Machine-translate with a glossary and a do-not-translate list

How to do it

An engine that translates your product name, or picks a different word for "invoice" on each screen, makes one product read like several. Keep the glossary and the do-not-translate list in the repository or in the translation management system that holds the message files. Pass them to engines that accept them, as glossaries in the larger translation APIs do. Reuse approved translations from translation memory, so one source sentence never gets two translations. Placeholders and markup pass through unchanged. A fluent speaker still reviews the result, as Give translators context and a review step says.

Examples

JavaScript example

for (const [key, source] of Object.entries(sourceMessages)) {
  const translated = messages.sv[key];
  for (const term of doNotTranslate) if (source.includes(term)) assert(translated.includes(term), `${key} lost ${term}`);
  for (const [term, sv] of Object.entries(glossary.sv)) if (source.includes(term)) assert(translated.includes(sv), `${key} must use ${sv}`);
}

Python example

import re
for key, source in source_messages.items():
    translated = messages["sv"][key]
    assert set(re.findall(r"\{\w+\}", source)) == set(re.findall(r"\{\w+\}", translated)), key
    for term in DO_NOT_TRANSLATE:
        if term in source:
            assert term in translated, f"{key} lost {term}"

How to check

A test reads every translated message. Each do-not-translate term in the source appears unchanged, each glossary term uses its listed translation, and every placeholder in the source is present.

Sources

• Google Cloud Translation: Create and use glossaries

Related rules

• LOC-21 Give translators context and a review step in Localization