/ Translation tools — document translation
Translate whole documents,
Translate whole documents,
get a translated copy back
Upload a .TXT, .DOCX, or .SRT subtitle file. The text is extracted, translated in segments, then reassembled into a translated file you can download — same format as what you put in.
/ 01 — SUPPORTED FORMATS
Three formats, one workflow
Each file type goes through a different extraction pipeline, but the end result is always the same: a translated copy in the original format.
.TXT
Plain text files
The entire file is read as a string, split into 500-character segments
at paragraph or sentence boundaries, translated in order, then written
back to a new
.txt file. No formatting to preserve —
the fastest format to process.
Output: translated.txt
.DOCX
Word documents
Text is extracted paragraph by paragraph using the mammoth.js
library. Each paragraph is translated individually, then all paragraphs
are reassembled into a new
.docx file using the
docx.js library with standard body formatting.
Output: translated.docx · Note: complex styles (tables, headers) are simplified
.SRT
Subtitle files
Each subtitle block is parsed separately — the index numbers and
timestamps are preserved exactly as-is, and only the dialogue lines
are sent for translation. The output is a valid
.srt
file ready to import into any video editor or player.
Output: translated.srt · Timestamps are never modified
/ 02 — TRANSLATION QUALITY
Where it works well, where it doesn't
Machine translation has improved enormously but it is not editing software. Knowing its limits helps you use it correctly.
Works reliably
- Informational or instructional text — manuals, guides, reports
- Subtitles for factual content — lectures, tutorials, interviews
- Correspondence and emails where tone is neutral
- Internal documents where a rough translation is enough to get the gist
- Languages with large training data — EN, ES, FR, DE, ZH, PT, AR, RU, JA, KO
Review carefully before using
- Marketing or persuasive copy — idioms rarely survive translation intact
- Legal contracts — machine translation is not a substitute for a certified translator
- Creative fiction — prose rhythm and wordplay are lost
- Low-resource language pairs — quality drops noticeably outside the major languages
- Highly technical jargon — domain-specific terminology can be mistranslated
/ 03 — QUESTIONS
Frequently asked
Does my document leave my device?
The file itself never leaves your browser — it is read into memory using the File API and never uploaded to any server. Only the extracted text segments are sent out, and they go directly to the translation engine (MyMemory or LibreTranslate). Neither service stores your text beyond the time needed to return a response. The translated file is assembled entirely in the browser and downloaded locally.
Why does my DOCX look different after translation?
Word documents can contain complex formatting — embedded images, tables, custom styles, headers and footers, tracked changes, and more. The mammoth.js extraction library pulls the plain text content reliably, but rebuilding the exact original layout in a new file is not feasible in the browser. The output is a clean, readable document with standard paragraph formatting. For layout-critical documents, manually paste the translated text back into your original file.
What is the maximum file size?
The tool accepts files up to 10 MB. In practice this covers virtually all text-heavy documents — a 10 MB plain text file contains roughly 10 million characters, and even a dense multi-hundred-page Word document rarely exceeds that. Very large DOCX files often contain embedded images that inflate the file size without adding to the text content; if your file is over 10 MB, try saving a copy with images removed.
Why does translation take a while for long documents?
Each translation API call handles a maximum of 500 characters, so a long document is broken into many sequential requests. A 10,000-character document generates around 20 API calls, each taking roughly one second. A small delay is added between requests to stay within the free usage limits of the translation service. The progress bar shows how far along the process is.
Can I translate between any two languages?
The tool supports over 40 languages. For the primary engine (MyMemory), many pairs route through English as a pivot — so a Portuguese-to-Japanese translation is actually two hops: PT→EN, EN→JA. This is standard practice for less common pairs and works well for informational content. If a specific pair produces poor results, try routing through English manually by doing two separate translations.
What happens if the translation service is down?
The tool uses two independent engines. If a request to the primary engine (MyMemory) fails or returns an error, it automatically retries using a LibreTranslate instance. If both fail for a given segment, the original untranslated text is kept in its place and the segment is marked in the status log so you know which parts need attention.
/ 04 — RELATED TOOLS
More tools from TopToolLab
🌐
Online Translator
Paste text and translate between 40+ languages instantly — no file upload needed.
Use tool → 🗜️ PDF CompressorShrink PDF file size before sharing translated documents with clients or colleagues.
Use tool → 📎 PDF MergerCombine the original and translated document into one PDF for side-by-side reference.
Use tool →