BookTranslator
BookTranslator

What Not to Translate: Build a Do-Not-Translate List Before You Start

Decide what to preserve, localize, or translate in books and documents. Download a scoped DNT worksheet and inspect an annotated source-and-target example.

BookTranslator

BookTranslator Team

10 min read

Preserve strings whose exact identity makes something work: a model number, a URL destination, a variable, or the label printed on an untranslated control. Decide names, brands, titles, and specialist terms by context and target market. Translate the explanations that help readers use those things.

That is more useful than “never translate proper nouns.” An official localized brand name may be the right choice; an ordinary-looking word such as HOME may need to stay exact because it is engraved on a device.

A do-not-translate list, or DNT list, records these exceptions before work starts. The important part is not just the protected string. It is where the rule applies, why it applies, and how you will check the delivered file.

Download the DNT worksheet with worked examples. The examples are fictional training material, not customer documents or measured translation results. Replace them with decisions for your own source file.

Make three decisions, not one long exclusion list

For each candidate, ask these questions in order:

  1. Would changing the characters break identification or function? Preserve the exact string in that context. Examples include LD-204, a citation identifier, and the variable max_retries.
  2. Is there a target-language or market-specific form the reader should use? Check the appropriate authority, then record a deliberate localization decision. Do not let each translator improvise.
  3. Is it language intended to explain something to the reader? Translate it, using an approved glossary term where necessary.

“Preserve” is not a permanent property of a word. It is a rule for a particular use. In a software manual, the command print stays exact; the instruction “print the report” should be translated.

ItemPreserve exactlyDecide by market or projectTranslate for the reader
People, places, organizationsAn authoritative identifier or source spelling required for lookupEstablished target-language name, transliteration, first-mention dual formA descriptive explanation of the entity
Brands and product namesThe approved form for this editionAn official localized name, if applicableProduct descriptions and benefits
Book and article titlesThe title of the specific edition being cited when required by citation styleThe title of your new translated edition; an established published translationDescriptive headings and surrounding commentary
URLs, email, pathsThe actual destination, address, or required pathA verified replacement destination or localized file structureLink labels and explanatory text
Code and variablesSyntax, identifiers, placeholders, and literal values required by the programUser-facing strings inside code, according to the example's purposeComments and explanations when they are part of the translation scope
Numbers, units, standardsThe quantity's meaning, model codes, standard identifiersDecimal notation, date presentation, or authorized unit conversionLabels, measurement explanations, and instructions
Tables, formulas, metadataStructural identifiers and functional expressionsEdition-specific identifiers, metadata language, field behaviorTable headers, captions, descriptions, and alternative text

For the detailed choice between retaining, transliterating, or adapting a name, use the names and place names decision guide. A DNT worksheet should record the resulting decision, not replace that research.

Separate the DNT list from the glossary and style guide

These assets answer different questions, even when they live in the same workbook.

AssetQuestion it answersExample decision
DNT listWhat must remain exact, and where?Keep HOME when referring to the button printed on the fictional LeafDeck device
Translation glossaryWhich target term expresses this concept here?Use the approved French term for a workspace in the product interface
Style guideHow should the translated language read and look?Use the formal form of address and the project's heading capitalization

Do not put a difficult technical term on the DNT list simply because nobody has resolved it. Mark it as a terminology question. Do not label an unfamiliar institution “untranslatable” when the reader needs an explanation.

For terminology decisions rather than exclusions, build a translation glossary. Keep unresolved decisions out of the approved handoff until someone has answered them.

Give each exclusion a scope and an acceptance check

A weak DNT entry says Home — do not translate. That could freeze an entire navigation concept, a household reference, and a hardware label at once.

A usable entry says:

  • Exact source: HOME
  • Scope: references to the physical button in the LeafDeck quick-start guide
  • Required target: HOME, with the same capitalization
  • Reason: the button itself is not localized in this fictional edition
  • Matching rule: whole label, case-sensitive; not the ordinary noun “home”
  • Owner: the person responsible for the device documentation
  • Check: compare every button reference with the approved label and the device artwork

The download also includes columns for the source location, target locale, decision status, evidence, and review note. Before sending it to anyone, filter out the demo rows and unresolved decisions. Freeze a version alongside the source document.

Use exact matching for codes and identifiers. Use contextual review for words that can appear as both a protected label and normal language. A substring rule for CAT, for example, should not freeze part of “category.”

Apply the rule differently to books, PDFs, and Word files

Books: preserve the reference without freezing the prose

In an illustrated publishing example, a manuscript discusses a fictional book called The Glass Orchard. The title in its bibliography identifies a particular source edition. The new chapter heading “Walking through the glass orchard” is a sentence with a different job.

