Most documentation is translated — either by human translators or machine translation engines like Google Translate, DeepL, or Microsoft Translator. The way you write in English directly affects how accurately and naturally your content reads in other languages.
The good news: writing that translates well is identical to writing that reads clearly in English. The Plain Language Principles guide establishes the core rules — short sentences, active voice, one idea per sentence, avoiding idioms. Follow those as your baseline for all content. This article covers what changes or additionally matters specifically for translation and localization.
Sentence Length and Structure
The plain language guideline of keeping sentences short takes on extra weight for translation. Sentences over 25 words are more likely to be split incorrectly by machine translation engines and produce ambiguous output. The 25-word ceiling is not just a readability recommendation — it is a translation quality requirement.
Similarly, compound sentences with multiple clauses are harder to translate accurately. If you find yourself joining two ideas with "which," "that," "although," or "however," consider whether they work better as two sentences.
Avoiding Idioms and Culturally Specific Expressions
Idioms do not translate. Phrases clear to English speakers are often meaningless or misleading in other languages. See the Idioms and Culture-Bound Language section of the Common Voice and Tone Mistakes guide, and the Culturally Neutral Language section of the Writing Bias-Free Content guide, for the full lists and reasoning. In a translation context the stakes are higher than in English-only contexts: a mistranslated idiom can produce content that actively misleads the reader, rather than simply sounding unnatural.
Also avoid directional language ("see above," "on the left") — these references break when text reflows in translation, when layout changes for right-to-left languages, or when content is restructured for mobile. See the Words That Can Cause Accessibility and Localization Problems section of the Words to Avoid quick reference for the full rule and alternatives.
Terminology Consistency for Translation Tools
Translation tools build glossaries from your source text. Inconsistent terminology — using "remove," "delete," and "erase" interchangeably for the same action — means inconsistent translations. Choose one term per concept and use it throughout; see the Terminology and Word Usage guide for the shared glossary this knowledge base already maintains, and its process for proposing a new term if one you need isn't covered yet. Whatever term is standardized there is also the term that should feed any translation tool's glossary.
Numbers, Dates, Currency, and Units
Localization goes well beyond words. See the Dates, Numbers, and Units for Localization guide for the full treatment — including translator-brief instructions, time zone abbreviation ambiguity, metric-vs-customary units, and phone number formatting — and the Numbers, Dates, and Time guide for the underlying formatting rules those localization practices build on.
Right-to-Left (RTL) Languages
If your knowledge base is localized into a right-to-left language (Arabic, Hebrew, and others), be aware that translation changes more than text direction:
- Numbered lists, breadcrumb navigation, and progress indicators may need to mirror horizontally, which is generally a platform/theme concern rather than something a writer manually adjusts per article — but it's worth confirming with whoever owns the platform that RTL rendering has actually been tested, not just assumed to work.
- Screenshots and diagrams that show directional UI elements (a sidebar on the left, a "next" arrow pointing right) may not match the mirrored layout of an RTL version of the product itself. See the Localized Screenshots section of the Images, Screenshots, and Video guide for how to handle this.
- Avoid embedding text inside images wherever possible, since embedded text can't be mirrored or translated the way real text can — this is also good general practice per the Avoid Screenshots of Text-Only Content guidance in that same article.
Writing for Machine Translation Quality
Beyond plain language, the following practices specifically improve machine translation (MT) output:
- Do not omit articles ("a," "an," "the") even when it feels natural in English — they provide grammar signals MT engines rely on.
- Write complete sentences. Fragment-style UI text ("Select to enable") is frequently mistranslated.
- Prefer positive constructions over negated ones: "To disable notifications, clear the checkbox" is clearer for MT than "Do not select the checkbox unless you want to receive notifications."
- Keep subject and verb close together. Long noun phrases between subject and verb confuse both translators and MT engines.
- Avoid parenthetical asides mid-sentence. Move them to a separate sentence or remove them.
Punctuation Signals
Translation tools use punctuation to determine sentence boundaries and clause structure. Inconsistent punctuation produces inconsistent translations. Use commas, periods, and sentence structure consistently throughout, per the Punctuation guide. Avoid excessive dashes, ellipses, or parenthetical nesting — these are particularly problematic for MT engines.
What to Flag for Human Review
Some content resists accurate machine translation and should be reviewed by a human translator regardless of MT quality elsewhere in the document:
- Warning and error messages — precision matters; mistranslation can mislead users into damaging actions.
- Legal or compliance language.
- Content involving humor or a conversational tone.
- Product names and trademarks — confirm they are handled correctly for each locale.
- Emotionally sensitive content.
Self-Audit Checklist
Before publishing content intended for translation, check the article against these questions:
- Are sentences kept under 25 words, treating this as a translation requirement, not just a readability preference?
- Have compound sentences with multiple clauses been split where it improves translatability?
- Is the article free of idioms and directional references ("see above," "on the left")?
- Is terminology for each concept consistent with the shared glossary, not varied across synonyms?
- Are numbers, dates, currency, and units presented in an unambiguous, locale-safe format?
- If the knowledge base supports RTL languages, have screenshots and embedded text been checked for compatibility?
- Are articles never omitted, and are sentences complete rather than UI-style fragments?
- Are constructions phrased positively rather than negated, where possible?
- Is punctuation used consistently, with dashes, ellipses, and nested parentheticals kept to a minimum?
- Has any warning, legal, humorous, trademarked, or emotionally sensitive content been flagged for human translation review?