Using language detection and dynamic machine translation in Virtual Agent NLU and LLM standard chat conversations

  • Release version: Yokohama
  • Updated March 6, 2025
  • 9 minutes to read
  • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Using language detection and dynamic machine translation in Virtual Agent NLU and LLM standard chat conversations

    This feature enhances the Virtual Agent chat experience by combining language detection with dynamic machine translation, allowing users to interact in their preferred or detected languages. It supports fallback languages for locales and translates languages not professionally localized or unsupported in topic discovery, thereby accommodating diverse users and improving accessibility.

    Show full answer Show less

    Key Features

    • Language Detection: Detects the user’s language on the first utterance and can prompt users to switch the conversational language or switch automatically based on configuration.
    • Dynamic Machine Translation: Translates conversations dynamically when professional translations are unavailable, storing translations temporarily to reduce calls to external services.
    • Fallback Language Configuration: Allows designation of fallback languages for unsupported locale-specific codes, ensuring consistent user experience (e.g., Spanish as fallback for Mexican Spanish).
    • Integration with NLU and LLM Conversations: Enables seamless language detection and translation in both Natural Language Understanding (NLU) and Large Language Model (LLM) standard chat conversations.
    • Support for Guest Users: Language detection is limited for guest users unless resumable guest sessions are enabled; dynamic translation helps in these scenarios.
    • Selective Use of Dynamic Translation: Can be disabled for professionally localized languages to prioritize expert translations.
    • Translation Storage and Refresh: Translated text is stored in a dedicated table and cleared every 24 hours to ensure updated translations from platform resources.
    • Channel Support: Available on ServiceNow chat widget, Microsoft Teams, Slack, Workplace from Facebook, and Facebook Messenger; not supported on mobile client, WhatsApp, LINE, Twilio SMS, Alexa, Google Assistant, or Siri.
    • Agent Chat Considerations: If dynamic translation is off for Agent Chat but enabled for Virtual Agent, conversation language mismatches may occur; enabling dynamic translation for Agent Chat resolves this.

    Practical Usage and Configuration

    • Configure language detection and dynamic translation settings in the ServiceNow instance to support multiple languages and fallback scenarios.
    • Use dynamic translation during localization efforts to cover untranslated content, then disable it once professional translations are complete for quality assurance.
    • Language detection applies only to the initial user input; language switching mid-conversation is not supported.
    • Users are prompted to switch languages when a new language is detected, with options to accept or decline, influencing conversation flow.
    • Default system language and user profile language preferences influence the starting conversational language.
    • Exclusion rules in dynamic translation can protect specific terms from being translated.

    Expected Outcomes

    • Improved user engagement by allowing conversations in users’ preferred or detected languages.
    • Smoother handling of multilingual conversations including those in unsupported or partially localized languages.
    • Reduced reliance on professional translations for all content while maintaining quality where expert translations exist.
    • More efficient translation service usage by caching translated text temporarily.
    • Consistent language experience across chat sessions, with language preferences preserved appropriately based on user actions.

    Enable a combination of language detection and machine translation for Virtual Agent to improve the chat experience for diverse users. You can designate fallback languages for locales or dynamically translate languages that are not professionally localized or are not supported in topic discovery.

    Chat users are more comfortable having conversations in their preferred language. Their preferred language is set in the user's profile language preference. For example, Virtual Agent greets users in the preferred language specified in their user profile. But if a user responds in another language, Virtual Agent can detect the change. Depending on your user experience, users either receive a prompt regarding whether they'd like to switch the conversational language to the detected language or the conversational language updates automatically to the detected language. After the language is switched to the detected language, that language becomes the conversation's language for the remainder of the conversation. But in many cases users don't have control over their preferred language, particularly if they are guest users.

    Guest conversations do not support language detection if resumable guest sessions are not enabled. This is why language detection and dynamic machine translation can improve the chat experience, particularly if you have conversations that have not been localized in some languages. For example, your organization may routinely localize into Spanish and French, but you may also want to support Dutch and Danish. You can enable Dynamic Translation on your ServiceNow® instance for those languages.

    You can use language detection to support custom or locale-specific language codes as well. For example, NLU topic discovery doesn't support Mexican Spanish (es-mx), but it does support Spanish (es). You can designate Spanish as the fallback language for Mexican Spanish, ensuring that these users have a good customer experience.

    Virtual Agent conversations can be dynamically translated. Virtual Agent uses platform translations and if no platform translations are found, then Dynamic Translation is used instead. If some conversations have been professionally translated, you should disable dynamic translation for those specific languages. Expert translations of your NLU models, keywords, and topics may be more accurate and of higher quality than machine translations. Prior to making a dynamic translation call, the Text Translations for Dynamic Translation [sys_cs_dynamic_translation_text] and Message [sys_ui_message] tables are checked to see if the text has been previously translated or already has a localization record. A dynamic translation call is only made if no translation exists in either table. After a dynamic translation call is made and that text is translated, that translated text is stored in the Text Translations for Dynamic Translation [sys_cs_dynamic_translation_text] table for future use. This table is cleared every 24 hours to ensure that the most up-to-date translations stored in the Message [sys_ui_message] table are being referenced. Storing previously translated text results in fewer dynamic translation calls to third-party translation services. For more information about localizing Virtual Agent conversations, see Localizing Virtual Agent conversations.

    You can use Dynamic Translation during the localization process as well. If you are localizing conversations with professional linguists, you can enable Dynamic Translation while some topics and strings remain untranslated. When your deployment in that language is complete, you can disable Dynamic Translation, since professional translations are likely more accurate.

    Video link Virtual Agent Academy: Dynamically translate languages in Virtual Agent Watch this video for a demonstration of using Virtual Agent with Dynamic Translation.

    Note:
    You can use language detection independently of Dynamic Translation.

    Language detection and Dynamic Translation process in Virtual Agent NLU and LLM standard chat conversations

    When you enable language detection and Dynamic Translation for Virtual Agent for other languages, a variety of diverse users gain access to your Virtual Agent chat environment.

    Language detection for NLU and LLM conversations are enabled in the same location. For more information on enabling language detection, see Enable Dynamic Translation for languages in Virtual Agent. Dynamic Translation for NLU and LLM conversations are enabled in different locations. For more information on enabling dynamic translation, see Enable Dynamic Translation for languages in Virtual Agent and Configure multilingual service for Now Assist applications.

    Language detection is only applicable to the first utterance entered into the conversation. Conversational language switching cannot occur mid-conversation. Language detection prompts users with a question regarding whether they'd like to switch the conversational language during Virtual Agent NLU or LLM standard chat conversations.

    When language detection is turned off, no conversational language switching can occur.

    The default system language sets all users’ default session language. That user's default session language is what sets the conversational language. It is also the preferred language in your NLU model, if NLU is enabled. When a user types an utterance that is not recognized as the system language, the results depend upon your configuration for that language.

    The following examples show how a conversational language can be impacted depending upon the language detection and dynamic translation settings:

    Note:
    If you create exclusion rules for Dynamic Translation, those excluded terms are honored during Dynamic Translation. For more information about the Exclusion Framework, see Exclusion Framework in Dynamic Translation.
    Table 1. Language detection and Dynamic Translation configuration settings and results
    Language configuration Results in Virtual Agent NLU or LLM standard chat conversations
    Language detection and Dynamic Translation are not configured. For unlocalized conversations, the language the topics appear in is dependent on the language they were authored in. Typically, the default system language is used for topics and topic discovery.
    Language detection is enabled, but not Dynamic Translation

    Language responses can show up differently depending on whether localization was done or not for those topics. The conversation language may continue in the detected language based on whether localization records for those topics were previously included. If localization records are not provided, the conversation language would continue in the preferred conversational language.

    If the detected language is a locale or other language code that has been configured with a fallback language, Virtual Agent asks the user if they want to switch the conversation to the fallback language.

    For example, if the detected language is Mexican Spanish (mx-es), and Spanish (es) has been designated as the fallback language, Virtual Agent asks: I have detected you are typing in Spanish. Would you like to switch the conversation to Spanish? (Yes/No)

    If the user selects Yes, the fallback language is used for topic discovery and topic display. This continues until the conversation ends.

    If the user selects No, Virtual Agent will not prompt them again during the conversation.

    Language detection and Dynamic Translation are enabled

    If the detected language is enabled for dynamic translation in Virtual Agent Chat Settings, Virtual Agent asks the user if they want to switch the conversation to the detected language.

    For example, the preferred conversational language is in English but the user types an utterance in Spanish. The user receives a prompt asking whether they'd like to switch the conversation to Spanish. The prompt reads, I have detected you are typing in Spanish. Would you like to switch the conversation to Spanish? (Yes/No)

    If the user selects No, Virtual Agent will not prompt them again during the conversation and the conversation continues in English.

    If the user selects Yes, the utterance is translated to the default language. The default language for NLU conversations is within the glide.sys.language system property. The default language for LLM conversations is within the com.glide.cs.llm.default.Language system property. All topic matches are translated dynamically into the user's conversation language and displayed. In this example, the conversation will now continue in Spanish and Virtual Agent expects the user to continue chatting in Spanish. After the conversation ends and the user restarts a new conversation by reloading their browser or initially navigating to a portal with no conversation in progress, the prior conversational language (Spanish in this example) continues.

    Figure 1. Example of a Virtual Agent LLM standard chat conversation using language detection
    Virtual Agent detects an utterance in Spanish and asks the user if they want to switch, in both English and Spanish.
    The language detection and dynamic machine translation process operates according to the following principles:
    • Language detection occurs only once per conversation, at the initial utterance.

      If a user tries to change the language in the middle of a conversation, Virtual Agent returns an I didn't understand message.

    • When the conversation has closed and restarted, the language used in the next conversation varies depending on how the user restarted the conversation. If the user restarted the conversation by reloading their browser, the preferred language is restored in the next conversation. If the user restarted the conversation by using the chat widget's restart feature, the previous conversational language is restored in the next conversation. Regardless of how a conversation has closed and restarted for NLU or LLM standard chat conversations, if a user did not accept to switch languages after a different language was detected in the previous conversation, the preferred language is restored in the next conversation.
    • If a user's detected language is the same as the preferred language, the user is not asked to switch languages.
    • If a user's detected language has no fallback language specified and dynamic translation is not enabled for it, the conversation continues in the preferred language.
    • If a user's detected language is not activated on the instance, the conversation continues in the preferred language. For more information, see Activate a language.
    Note:
    AI Search results are not translated.

    Dynamic Translation and Agent Chat

    If Dynamic Translation is enabled on Virtual Agent but turned off for Agent Chat, the live agent will see what the user sees. If the user started the conversation in another language and the conversational language switches, then the live agent will see the conversation in the user's detected language. This may cause some confusion for both the agent and the customer. To resolve this issue, consider turning on Dynamic Translation for Agent Chat as well. For details, see Configuring Dynamic Translation for Agent Chat.

    Supported languages

    For better results, install the ServiceNow language plugins for each language that uses dynamic translation. For a detailed list of language plugins, see Activate a language.

    You can enable dynamic translation for languages that are not currently supported on the ServiceNow platform. For details, see Enable Dynamic Translation for an unsupported language.

    Supported channels

    Dynamic Translation for Virtual Agent is currently supported in the following channels:
    • ServiceNow chat widget
    • Microsoft Teams
    • Slack
    • Workplace from Facebook
    • Facebook Messenger
    Note:
    The Virtual Agent topic test window may identify user utterances from Microsoft Teams or other adapters as sysparm_search_text. Dynamic Translation will translate this content as well.
    The following channels do not support Dynamic Translation for Virtual Agent:
    • ServiceNow mobile client
    • WhatsApp
    • LINE
    • Twilio SMS
    • Alexa
    • Google Assistant
    • Siri