DoclingDocument Text Formatting#

Text formatting metadata in DoclingDocument is modelled by two types defined in docling_core/types/doc/common/formatting.py:

• Script — a str enum with three values: "baseline" (default), "sub" (subscript), "super" (superscript).
• Formatting — a Pydantic BaseModel with four bool fields (bold, italic, underline, strikethrough, all defaulting to False) and one Script field (script, defaulting to Script.BASELINE).

Formatting lives as an optional field on TextItem (and all its subclasses — TitleItem, SectionHeaderItem, ListItem, FormulaItem, etc.). When None, no formatting is applied. See the DoclingDocument Builder API for how to pass a Formatting instance when constructing items.

JSON Serialization#

Because Formatting is a Pydantic BaseModel, it round-trips to/from JSON automatically via model_dump(). When present, the "formatting" key appears inline in the text item object with all five fields explicitly serialized 1:

// subscript example
"formatting": {
  "bold": false,
  "italic": false,
  "underline": false,
  "strikethrough": false,
  "script": "sub"
}

All five fields are always emitted — there is no sparse serialization. The script field takes the string literal values "baseline", "sub", or "super" matching the Script enum 2.

DocLang XML Serialization#

In DocLang XML, each active formatting attribute wraps the text content in a dedicated XML element. The mapping is defined by DocLangToken and implemented in DocLangDocSerializer:

Tags can be nested for combined formatting. The fixture content_all.gt.dclg.xml demonstrates this, including a case where bold, italic, underline, and strikethrough are all applied simultaneously via nested wrapping.

The serialization order and nesting are controlled by the post_process() method on the base DocSerializer. It applies wrappers in a fixed sequence — bold → italic → underline → strikethrough — and then script (subscript/superscript are mutually exclusive, handled via elif). Individual serializers override serialize_bold(), serialize_italic(), etc.; the base class stubs are no-ops that return text unchanged 3.

The post_process() hook runs only when the serialization params have include_formatting=True (the default). Setting include_formatting=False strips all formatting tags from output.

Key Source Files#