Arabic text breaks silently. Not with an exception — with a string that still looks like Arabic, passes a review, and is a different thing underneath. Paste some below, then look at what a real tokenizer does with it.
Loading AraBERT's tokenizer in your browser…
This page is a demonstration. The real tool is pip install arabic-lint — it also scans source code for the get_display(reshape(text)) recipe, which needs a Python AST and cannot be done honestly in a browser.
Half the value is the silence: it reports nothing when nothing is wrong, and it stays quiet for ReportLab, for Pillow without Raqm, and for terminal print(), because the recipe is still correct there.
Built from an audit of 341 public Arabic datasets on this Hub. PyPI · GitHub · audit dataset · DOI · Syamjith NK, Abu Dhabi.