Record the cited title's treatment and the edition it identifies. Translate the chapter heading. If the target edition uses a translated title for the cited work, document that choice and keep the citation traceable. Do not replace every occurrence of the phrase because one bibliography entry is protected.

A quotation also needs an explicit policy: original language, a published translation, your own translation with attribution, or both. Quotation marks alone do not mean “leave this entire passage untranslated.”

PDFs: distinguish visible labels from technical data

Imagine a PDF specification table containing the model LD-204, the heading “Input voltage,” and a value of 24 V.

The model remains exact. The heading is translated. The voltage must retain its meaning; it is not a phrase to creatively rewrite. If a project authorizes unit conversion or localized numeric presentation, record the transformation and independently check it. A DNT list is not a substitute for numerical QA.

For a scanned PDF, inspect the recovered text before checking preservation. If OCR reads LD-204 as LD-2O4, keeping the extracted string unchanged preserves the error. Compare identifiers with the source image, not only with the OCR text.

DOCX: protect fields and destinations, not every visible result

In a Word manual, a hyperlink's destination and its visible label have different jobs. The destination may remain unchanged while “Download the reference guide” becomes target-language text.

Likewise, protect the structure behind a cross-reference while translating its surrounding sentence. Recheck the displayed reference after export; do not assume it still points to the right section because the sentence is fluent. Formula symbols can stay exact while their descriptions and table headings change.

This is a useful extension of source-file preparation: classify content by function before handing off the complete file.

Inspect a marked source and an expected result

Here is an original training sentence from a fictional device guide:

Press HOME on LeafDeck LD-204. Save the log to logs/session.txt, then set max_retries to 3.

An unmarked sentence leaves the translator to infer whether HOME is a physical label, whether LeafDeck has a localized name, and whether the path and variable are literal.

For an HTML-aware workflow, the following source makes the intended boundaries explicit:

<p>
  Press <span translate="no">HOME</span> on
  <span translate="no">LeafDeck LD-204</span>. Save the log to
  <code translate="no">logs/session.txt</code>, then set
  <code translate="no">max_retries</code> to 3.
</p>

The W3C explains that HTML's translate="no" marks an element's content as not to be translated and that the setting is inherited by its descendants. It is a processing instruction, not a visible lock. A tool must support and retain it for the intended protection to work. See W3C guidance on the translate attribute.

A manually authored French expected result is:

Appuyez sur HOME sur LeafDeck LD-204. Enregistrez le journal dans logs/session.txt, puis réglez max_retries sur 3.

The surrounding instruction changes; the label, product/model, path, variable, and value remain. This is an acceptance example, not output from an AI translation test. It demonstrates what to inspect, not a measured tool capability.

Open the complete unmarked source, marked source, and manually authored expected result. The files include a link example: its reader-facing text changes while its destination does not.

Do not paste HTML tags into a PDF or DOCX and expect equivalent behavior. Use the source format's supported workflow, or keep the worksheet as a handoff and review reference. We are not claiming that BookTranslator consumes this markup or imports the worksheet.

Catch overprotection as well as unwanted translation

A DNT review has two failure directions:

  • Underprotection: max_retries becomes a translated phrase or LD-204 changes character.
  • Overprotection: the entire troubleshooting paragraph stays in English because it contains a protected model number.

Protect the smallest meaningful span. A blanket rule on a whole paragraph or table can also leave reader-facing descriptions untranslated. HTML attributes such as alternative text and titles require attention too; the W3C's attribute guidance distinguishes translatable text from structural attributes. Inspect both visible and non-visible content.

Before accepting the deliverable:

  1. Verify the source. Resolve OCR errors and confirm that each DNT decision applies to this edition and locale.
  2. Compare protected strings. Check spelling, case, punctuation, digits, and any approved substitutions against the worksheet.
  3. Inspect context. Confirm that the right occurrence was protected and that surrounding instructions were translated.
  4. Exercise the result. Open links, follow cross-references, inspect code or formula examples, and compare hardware labels where relevant.
  5. Review the final export. Check the actual PDF, EPUB, or DOCX, not only an intermediate text extraction.

Counts can identify suspicious changes, but they are not proof. A glossary appendix can increase the number of occurrences; a missing chapter can reduce it. Explain the difference instead of adjusting the count to make a check pass.

Take the list into translation, then keep it for review

Once the source and exceptions are ready, use a complete-file translation workflow for the supported book or document you need to translate. Keep the DNT worksheet beside the source as your acceptance reference. Do not assume an automatic glossary is equivalent to your approved exclusions, or that every uploaded format carries the same markup controls.

After translation, use the terminology-consistency workflow to check approved and rejected forms across the complete work. The DNT list answers what should remain exact; the wider review answers whether the translated document still means and does what it should.

Gerelateerde berichten