DeepL is, by most benchmarks and popular consensus, the strongest general-purpose machine translation service. This is not a challenge to that - it is a look at a specific gap: translating technical documents where code must survive translation untouched.
What DeepL Does Well
For natural-language text, its fluency and nuance are excellent, document-format support is broad, and the glossary feature pins specific terms. For a business email or a marketing page, it is a safe recommendation.
Where Translation Meets Code
Translation models translate text - and in technical documents some of that text is not prose:
-
Code blocks: a fenced
``
python block may come back with translated comments, translated string literals, or reflowed formatting - any of which breaks it - **Identifiers:**getUserName,MAX_RETRIES,user_idcan get casing-corrected or partially translated - **Technical terms:** "commit", "deploy", "race condition" have precise meanings; a fluent paraphrase can quietly change the meaning - **Commands and paths:**cd /var/log/app` is not a sentence and should never be touched
The Two-Step Alternative
A code-aware translation flow splits the job: skip the code, translate only the prose, pin identifiers and terms. Lint's translator works this way - code blocks, commands, and technical terms pass through untranslated, so output goes straight back into your docs.
| General translator | Code-aware translator | |
|---|---|---|
| Best at | Natural-language documents | READMEs, tutorials, API docs |
| Code blocks | May be altered | Untranslated, byte-identical |
| Identifiers/terms | Depends on glossary setup | Skipped by design |
Which to Use
Contract, email, article? Use the strongest general translator - that is DeepL's home turf. README or tutorial with code in it? Use a code-aware flow, or plan a manual pass to repair every snippet.
👉 Try Lint's code-aware translator - code stays code, free tier available.
Top comments (0)