Multiple OMML equations in one paragraph concatenated into a single display block

[smroels]

omml_multi_equation_paragraph.docx

Bug

When a single <w:p> paragraph contains multiple <m:oMath> sibling elements, docling
concatenates all of them into one $$ display block instead of emitting each as a separate
equation. The equations are also output in an unexpected order.

Given a paragraph with three sibling display equations (a = b, c = d, e = f), the
expected output is three separate display blocks:

$$a = b$$

$$c = d$$

$$e = f$$

Actual output is one block with all content merged (and order scrambled):

$$c=de=fa=b$$

This likely originates in the paragraph-level equation handling in
docling/backend/docx/ms_word_backend.py, where sibling <m:oMath> nodes within one <w:p>
are not iterated and split into individual equation items.

Steps to reproduce

1. Download the attached DOCX file.

2. Run:

   docling --from docx --to md --output . omml_multi_equation_paragraph.docx

3. Inspect omml_multi_equation_paragraph.md. The three equations (a = b, c = d, e = f)
   appear concatenated in a single $$ block rather than as three separate blocks.

The DOCX contains a single paragraph with three sibling <m:oMath> elements:

<w:p>
  <m:oMath>  <!-- a = b -->  </m:oMath>
  <m:oMath>  <!-- c = d -->  </m:oMath>
  <m:oMath>  <!-- e = f -->  </m:oMath>
</w:p>

This structure is produced naturally by Microsoft Word when a user places multiple display
equations in the same paragraph (e.g., by pressing Enter within an equation block and
continuing to type).

Docling version

Docling version: 2.79.0
Docling Core version: 2.69.0
Docling IBM Models version: 3.12.0
Docling Parse version: 5.5.0
Python: cpython-312 (3.12.12)
Platform: Linux-5.14.0-611.16.1.el9_7.x86_64-x86_64-with-glibc2.34

Python version

Python 3.12.12

Attachments

• omml_multi_equation_paragraph.docx — minimal three-equation paragraph that triggers the bug