[ducduong97hn]

Fixes #329

Summary

Mammoth's default style map has no rules for Word's built-in Title and Subtitle paragraph styles, so DocxConverter emitted them as plain paragraphs and the document title lost its heading. For example, the existing test.docx fixture converted to:

AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation

Qingyun Wu , Gagan Bansal , ...

# Abstract

This adds two default style-map rules, Title → h1 and Subtitle → h2 (the mapping suggested as a workaround in the issue thread), so the same document now starts with:

# AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation

Details

• The rules match the style name, not the style ID. Localized Word versions write their own style IDs: test.docx stores Title as a3 and Subtitle as a5, but the names are still Title and Subtitle.
• The rules come after the caller's style_map and the document's embedded style map, so either one can still override them, e.g. style_map="p[style-name='Subtitle'] => p:fresh".
• Title-styled paragraphs inside table cells still render as plain cell text, the same as Heading 1 does today.

Tests

• The test.docx vector now expects # AutoGen: .... This fixture is a real Word document whose title uses a localized style ID, and the vector runs through the module and CLI entry points. It fails on main.
• test_docx.py: new tests for Title/Subtitle → #/## (using the fixture's localized IDs) and for a caller style_map overriding the default.
• markitdown-ocr: updated four DOCX snapshots whose fixtures start with a Title-styled paragraph (# Report, # Complex Document, ...).

Local runs (pytest, Python 3.10 on Windows): packages/markitdown 1174 passed, 3 skipped; packages/markitdown-ocr 108 passed. Black 26.10.1 reports no changes.

[ducduong97hn]

@microsoft-github-policy-service agree