email / message | Medium | Match the author’s voice; pair with t1k-human-writing. |
Precision always wins for: numbers, dates, currency, amounts, IDs, defined/glossary terms, proper nouns, code — carry verbatim; only reformat presentation when localization is wanted, never the value.
Keep-in-English (loanword) rule — critical for VI
Section titled “Keep-in-English (loanword) rule — critical for VI”VI tech/gaming/business/marketing speech is heavily code-switched. Over-translating is the #1 failure mode — forcing invented Vietnamese for terms professionals never translate reads stiff and wrong.
- Keep English when that is what practitioners say out loud: playable, landing page, microsite, brief, campaign, dashboard, deliverable, KPI, CPM/CPI/ROAS, deadline, feedback, deploy, build, P&L, COGS, markup, margin, sign-off, SOW, publisher.
- Translate where a settled native term is what the team actually uses: revenue → doanh thu, gross profit → lợi nhuận gộp, deposit → đặt cọc, discount → chiết khấu/giảm giá, data capture → thu thập dữ liệu, consent → sự đồng ý.
- Test: “Would a VI professional in this field say the English or the VI word out loud?” Match that.
- Never transliterate (invent VI spellings of English words). Keep the English, or use the real VI term.
Full term-by-term table (with the ⚠️ traps like discount floor = max discount, transfer price ≠
“chuyển giá”): references/vi-business-register.md.
Anti-patterns (the machine-translation tells)
Section titled “Anti-patterns (the machine-translation tells)”- Word-for-word / mirrored source syntax; no topic-fronting.
- VI: stamping đã/đang/sẽ on every verb; generic tôi/bạn everywhere (register never shifts); over-inserting của/một/các/những; English-shaped or wrong-valence passives (được for a bad thing); wrong/absent classifiers; missing or wrong diacritics.
- Translating do-not-translate terms (playable → “quảng cáo chơi thử”).
- Broken diacritics / tofu boxes in the rendered PDF (wrong font).
- Em-dash slop and other AI tells (see
t1k-human-writing).
Deep EN↔VI linguistics + the full AI-tells list + QA checklist:
references/en-vi-linguistics.md.
Numbers, currency, dates (VI)
Section titled “Numbers, currency, dates (VI)”VI inverts the Anglo separators: thousands = ., decimal = , (e.g. 1.234.567,89). VND is
zero-decimal (70.000.000 ₫). In prose use magnitude words — “70 triệu”, “1,5 tỷ” — not
full digits. Dates DD/MM/YYYY. ⚠️ 20.000 = twenty-thousand in VI but twenty-point-zero in the
US — always label the currency when both appear. Details:
references/vi-business-register.md § Formatting.
Render workflow (Vietnamese-safe PDF/DOCX)
Section titled “Render workflow (Vietnamese-safe PDF/DOCX)”When --render (or the source is a rendered doc), produce a VN-safe PDF. Wrong fonts break
Vietnamese (missing precomposed glyphs → tofu / “ransom note”).
- Save
.md/.htmlas UTF-8; normalize text to NFC (precomposed). - Use a Vietnamese-complete font: Be Vietnam Pro or Inter (embed via
@font-face), or the installed Noto Sans / Noto Serif fallback. NEVER Poppins (no VI subset — known broken). - Render
md → HTML(+VN-font CSS) → PDFvia weasyprint (deterministic embedding). Avoid thepdflatexengine for VI; if LaTeX is required usexelatexwith a VI main font. - Verify: render the test string and eyeball zero tofu / uniform weight, then confirm embedding.
(covers ĩ ỹ ướ ế ị ả ự đ ữ ắ — the horn+tone and circumflex+tone stacks). ThenNghĩ kỹ trước khi quyết định; ưu tiên chất lượng. Tự động hoá quy trình, giữ nguyên bản sắc.
pdffonts out.pdf→ each VI font showsemb=yes.
Full pipeline + font table + gotchas: references/typography-rendering.md.
References
Section titled “References”references/en-vi-linguistics.md— EN↔VI grammar pitfalls, AI-tells, idioms/jargon, QA checklist, reverse (VI→EN) direction.references/vi-business-register.md— VI business register, finance/marketing glossary (keep-EN vs translate, ⚠️ traps), number/currency/date formatting, before/after examples.references/typography-rendering.md— why fonts break VI, the safe-font table, the NFC +@font-face+ weasyprint pipeline, verification harness.
Related
Section titled “Related”t1k-human-writing— strip AI tells / match voice (run after translating prose).t1k-contract— render Markdown → HTML/DOCX/PDF (use the VN font stack here for VI docs).t1k-contract-review— for translated contracts, verify no legal-meaning drift.