Traditional Chinese (zh-TW) queries always get Simplified Chinese output #117
Description
When I enter a question in Traditional Chinese (zh-TW), the output is always in Simplified Chinese, even if the prompt or system language is set to zh-TW. This makes it impossible to get a proper Traditional Chinese result.
Possible cause:
Most LLMs and language detectors tend to default to Simplified Chinese, since there's much more zh-CN content and data on the internet compared to zh-TW. In the codebase, there are only schema examples for generic Chinese (zh), so the model almost always picks that (which is handled as zh-CN), and never zh-TW, even if zh-TW is in the i18n mapping.
Suggestion:
- Change any generic Chinese (zh) example to zh-CN, to make the intent clear
- Add a dedicated example for zh-TW, so queries can be matched and handled properly
- Double check that the language detection and mapping logic lets zh-TW queries actually get zh-TW output
Would like to hear thoughts from maintainers and other users, thanks!