About multilingual support

Overview

Your AI Agent can communicate with end users in their preferred language. Use multilingual functionality to provide a personalized experience for end users across different regions and languages.

For setup instructions, see Multilingual support setup.

Limitations

Multilingual support has the following constraints:

English is always enabled: English is the default language and cannot be disabled. Your AI Agent always falls back to English when it cannot determine the end user’s language or find content in their language.
Dashboard language: The Ada dashboard is only available in English. Browser translation tools may translate some items inaccurately.
Channel support varies: Not all languages are supported across all channels. See Supported languages for details.
Knowledge language support: Only a subset of languages support non-English Knowledge articles. See How Ada uses multilingual knowledge.
Language detection limitations: Some languages cannot be detected automatically. See Language switching for details.

Use cases

Multilingual support enables your AI Agent to serve end users across languages and regions.

Global end user support: Communicate with end users in their preferred language.
Seamless language switching: Allow end users to change languages mid-conversation based on their preference.
Consistent brand experience: Customize Greetings and Handoff messages with manual translations to maintain brand voice across languages.

Capabilities & configuration

Multilingual support provides flexible language handling across channels.

60+ supported languages: Communicate with end users in a wide range of languages across Web Chat, Email, and Voice channels.
Automatic language detection: Your AI Agent detects the end user’s language and responds accordingly.
Language switching: End users can switch languages mid-conversation via autodetect, typing a language name, or using chat settings.
Translation methods: Ada uses either Google Translate or LLM-based translation depending on the language.
Knowledge localization: Serve region-specific Knowledge content to end users based on language and locale.
Manual translations: Customize translations in Greetings and Handoffs for specific languages.

Supported languages

Ada supports many languages, though support across channels and features varies. In the table below:

Web chat, Email, and Voice indicate whether the AI Agent can speak that language in those channels.
Knowledge indicates whether the AI Agent can read Knowledge articles in that language. If a language is not supported for Knowledge, articles in that language will not be used by your AI Agent.
Language detection indicates whether Ada can automatically detect and switch to conversing in this language. Some languages are not supported by the language detection model.
Translation indicates the service that Ada uses to translate inquiries and knowledge content. This happens when your AI Agent does not have, or cannot use, knowledge content in the end user’s language. In such cases, the AI Agent falls back to English articles, and responses are translated into the end user’s language. Depending on your configuration, your AI Agent may use either Google Translate or an LLM to generate the translated reply.

For supported right-to-left languages, web chat is tailored for the best end user experience. This includes flipped UI components such as settings, loading bar, and more.

Language	Web chat	Email	Voice	Knowledge	Language detection	Translation
Afrikaans	Yes	No	No	No	No	Google Translate
Albanian	Yes	Yes	No	No	No	Google Translate
Arabic	Yes	Yes	No	Yes	Yes	LLM generation
Belarusian	Yes	Yes	No	No	Yes	Google Translate
Bosnian	Yes	No	No	No	No	Google Translate
Bulgarian	Yes	Yes	No	No	Yes	Google Translate
Burmese	Yes	Yes	No	No	Yes	Google Translate
Catalan	Yes	No	No	No	No	Google Translate
Chichewa (Nyanja)	Yes	No	No	No	No	Google Translate
Chinese (Simplified)	Yes	Yes	No	Yes	Yes	LLM generation
Chinese (Traditional)	Yes	Yes	No	No	Yes	Google Translate
Croatian	Yes	No	No	No	Yes	Google Translate
Czech	Yes	Yes	No	No	Yes	Google Translate
Danish	Yes	Yes	No	No	Yes	Google Translate
Dutch	Yes	Yes	Yes	Yes	Yes	LLM generation
English	Yes	Yes	Yes	Yes	Yes	LLM generation
Estonian	Yes	Yes	No	No	Yes	Google Translate
Filipino (Tagalog)	Yes	No	No	No	Yes	Google Translate
Finnish	Yes	Yes	No	No	Yes	Google Translate
French	Yes	Yes	Yes	Yes	Yes	LLM generation
German	Yes	Yes	Yes	Yes	Yes	LLM generation
Greek	Yes	Yes	No	No	Yes	Google Translate
Haitian Creole	Yes	No	No	No	Yes	Google Translate
Hebrew	Yes	Yes	No	No	Yes	Google Translate
Hindi	Yes	Yes	No	No	Yes	Google Translate
Hungarian	Yes	Yes	No	No	Yes	Google Translate
Icelandic	Yes	Yes	No	No	Yes	Google Translate
Indonesian	Yes	Yes	No	No	Yes	Google Translate
Italian	Yes	Yes	Yes	Yes	Yes	LLM generation
Japanese	Yes	Yes	No	No	Yes	Google Translate
Kazakh	Yes	Yes	No	No	No	Google Translate
Khmer	Yes	Yes	No	No	Yes	Google Translate
Kinyarwanda	Yes	Yes	No	No	No	Google Translate
Korean	Yes	Yes	No	No	Yes	Google Translate
Latvian	Yes	Yes	No	No	Yes	Google Translate
Lithuanian	Yes	Yes	No	No	Yes	Google Translate
Macedonian	Yes	Yes	No	No	No	Google Translate
Malay	Yes	No	No	No	Yes	Google Translate
Ndebele (South)	Yes	No	No	No	No	Google Translate
Norwegian	Yes	No	No	No	Yes	Google Translate
Polish	Yes	Yes	No	No	Yes	Google Translate
Portuguese	Yes	Yes	No	Yes	Yes	LLM generation
Punjabi	Yes	Yes	No	No	Yes	Google Translate
Romanian	Yes	Yes	No	No	Yes	Google Translate
Russian	Yes	Yes	No	No	Yes	Google Translate
Serbian	Yes	Yes	No	No	No	Google Translate
Sesotho	Yes	No	No	No	No	Google Translate
Shona	Yes	No	No	No	No	Google Translate
Slovak	Yes	Yes	No	No	Yes	Google Translate
Slovenian	Yes	No	No	No	Yes	Google Translate
Spanish	Yes	Yes	Yes	Yes	Yes	LLM generation
Swahili	Yes	Yes	No	No	No	Google Translate
Swedish	Yes	Yes	Yes	No	Yes	LLM generation
Tamil	Yes	Yes	No	No	Yes	Google Translate
Thai	Yes	Yes	No	No	Yes	Google Translate
Turkish	Yes	Yes	No	No	Yes	Google Translate
Ukrainian	Yes	Yes	No	No	Yes	Google Translate
Vietnamese	Yes	Yes	No	No	Yes	Google Translate
Xhosa	Yes	Yes	No	No	No	Google Translate
Zulu	Yes	Yes	No	No	No	Google Translate

