Add more language support for humanization

[rudra496]

Description

Currently humanization works best in English. Add better support for other languages.

What to do

• Test with Spanish, French, German, Bengali, Hindi text
• Add language-specific prompts in lib/prompts.ts
• Ensure the UI language dropdown works for all languages
• Document supported languages in README

Difficulty

Beginner-friendly

[voidborne-d]

Great initiative! One language worth prioritizing is Chinese (Mandarin) — the AIGC detection landscape in China is completely different from English.

A few things to know if you go down that path:

• Chinese AI detectors (CNKI AIGC, Wanfang, VIP/Weip) use different fingerprinting than GPTZero or Originality.ai. They look for sentence-level statistical patterns and perplexity distributions specific to Chinese grammar.
• Prompt-based rewriting through LLM APIs often does not work well for Chinese academic text because the detectors flag the output style of major models specifically.
• Chinese text humanization needs to handle classical idiom insertion, sentence structure variation (把/被 constructions, topic-comment reordering), and register shifts between formal/academic and natural Chinese.

For reference, there is an open-source Python library focused specifically on this problem: humanize-chinese. It does 0-100 AI probability scoring and has 7 style transforms designed for Chinese text patterns. Pure Python, zero dependencies. Might be useful as a reference for how Chinese-specific humanization differs from English approaches.

Happy to help if you want to discuss Chinese language specifics!

[rudra496]

Thanks for the detailed insights on Chinese humanization — this is exactly the kind of domain knowledge that makes open source projects better. 🙌

A few thoughts:

Chinese support is a priority. We definitely want to include Mandarin (Simplified & Traditional) in the next phase. The challenges you outlined — detector fingerprinting differences, LLM output style flagging, and the need for structural variations like 把/被 constructions — are super relevant and not something I'd have caught without this input.

The reference library helps. I'll look into humanize-chinese as a reference for how Chinese-specific transforms differ from our current English-first approach. Different grammar patterns need different strategies, and that project seems to validate exactly what you're describing.

I'd love your help on this. If you're interested in contributing, here's what would be most valuable:

• Chinese-specific prompts in lib/prompts.ts (sentence structure variation, register shifts, idiom insertion)
• Testing against CNKI AIGC / Wanfang detectors if you have access
• Guidance on what "natural" vs "AI-generated" Chinese looks like in academic contexts

No pressure either way — even just pointing out edge cases or reviewing PRs would be a huge help. Let me know if you'd like to be assigned to a Chinese-specific issue once I create one.

Thanks again for stopping by! 💪