What should I do to contribute to translating Three.js documentation into Turkish?

I would like to contribute to translating the Three.js documentation into Turkish. How can I get involved in this process?

Pls see this thread (also this answer in a separate thread, for a shorter explanation.)

The best way to create a Turkish version of the docs would be to create and maintain it in a separate repo dedicated just for that - then update it every new three.js release to keep it up to date (that’s the hard part usually, translating stuff initially is exciting and fun - tracking changes for an indefinite amount of time is a chore :smiling_face_with_tear:)

1 Like

True, dare I say a LLM may be an ideal candidate to automate this process?

EDIT: @mirzasahin does this check out…? (taken from here and translated through gpt)

Kurucu
BufferGeometry()
Bu, yeni bir BufferGeometry oluşturur. Ayrıca birkaç özelliği varsayılan bir değere ayarlar.
Özellikler
.attributes : Object
Bu hashmap, ayarlanacak özelliğin adını id olarak ve değeri olarak ayarlanacak tamponu içerir. Bu özelliğe doğrudan erişmek yerine, bu geometrinin özelliklerine erişmek için .setAttribute ve .getAttribute kullanın.

1 Like

For Western languages GPT may work - but for Polish it’s already sounding incorrect-ish (ie. grammar and words are sometimes situational, so while translations are understandable from context, words used are incorrect.)

Also asked my friend to check English-to-Thai translation by GPT - it was 100% incorrect :smiling_face_with_tear:

But from positives - there could be better models than GPT, which is very generalised, trained specifically for translation prompts :eyes:?

3 Likes

Yes, I have a feeling that training a custom model with Ludwig on current (+ legacy) 3js documentation, + any currently maintained (+ outdated) translated documentation could definitely cover a lot of cases for an automated bilingual documentation framework :thinking::eyes: @mirzasahin this would be an approach I’d personally look into because as @mjurczyk has suggested, 3 breaking iterations in and I’m sure any sane person would eventually crack having to manage migratory docs without some sort of automated system in place…

It’s kinda mindblowing that GPT which is the state of the art, can’t do perfect translation yet, when that was the primary usecase for AI up until LLMs got so crazy. Like… did they forget to train against the same corpus that google translate used?

1 Like