Arabic text extracted from PDF is reversed (both words and word's chars)

[Mahmoud-A-Noor]

Bug

when i extract text from pdf that contains arabic content i get arabic text in the wrong direction and chars of each word is reversed
...

Steps to reproduce

just try to extract text from any arabic pdf
...

Docling version

latest version
...

Python version

3.13
...

[dosubot]

Hi @Mahmoud-A-Noor! I'm Dosu and I’m helping the docling team.

This happens because Docling extracts text from PDFs using the PyPDFium2 library, but does not handle right-to-left (RTL) text direction or Arabic character shaping. As a result, Arabic text often appears reversed, both at the word and character level. This is a common issue with many PDF extraction tools, not just Docling, because PDFs often store RTL text in a visually ordered way that needs special handling to display correctly see code reference.

Currently, Docling does not have built-in support for RTL correction or Arabic text shaping. To fix the extracted text, you can post-process it using Python libraries like python-bidi (for bidirectional text reordering) and arabic-reshaper (for correct character shaping). Here’s a quick example of how you might do this:

from bidi.algorithm import get_display
import arabic_reshaper

def fix_arabic_text(text):
    reshaped_text = arabic_reshaper.reshape(text)
    return get_display(reshaped_text)

# Usage:
fixed_text = fix_arabic_text(extracted_text)

You’ll need to run this on each extracted text segment before using or displaying it. If RTL support is important for your workflow, you might consider integrating this step into your Docling pipeline.

There’s no official workaround or built-in feature for this in Docling yet, but you can track related development in PRs like #903, which touches on RTL handling for programmatic PDFs, though it doesn’t fully resolve Arabic extraction issues.

To reply, just mention @dosu.

---

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