Translate pptx

Translate PPTX

Translate PowerPoint presentations while preserving all formatting and design.

All intermediate files (unpacked XML, JSON) are written to a temporary directory and cleaned up automatically.

Prerequisites

• defusedxml — XML parsing
• lxml — XML validation
• markitdown[pptx] — text extraction for review

Workflow

1. Analyze the source presentation

python -m markitdown input.pptx

Identify the source language and total text volume.

2. Confirm the target language

If the user has not specified a target language, ask them to choose using AskUserQuestion:

Options:

1. Japanese
2. Chinese
3. Spanish
4. Others (ask the user to specify)

Map the selection to a language code: ja, zh, es, etc.

3. Create a temporary working directory

Use Python's tempfile.mkdtemp() to create a temp directory for all intermediate files. Store the path in a variable for use in subsequent steps.

import tempfile
tmpdir = tempfile.mkdtemp(prefix="translate_pptx_")
unpacked_dir = os.path.join(tmpdir, "unpacked")
texts_json = os.path.join(tmpdir, "texts.json")

4. Unpack

python scripts/office/unpack.py input.pptx $tmpdir/unpacked/

5. Extract translatable text

python scripts/extract_texts.py $tmpdir/unpacked/ $tmpdir/texts.json

This produces a JSON file with all <a:t> text segments from slides and speaker notes, each with its file path and index.

6. Translate

Read texts.json and translate each text segment. Add a "translated" field to each entry.

Translation guidelines:

• Preserve the meaning, tone, and register of the original
• Keep proper nouns, brand names, and technical terms unchanged unless a standard localized form exists
• Maintain placeholders, numbers, URLs, and email addresses as-is
• Adapt idioms and cultural references naturally rather than translating literally
• Keep text length comparable — significantly longer translations may overflow text boxes
• For CJK to Latin translations, expect ~30% length variation; abbreviate if needed to fit

CRITICAL: Handling word-by-word segments

Text in PPTX is split across individual <a:t> XML elements, often word-by-word. When translating:

1. Group consecutive segments within each slide that form a logical phrase or sentence
2. Translate the FULL phrase/sentence as a unit
3. Put the complete translated text in the first segment of the group
4. Set remaining segments in the group to empty string "" — the apply_translations.py script will replace these with empty text in the XML, removing the original English

For standalone segments (single words that are independent labels), translate directly.

Process for large presentations (>50 segments):

1. Read texts.json
2. Split into batches by slide groups
3. Use subagents for parallel translation — each batch handles ~5-6 slides
4. Merge results back into texts.json

Process for small presentations (≤50 segments):

1. Read texts.json
2. Translate all segments at once
3. Write the completed file

7. Apply translations

python scripts/apply_translations.py $tmpdir/unpacked/ $tmpdir/texts.json --lang <target>

The --lang flag sets the language attribute on text runs. Use short codes: en, ja, zh, ko, fr, de, es, pt, it, ru, ar, hi, th, vi — or a full BCP 47 tag like en-US.

8. Clean and pack

python scripts/office/clean.py $tmpdir/unpacked/
python scripts/office/pack.py $tmpdir/unpacked/ translated.pptx --original input.pptx

The output file should be placed in the same directory as the input file, with a language suffix appended (e.g., input_JA.pptx).

9. Cleanup

Remove the temporary directory:

import shutil
shutil.rmtree(tmpdir)

10. QA

python -m markitdown translated.pptx

Verify:

• All text segments are translated (no source language remnants)
• No placeholder or template text remains
• Proper nouns and brand names are preserved
• Numbers and formatting are intact