Maintain multilingual capabilities by adding languages, updating language-specific knowledge and voices, and managing pronunciation rules. Keep translations and behavior rules up-to-date as your business expands to new markets.
Quick reference
| I need to… | Action |
|---|
| Add a new language | Configure > General → Additional languages |
| Update language-specific knowledge | Edit topics with language variants or use <language:xx> tags |
| Change voice for a language | Channels > Voice > Agent Voice → select language card |
| Fix a translation issue | Channels > Response Control > Translations → edit the card |
| Add language-specific pronunciations | Channels > Response Control > Pronunciations → add rules per language |
| Test language switching | Agent Chat → select language from dropdown |
Adding or removing languages
Adding a new language
- Go to Configure > General and add the language under Additional languages
- Configure a voice for the new language in Channels > Voice > Agent Voice
- Add language-specific knowledge (see below)
- Add any necessary translation overrides
- Update behavior rules for the new language – use
<language:xx> tags to scope rules to specific languages
- Add pronunciation rules for the new language
- Test in Agent Chat using the language dropdown, then publish
Use a multilingual voice model (such as ElevenLabs multilingual or Cartesia sonic) for proper pronunciation across languages. See
Voice for available providers including ElevenLabs, Cartesia, Hume, Rime, Minimax, PlayHT, and Google TTS.
Removing a language
Remove the language from Configure > General under Additional languages, then update any language-specific knowledge, translations, pronunciation rules, and functions. The agent will fall back to the main language for callers speaking the removed language.
Language-specific knowledge
Managed Topics with language variants
- Go to Build > Knowledge > Managed Topics
- Create or edit a topic
- Add language variants for each supported language with translated or culturally-appropriate content
- Translate both sample questions and content for each variant
- Save and test
Sample questions must be in the same language as caller inputs – they are compared with user inputs during the retrieval process.
Conditional content with <language:xx> tags
For content where you don’t need full language variants, use <language:xx> tags to serve language-specific content in a single prompt:
<language:en>
Please hold while I check your account.
</language:en>
<language:es>
Por favor espere mientras reviso su cuenta.
</language:es>
Connected Knowledge per language
Add separate sources per language:
- URL sources – add the language-specific version of your website
- File sources – upload language-specific documents
- Integration sources – configure integrations (such as Zendesk or Gladly) to return language-specific content
Shared vs. language-specific knowledge
| Use shared knowledge when… | Use language-specific knowledge when… |
|---|
| Information is universal (phone numbers, addresses) | Content needs translation |
| Data is language-agnostic (product IDs, prices) | Cultural context matters |
| Maintaining multiple versions is impractical | Local regulations differ by region |
Updating voices per language
- Use native voices – don’t use an English voice for Spanish
- Match regional accents – Mexican Spanish for Mexico, Castilian for Spain
- Test pronunciation of language-specific characters
- Multilingual TTS models are convenient but may have slightly lower quality than language-specific models
Language detection and switching
By default, the agent detects the caller’s language automatically through ASR and responds in that language.
If detection is too aggressive or not sensitive enough, adjust in behavior rules – for example, instruct the agent to only switch after the caller has spoken consistently in a different language for multiple turns.
For explicit selection, create a Managed Topic that lets callers choose their language and use a function to set it:
def set_language(language_code):
conv.set_language(language_code)
return {"utterance": f"Switching to {language_code}."}
Language-specific functions
Accessing the current language
Access the current language in functions with conv.language:def dynamic_response():
current_language = conv.language
if current_language == "es":
return {"utterance": "Respuesta en español"}
else:
return {"utterance": "Response in English"}
Using translation cards in functions
For hard-coded utterances that need language-specific versions, use the conv.translations object instead of if/else branching. Create a translation card, then reference it by key:conv.translations.tn_greeting
getattr(conv.translations, "key with special chars")
What to translate
Not all project content needs translation. See the full reference table in Multi-language setup. Key rules:
- Keep instructions in English (e.g., “Ask for the user’s phone number”)
- Translate example utterances and scripted responses
- If content is directed at the agent, keep it in English. If it will be spoken aloud to the customer, translate it.
- Topic names, actions, function names, and Python code should stay in English.
Fixing translation issues
| Issue | Solution |
|---|
| Awkward phrasing | Add a manual override on the Translations page |
| Cultural mismatches | Use culturally-appropriate equivalents for idioms |
| Incorrect terminology | Use domain-specific terms with a glossary |
| Formatting issues | Localize date/time/number formats per language |
| Pronunciation issues | Add language-specific rules on the Pronunciations page |
Testing
- Open Agent Chat and select a language
- Verify the agent responds correctly
- Switch languages mid-conversation and confirm detection works
- Test with native speakers for naturalness and cultural appropriateness
Maintenance checklist
- Monthly: Audit content parity across languages
- Quarterly: Review voices for naturalness
- Ongoing: Update all languages together – don’t let translations lag behind
Related pages
