Multilingual service for Now Assist

  • Release version: Australia
  • Updated January 30, 2025
  • 5 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 Multilingual Service for Now Assist

    Now Assist applications leverage large language model (LLM) multilingual capabilities to translate user-generated content and communicate with users in their preferred languages. Language detection is based on the user's Language & Region preferences or the text content. The multilingual service supports multiple LLM providers, including Now LLM Service, Microsoft Azure OpenAI, Anthropic Claude on AWS, and Google Gemini, with Azure OpenAI offering the most extensive language support on the ServiceNow AI Platform.

    Show full answer Show less

    How Multilingual Services Work

    Two main translation methods are used:

    • Native Translation: The LLM processes text in the user’s language directly, responding in the same language without converting to English. This method prioritizes supported languages and delivers higher quality and lower latency.
    • Dynamic Translation: Non-English input is first translated to English, processed by the LLM, then translated back to the user's preferred language. This method is used for unsupported languages or as a fallback.

    If both methods are enabled, native translation takes precedence, improving performance and reducing latency.

    Supported Languages and Models

    Languages are categorized into three groups based on native support quality:

    • P1: Strong native support with high-quality translation (e.g., English, French, German, Italian, Spanish, Brazilian Portuguese).
    • P2: Good support with occasional gaps in nuance (e.g., Canadian French, Japanese, Dutch).
    • P3: Broader language coverage with less training depth, potentially affecting idiomatic accuracy (e.g., Swedish, Finnish, Hebrew, Korean, Russian, Chinese, Arabic).

    Support varies by model, with ServiceNow’s own large models supporting mostly P1 and P2 languages, while providers like Microsoft Azure OpenAI cover over 80 languages including P3.

    Adding Additional Languages

    You can add languages beyond the predefined P1, P2, and P3 groups by ensuring the language is supported by your LLM provider and present in the ServiceNow syslanguage table with the Active flag enabled. The language must then be enabled in the generative AI model configuration (sysgenerativeaimodelconfig table) and activated in the Now Assist Admin settings under the Multilingual service tab. After adding a language, reindexing all sources in that language is necessary, and if semantic search issues arise, clearing the embedding model cache may be required.

    Dynamic Translation Details

    Dynamic Translation uses Microsoft Azure OEM translation service through ServiceNow and supports all languages available as language packs on the ServiceNow platform, including Swedish, Finnish, Czech, Hebrew, and others.

    Choosing and Enabling Translation Services

    Choosing between native translation and Dynamic Translation depends on factors like latency and translation quality. Native translation is included with Pro Plus and Enterprise licenses at no additional cost. You can manage and monitor usage and enable multilingual services for Now Assist applications through the platform settings.

    Now Assist applications use the multilingual capabilities of large language models (LLMs) to translate user-generated content.

    Multilingual service overview

    You can configure the multilingual services for Now Assist applications to communicate with your users in their preferred languages. The LLM detects the language used based on the user's language choice in their Language & Region preferences or the contents of the text, such as a conversation message.

    Multilingual support is enabled for multiple LLM providers, such as Now LLM Service, Microsoft Azure OpenAI, Anthropic Claude on AWS, and Google Gemini. Azure OpenAI offers the most language support on the ServiceNow AI Platform.

    How the different multilingual services work

    The inherent multi-language capabilities of the LLM is called native translation. In native translation, text is sent as-is to the model and the model responds in the user's language. For example, if a user sends a message in Spanish, the multilingual LLM supports Spanish, so it receives the Spanish input directly from the user. It then generates a response in Spanish, regardless of the language used in the source material it uses to create that response.

    For Dynamic Translation, once non-English content is detected, the translation service translates the content to English before sending the request to the large language model (LLM). After the LLM returns a response, the response is translated back into the user's preferred language for them to see. For example, a user enters a message in Slovenian. Dynamic Translation translates the Slovenian to English before giving the message to the LLM. The LLM generates a response in English, and then Dynamic Translation translates that response to Slovenian to show to the user.

    If you enable both native translation and Dynamic Translation, native translation has preference and is used first.

    If you have native translation enabled, the LLM checks the user's language to see if it is supported. If the language is supported, only native translation is used and no Dynamic Translation occurs. A small number of languages are not yet natively supported by all models. Now Assist Dynamic Translation is automatically applied to address this for unsupported languages. Additionally, if you have native translation enabled, then LLM requests that return outputs in the fallback language will not be translated by Dynamic Translation. Otherwise, they’ll be returned as-is to the user. This process helps increase skill performance and decrease latency.
    Note:
    If you want to add a language that is not part of the regularly configured languages, you can do so. See the section below, Adding languages beyond the P1, P2, and P3 language groups.

    Native translation

    Supported languages are categorized into three basic groups.

    P1 languages have the strongest native support across most model providers, meaning high-quality translation and generative capabilities with fewer errors. Nuance, idioms, and context are usually handled well because these languages are deeply integrated into the models.

    P2 languages are supported natively by smaller ServiceNow models and most external providers, but not always by every model (e.g., some exceptions for Canadian French and Dutch). Accuracy is good, but there may be occasional gaps in nuanced understanding compared to P1 languages.

    P3 languages include a broader set of languages supported by major providers like Microsoft Azure OpenAI, Google Gemini, and Anthropic Claude on AWS, but not natively by ServiceNow’s own large models. While coverage is comprehensive, these languages generally have less training depth, so subtle tone or idiomatic expressions may be harder to capture perfectly.

    • P1
      • English
      • French
      • German
      • Italian
      • Spanish
      • Brazilian Portuguese
    • P2
      • Canadian French
      • Japanese
      • Dutch
    • P3
      • Swedish
      • Finnish
      • Czech
      • Hebrew
      • Hungarian
      • Korean
      • Norwegian
      • Polish
      • Portuguese
      • Russian
      • Simplified Chinese
      • Traditional Chinese
      • Thai
      • Turkish
      • Arabic
      • Danish

    Specific models vary in their support.

    • ServiceNow Large Language Model v1 (Llama 3.3 70B) only supports English natively.
    • ServiceNow Large Language Model v2 (GPT-OSS) supports English, French, German, Spanish, and Italian natively.
    • ServiceNow Small Language Model v1 (Mistral Nemo 12B) and v2 (Apriel) supports all P1 and P2 languages.
    • Microsoft Azure OpenAI GPT-4.1 and GPT-4.1-mini support more than 80 languages, including P1, P2, and P3.
    • Google Gemini 2.5 Flash/2.5 Pro currently does not support French Canadian natively.
    • Anthropic Claude 4.0 Sonnet currently does not support French Canadian or Dutch natively.

    Adding languages beyond the P1, P2, or P3 language groups

    Some providers, such as Azure OpenAI, offer additional language support beyond the P1, P2, and P3 language groups previously described. You can configure your instance to include additional languages.

    Note:
    Before adding a language that is not in the P1, P2, or P3 groups, you should confirm whether the language is supported by the provider. If you add a language that isn't supported, you may get unexpected results. The LLM could fail to detect the language, and/or it could not be able to translate effectively.
    Ensure the language is on the sys_language table.
    Create a record for the language if necessary. The record should include the language name and code. The Active field should be set to true.
    Enable the language for the Now Assist model.
    • Go to the sys_generative_ai_model_config table and open the record for the model you want to add the language to, such as gpt_small by Azure OpenAI.
      Note:
      The model for AI Search is named E5FT. If you want to make changes to the language support for AI Search, you can configure that record. After, you need to reindex all sources in the new language.
    • In the Supported languages list, add the language and save the record.
    Activate the language for Now Assist applications.
    • Go to All > Now Assist Admin > Settings and open the Multilingual service tab.
    • Scroll to the provider you edited, then select the edit button.
    • Select the language you added and save.
      Note:
      If you don't see your language as an option, ensure that the language is on the sys_language table and Active is set to true.
    Note:
    If semantic search is not working even after installing a language and enabling it for AI Search, you may need to clear the embedding model cache. To do so, run the following: GlideCacheManager.flush("ais_semantic_embedding_model_cache");.

    Dynamic Translation

    Dynamic Translation for Now Assist uses the Microsoft Azure OEM translation service through ServiceNow. Every language available as a language pack on the ServiceNow platform is supported by Microsoft Azure OEM. These languages are:
    • Swedish
    • Finnish
    • Czech
    • Hebrew
    • Hungarian
    • Korean
    • Norwegian
    • Polish
    • Portuguese
    • Russian
    • Simplified Chinese
    • Traditional Chinese
    • Thai
    • Turkish
    • Arabic
    • Danish
    For more information, see Microsoft Azure OEM for Dynamic Translation in Now Assist.

    Choosing a translation service

    There are different factors that can go into your decision to use one translation service over the other, such as latency and quality of translation. Your entitlements determine what is available to you, but there are no additional costs associated with native translation service.
    Note:
    The native translation service is included in Pro Plus and Enterprise licenses, by default.
    For more information on tracking Now Assist usage, see Monitoring Now Assist usage.

    Enabling translation for Now Assist

    For more information on turning on multilingual services for Now Assist applications, see Enable translation for Now Assist applications.