How does Docling assign <loc_xxx> tags in export_to_doctags()? Method or formula?

[HashirSaikhani]

Hi team,

When using 'DocumentConverter' --> 'export_to_doctags()', I get tags like '<loc_80>', '<loc_220>', etc.

Could you please explain how these loc numbers are assigned?
– Are they based on row numbers, text order, coordinates, or something else?
– Why do some text blocks and tables get the same loc?
– Is there a way to influence or read how they are computed?

Also, is there a link or file in the docling codebase where this tagging happens?

Thanks!

[dosubot[bot]]

The <loc_xxx> tags in export_to_doctags() are assigned based on the spatial coordinates (bounding box) of each document element, not on row numbers or text order. For each element, Docling normalizes its bounding box (left, top, right, bottom) to the page width and height, then scales these values to a fixed grid (default 500x500, controlled by the xsize and ysize parameters). The formula is:
<loc_xxx> = round(xsize * (coordinate / page_width)) for x-coordinates, and similarly for y-coordinates, clamped between 0 and xsize-1 (or ysize-1). Four such tags are concatenated to represent the bounding box of an element.

If two elements have identical or overlapping bounding boxes, they may get the same loc tag—this is common for text blocks and tables that share provenance or spatial coordinates.

You can influence how loc tags are computed by modifying the provenance bounding boxes (prov.bbox) of elements or by changing the xsize and ysize parameters when calling export_to_doctags().

The tagging logic is implemented in:

• docling_core/types/doc/document.py (export_to_doctags methods)
• docling_core/transforms/serializer/doctags.py (DocTagsDocSerializer and serializers)
• docling_core/types/doc/tokens.py (DocumentToken.get_location and get_location_token methods)

You can see the exact logic and adjust parameters as needed in these files.
See code reference

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}