[wittjeff]

Closes #745. Related: #868 (full BCP 47 tag on LanguageMetaField), docling-project/docling#4741 and docling-project/docling#4781 (PDF /Lang attached to body.meta, opt-in until this lands).

Problem

BaseMeta.language is metadata about the text, but once set it is treated as content:

1. Markdown and plain text print the field's pydantic repr (MarkdownMetaSerializer._serialize_meta_field has no LanguageMetaField branch and falls through to str(field_val)), so the document gains a paragraph such as confidence=0.99 created_by='langdetect' code=<HumanLanguageLabel.DE: 'de'>. EntitiesMetaField leaks the same way.
2. HTML has no lang attribute on the elements that carry the text.
3. HTMLParams.html_lang is declared but never emitted; the root is the hard-coded <html>.

Change

Markdown / plain text

• LanguageMetaField and EntitiesMetaField are skipped unless mark_meta=True, where they render as [Language] de and [Entities] IBM, Zurich.
• Any other structured field (a BaseModel) without a renderer is skipped instead of printing its repr. Custom scalar fields keep rendering as before.

HTML

• Root element is <html lang="...">. html_lang (params and export_to_html) now defaults to None, resolved by resolve_document_language(): doc.body.meta.language, else the character-weighted majority of the text items' meta.language, else "en". The resolution happens once in HTMLDocSerializer.model_post_init, so every element serializer compares against the same page language.
• <p>, <h1>..<h6> and <li> get lang="<code>" when the item's meta.language differs from the page language, compared by primary subtag (en-US vs en is the same language). This is WCAG 3.1.1 (page) and 3.1.2 (parts).
• The language row in the <details class="docling-meta"> block is unchanged: existing tests pin it, and it does no harm.

Ground truth

All .gt.html / .html.gt files are regenerated; the only change in each is <html> -> <html lang="en">. (The regeneration also rewrote the visualization PNGs through local font drift; those are reverted and not part of this PR.)

Tests

tests/test_metadata.py: Markdown and plain text exclude language and entities by default and list them with mark_meta=True; a structured custom field without a renderer is skipped while a scalar one is kept; HTML lang attributes with the majority default, an explicit html_lang="en-US", and a body-level language; <html lang="en"> without any language metadata. Full suite and pre-commit (ruff, mypy, pytest, docs) pass.

🤖 Generated with Claude Code

[github-actions]

✅ DCO Check Passed

Thanks @wittjeff, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🔴 1 of 2 protections blocking · waiting on 👀 reviews

	Protection	Waiting on
🔴	Require two reviewer for test updates	👀 reviews
🟢	Enforce conventional commit	—

🔴 Require two reviewer for test updates

Waiting for

• #approved-reviews-by >= 2

This rule is failing.

When test data is updated, we require two reviewers

• #approved-reviews-by >= 2

---

Show 1 satisfied protection

🟢 Enforce conventional commit

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?: