You can give your agents the power to automatically translate ticket conversations if they so choose. This allows agents to communicate with end users and other agents, even if they are using different languages. This article describes how conversation translation works. To activate conversation translations for your agents, seeTurning on and off AI translations for ticket conversations.

What's my plan?
All Suites Team, Growth, Professional, Enterprise, or Enterprise Plus
Support Team, Professional, or Enterprise

You can give your agents the power to automatically translate ticket conversations if they so choose. This allows agents to communicate with end users and other agents, even if they are using different languages. This article describes how conversation translation works. To activate conversation translations for your agents, see Turning on and off AI translations for ticket conversations.

For information about using conversation translation after it is activated, see Translating conversations in the Zendesk Agent Workspace.

This article includes the following topics:

  • About conversation translation
  • How conversations are translated
  • The end-user experience
  • Translation limitations

About conversation translation

Conversation translation gives agents the option to automatically translate inbound messages (coming from the end user) and outbound messages (going to the end user). It works in conversations from all Zendesk channels, including:

  • Zendesk messaging
  • Live chat
  • Social channels
  • Sunshine Conversations channels
  • Email, web form, or API channels

If the agent’s and end user’s languages differ, a translation banner is displayed informing the agent of the language difference, and they can choose to enable translation for that conversation.

The following video gives you an overview of how conversation translation works:

How conversations are translated

An end user’s language is determined based on recent comments and replies they have sent, then our translation services are used to translate text. The length and number of the comments and replies can affect the language analysis. There's no guaranteed length at which the language will always be detected, but longer messages produce better results.

For agents, conversation translation works based on the language set in the agent’s Support user profile. For the most accurate results, the language of the text the agent types should match that profile language. For example, if an agent’s profile language is set to English, the conversation should be written in English before it is translated into the end user's language. If the agent enters text in a different language than the one configured in their profile, translations may be inaccurate or behave unexpectedly.

If the text scan detects that the agent’s and end user’s languages are different, the translation banner is displayed. If the text scan can't detect the language (for example, the message was too short to determine the language), the banner doesn’t appear.

If you don't see the translation banner, click the ticket options menu, then click Translate.

Both the original and translated messages sent by the agent appear in the Event log. End-user comments and replies appear only in the original language.

The end-user experience

When an agent turns on automatic translation for a conversation, the end user does not receive an activation notification—agent messages simply appear in their detected language. However, translated agent messages are identified in the conversation. In messaging conversations, AWS translated messages are identified by the word Translated in their detected language.

For AI translations, the message says Translated by AI in their native language, if the disclosure setting is activated by an admin.

Translation limitations

Translation limitations include:

  • SunCo Shorthands are not supported for translations.
  • Translations for third-party widgets are not supported.
  • Our translation service cannot accurately translate texts written in a phonetic manner. For example, translation will only work for Hindi text in Devanagari script but won't work when it is written in Roman script.
  • If the translation fails when an agent sends an outgoing email, the email will still be sent in the original language after five minutes.
Powered by Zendesk