Support automatic fill-in of internal TOCs for non-English Languages

[wonhyeongseo]

Description

Hello @mishig25 ,

I am writing to raise a concern about the lack of support for automatic fill-in of internal TOCs (Tables of Contents) for non-English languages in the huggingface documentation.

For example in the vietnamese course, Giới thiệu(introduction) is converted as gii-thiu(good-natured) in the url.

Putting an english anchor should be necessary for all of the documentation regardless of language. This is because when people switch languages (from en->vi), the current system disallows them to see the same part of the documentation. This is troublesome for long api docs. Ideally, when a user makes a simple change in the language code, they will get to the section they want fast. Note: Right-to-Left languages like Arabic have a unique linking structure, and I couldn't figure out where to put the custom anchor.

Please refer to huggingface/course#376 for details.

Expected Behavior

Ideally, the software should be language agnostic. The automatic TOC fill-in functionality should recognize headers in any language, extract them, and populate the TOC accordingly. But as a first step, extraction of the English anchor and automatic fill-in to other languages is fine.

Current Behavior

As of now, when working with non-English languages, we have to manually add in these custom anchors inside double brackets [[anchor]] when translating. The internal TOC remains blank or fails to function if not handled this way. Contributors are left to manually create TOCs, which can be a tedious process for extensive documents.

Steps to Reproduce

1. Navigate to https://huggingface.co/learn/nlp-course/vi/chapter1/1#gii-thiu
2. Replace nlp-course/vi to nlp-course/en in the URL.

Possible Solution

This problem could potentially be resolved by inserting the custom English headers only at build-time.

I am hoping for a positive and prompt response. Please let me know if you need any further information. I intend to work on this issue with @jinnsp and @gabrielwithappy to send a PR.

Thank you for your attention to this matter.

[gabrielwithappy]

Dear @wonhyeongseo
I do believe this will help many contributors! :-)