Summary
Add a Response Language setting that lets users select the language DeepSeek should respond in (e.g., English, Chinese, Hindi, Japanese, etc.), paired with a Strict Language Mode that prevents the model from unexpectedly switching languages mid-response.
Problem
Currently, DeepSeek can unexpectedly mix in text from another language (e.g., Chinese) even when the user is conversing in English or another selected language. This breaks predictability, especially in technical, professional, or multi-user contexts, and forces users to repeatedly re-ask or manually correct the output.
Proposed Solution
- Response Language Setting — a dropdown/selector where users choose their preferred output language.
- Strict Language Mode (toggle) — when ON, the model must not switch to another natural language unless the user explicitly requests it.
- No unexpected Chinese text when English is selected.
- No unexpected English, Chinese, or other-language text when a different language is selected.
- Persistence — the selected language preference persists across conversations until changed, with the option to override it on a per-conversation basis.
- Exceptions (always allowed regardless of mode):
- Code, programming keywords, package/library names
- URLs
- Mathematical notation
- Proper nouns
- User-supplied quoted text (should not be unnecessarily translated or modified)
- Explicit override — if the user explicitly asks for a different language, the model should temporarily comply with that request.
- Language-agnostic mechanism — this should work uniformly for all supported languages, not be a Chinese-specific patch.
Example
Settings:
- Response Language: English
- Strict Language Mode: ON
Expected:
"The Kubernetes pod is failing because the container cannot resolve the DNS server."
Not expected:
"The Kubernetes pod is failing because... [unexpected Chinese text]"
Benefits
- Predictable, consistent language behavior
- Reduces need to repeatedly re-ask or correct output
- Greater user control over communication style
- Improves usability for non-native speakers and multilingual teams
Summary
Add a Response Language setting that lets users select the language DeepSeek should respond in (e.g., English, Chinese, Hindi, Japanese, etc.), paired with a Strict Language Mode that prevents the model from unexpectedly switching languages mid-response.
Problem
Currently, DeepSeek can unexpectedly mix in text from another language (e.g., Chinese) even when the user is conversing in English or another selected language. This breaks predictability, especially in technical, professional, or multi-user contexts, and forces users to repeatedly re-ask or manually correct the output.
Proposed Solution
Example
Settings:
Expected:
Not expected:
Benefits