Format-Preserving Document Translation

Problem

Translating office documents is not just swapping language. The result still needs to be editable and structurally aligned: Word paragraphs, Excel cells, PPT text boxes, and PDF layout should map back to the source positions, while terms, field keys, function names, and code snippets must not be mistranslated.

How It Works

The skill follows an “extract text—translate—replace text” workflow for docx, xlsx, pptx, and pdf:
- Format detection: selects the matching script by file extension and emits JSON extraction results.
- Text extraction: supports paragraphs, table cells, slide text boxes, and Word sdt structured document tags.
- Controlled translation: reads references/terminology.md, keeps listed terms unchanged, and leaves identifiers such as user_id, api_key, and getUserInfo() intact.
- Replacement: writes translated text back by element_index, preserving paragraph, cell, or text-box placement.

Boundaries And Notes

• Office documents: length changes may cause automatic line reflow.
• PDF: higher risk of layout loss; review the output before delivery.
• Special content: text inside images is not processed by default; math formulas are not translated, while explanatory annotations under formulas can be.
• Fields: update table of contents and page numbers manually after translation.