Translate Word documents. Formatting intact. Tracked changes survived.
Drop your .docx and get it back translated with styles, tables, headers, footers, comments, and tracked changes preserved. Open it in Word — looks identical, except translated.
What is Word DOCX?
A .docx is a ZIP container with document.xml at its core plus styles, settings, headers/footers, comments, and tracked-change history. Most AI translation tools either flatten to plain text (losing every style, bullet, table cell, run boundary), translate XML naively (breaking every reference between runs and styles), skip headers/footers, or drop comments and tracked changes.
Tracked changes that round-trip across languages.
An LSP came to us with this real workflow: client edits the English document with Word tracked changes, and you've already translated v1 to Spanish. Now the client wants v2 in Spanish — with the same tracked changes applied — not a fresh re-translation that loses the audit trail.
We built a delta-aware DOCX pipeline. Upload three files: the original English, the English with redlines, and Spanish v1. Get back Spanish v2 with equivalent insertions and deletions marked in Spanish. The reviewer's audit history survives. Hours instead of days per review round. No other AI translation tool ships this.
Why Word DOCX is tricky for AI translation
- Run boundary respect: a sentence may be split across 5 <w:r> runs because of inline formatting (bold, italic, hyperlinks). Translating each run independently produces fragmented translations.
- Style preservation: <w:rPr> and <w:pPr> must survive untouched.
- Hyperlink handling: <w:hyperlink> wraps runs that contain display text. The URL stays bound; the display text is translated independently.
- Tracked changes: <w:ins> and <w:del> blocks need special handling — preserve, accept, or generate new tracked changes against the original.
- Comments: <w:commentRangeStart/end> reference comment IDs in comments.xml. Both pieces need to be translated and re-linked.
- Headers and footers: stored as separate parts (header1.xml, footer1.xml) and must be translated independently.
- Tables: cell-level translation with row/col structure preserved.
- Table of contents: <w:fldSimple> and <w:instrText> fields contain non-translatable instructions mixed with translatable display text.
How Fily handles Word DOCX
- Run-aware extraction: sentences reassembled from runs into translatable units, then re-split back into runs preserving every <w:rPr> and inline formatting boundary.
- Zone-based parsing: runs inside <w:hyperlink>, <w:sdt>, <w:ins>, and <w:del> are detected and routed correctly.
- Headers and footers: each header*.xml and footer*.xml part is translated separately and reattached.
- Comments: translated and re-linked via comment ID. Reviewer's voice stays.
- Tracked changes (insert/delete): default preserves <w:ins>/<w:del> as-is. Tracked-delta mode can ingest EN + tracked changes + ES_v1 and emit ES_v2 with the same tracked changes applied in Spanish.
- Tables: cell-by-cell translation; gridSpan and merge structure preserved.
- Fields and TOC: field instructions skipped, display text translated, links re-rendered on next Word open.
Pipeline: word_qa_12step_v2@1.0.0 · docx_tracked_delta_v2@1.0.0 (delta mode)
The Word DOCX workflow with Fily
Upload
Drop your .docx (single or batch ZIP). Optional: glossary, TM, style guide.
Process
Fily runs the Word DOCX pipeline + 12 QA steps. Typical job: 10–20 minutes.
Download
Same format, ready to deliver. QA report HTML attached.
Standard use is uploading a Word document and getting it back ready to deliver. For documents that already have a reviewed translation (ES_v1) and a new English edit with tracked changes, Fily's delta mode produces ES_v2 with the same tracked changes applied — saving the reviewer from re-translating the whole document.
Beyond the standard pipeline
What we've built around Word DOCX
Edge cases clients brought us for this format — and the pipelines we shipped to solve them.
docx_tracked_delta_v2
Tracked changes that survive translation
An LSP had already translated v1 of a Word doc to Spanish. The client edited the English with tracked changes and wanted v2 in Spanish with the same redlines applied — not a fresh re-translation.
We built a delta-aware DOCX pipeline that ingests EN + redlines + ES_v1 and emits ES_v2 with the equivalent insertions and deletions in Spanish. The reviewer keeps the audit trail — hours instead of days per round.
mendoza_fda_delta
Delta translation against an approved TM
A regulated-content client updates FDA-reviewed documents constantly, but only a few lines change each cycle. Re-translating the whole file risked drifting away from language a regulator had already approved.
A delta pipeline that diffs the new source against the previous version, translates only what changed, and locks everything else to the approved TM byte-for-byte. New wording gets the AI; approved wording never moves.
docx_pair_review
Two Word docs in, a reviewed delta out
A client sent the original DOCX plus a reviewer's corrected copy and asked, in effect, 'what did we actually change, and can you apply the same corrections going forward?'
A pair-aware pipeline aligns both files, extracts the human corrections as tracked changes, and surfaces them in the browser review editor — billed only for segments that actually changed.
Golden TM · auto-build
A translation memory that builds itself
Clients wanted TM leverage but didn't want to maintain a TM — exporting, cleaning and re-importing TMX after every project is a job nobody has time for.
Every confirmed segment is promoted into the organization's Golden TM automatically, scoped per client and language pair, and reused on every future job. Reviewer edits in the dashboard feed straight back in. Exact matches then cost $0.
Frequently asked about Word DOCX
Other formats Fily translates
Ready to translate a Word DOCX file?
No card. No setup. Upload one file and see the output.