Multilingual behavior

Your AI Agent can understand end user questions in any of Ada’s supported languages, regardless of whether they are enabled in your AI Agent. However, it only responds in the languages that are enabled in the multilingual settings.

Default language

English is the default support language in all Ada AI Agents. You cannot disable it—your AI Agent always replies to English-speaking end users.

English is also the fallback language that your AI Agent uses if it cannot determine what language to use with an end user or if it cannot find Knowledge articles in the end user’s spoken language.

How multilingual knowledge works

Your AI Agent can always use English Knowledge articles to respond to end user inquiries, because English is Ada’s default language. It can also use Knowledge articles in some non-English languages:

English
Arabic
Chinese
Dutch
French
German
Italian
Portuguese
Spanish

Contact your Ada representative to enable using non-English articles in Ada.

For non-English Knowledge articles, your AI Agent does the following:

Determines the language the end user is writing in and identifies how best to respond using Ada’s Reasoning Engine.
If your knowledge base contains ingested content in the same language as the end user’s inquiry, the generated response only uses this matched language content.

This means that when multilingual knowledge content is ingested, non-English inquiries are limited to only access corresponding non-English articles. This overrides any Coaching to use an English article instead for non-English inquiries.
If your knowledge base does not contain content in the same language as the end user’s inquiry, the inquiry must be translated using either Google Translate or an LLM before the Reasoning Engine can generate a response.
If the language is not an enabled language, the AI Agent replies in Ada’s default language, English.

Knowledge localization

Ada supports ingesting localized articles in languages supported for knowledge. This ensures that you can serve region-specific content to your end users. For example, if your knowledge base contains different versions of English articles for end users in different regions, like American users (en-US) versus Canadian users (en-CA), Ada ingests region-specific articles from all of these locales.

Ada serves localized content to your end users based on the language that they speak. For end users speaking en, your AI Agent uses en and all localized en articles (e.g. en-CA, en-US, etc.) when serving them.

When Ada automatically switches languages, it always selects the non-localized variant (e.g., switching to French sets the end user’s language to fr, not fr-CA).

To ensure that localized content goes to the right end users, set availability rules on your localized articles.

To set availability rules for localized articles:

On the Ada dashboard, navigate to Config > AI AGENT > Knowledge > Articles.
At the top of the articles table, click Filter and select Language as the filter option.
Select the locale that you would like to set availability rules for, then click Apply.
At the top-left of the articles table, click on the check box to bulk-select all of the articles you can see in the list.
- You may need to scroll to the bottom and click through each “page” of your table view to continue selecting all available articles on all pages (previously selected articles will stay selected when you click to a new page).
After selecting all articles under this locale, click Set availability in the bar at the bottom of your screen.
Next, set rules so that these articles are only available to end users in this region.

A common setup is to use initialurl to do this. For example, set a rule like initialurl contains '.ca' to ensure en-CA articles only show up to Canadian end users chatting with the AI Agent on your .ca website.
Lastly, click Save to apply these rules.

Now, your AI Agent will always use these localized articles with the right users, even if their language field isn’t localized (i.e. no region).

Translation

Sometimes, your AI Agent must translate end user inquiries and Knowledge content before it can respond. This happens when the end user is speaking a non-English language, but Ada cannot use Knowledge content in a matching language.

In these situations, Ada references your English Knowledge content and then uses either Google Translate or an LLM to draft a reply in the end user’s language:

For languages supported by Google Translate: End user inquiries are translated into English before Ada’s Reasoning Engine generates a response. The response is then translated into the end user’s language by Google Translate before being sent.
For languages supported by LLM generation: End user inquiries remain in the original language, but Ada’s Reasoning Engine translates English Knowledge content into the end user’s spoken language when generating a reply. This approach enables better understanding of inquiries and higher-quality responses.

Channel-specific behavior

Your AI Agent can respond to end users in many languages in Web Chat, Email, and Voice, but there are differences in how each channel handles language:

In Web Chat, conversations can happen in any of Ada’s supported languages. You can configure your AI Agent to start the conversation in a specific language. End users can change the language using the web chat widget settings or by speaking in a different language.
In Email, conversations always start in English. End users can switch to another language by replying in that language, but there are limitations for switching to some languages.
In Voice, conversations can happen in a limited subset of Ada’s supported languages. You can let your end users choose the language they would like to use at the start of a conversation. You can also configure your AI Agent to start the conversation in a supported language by:
- Assigning a default starting language to a phone number or SIP address used to receive a phone call
- Passing the language identifier in the SIP header if the AI Agent is using SIP to receive calls
- Passing the language identifier value to the end user profile API

Multilingual support setup: Configure languages for your AI Agent.
Language switching: Understand how end users can switch languages during a conversation.
Manual translations: Customize translations in Greetings and Handoffs.
Knowledge: Manage your AI Agent’s knowledge base.