Multilingual model management
Summarize
Summary of Multilingual model management
Multilingual model management enables the use of Natural Language Understanding (NLU) models to comprehend user input in multiple languages. The NLU Workbench helps maintain a consistent structure across these languages, ensuring a unified user experience.
Show less
Key Features
- Primary and Secondary Languages: Primary models serve as the source language and can be translated into various secondary languages. Each primary model can have multiple secondary models, but they must be distinct in language.
- Language Grouping: Grouping languages is optional but aids in organizing models. You can add languages to primary models, automatically creating secondary models with translation options.
- Model Interactions: All intents and entities in secondary models mirror those in the primary model, ensuring consistency. New intents can only be created in primary models, and updates are reflected in secondary models.
- Model Duplication: Users with the nluadmin role can duplicate models, allowing for the creation of new model groups while retaining the original model structure.
- Intent Management: Enable or disable intents in your models to control their activity status, assisting in content editing and updating.
Key Outcomes
By leveraging multilingual model management, ServiceNow customers can effectively manage NLU models across languages, ensuring accurate translation and consistent user interaction. This capability allows for streamlined updates and maintenance of multilingual content, enhancing overall system performance and user satisfaction. The ability to assign editors for model maintenance further optimizes the model management process, ensuring high-quality translations and content accuracy.
Use multilingual Natural Language Understanding (NLU) models for the system to understand user input in several languages. The NLU Workbench helps you manage and maintain a consistent structure for content across languages to provide a unified experience.
Primary and secondary languages
A primary language is the source language you choose when creating a model. These models are considered primary models. Primary models can then be translated into different languages. Those translated models are referred to as secondary models. The languages in which they are translated are referred to as secondary languages.
The NLU Workbench home displays primary and secondary language models nested under the model name. Select the arrow to the left of the model name to expand the language group.
- Primary models have a language you assign to them during model creation and listed as English (Primary). The language of the primary model is the source language for the translations that follow later in the secondary models.
- Secondary models are translated copies of the primary model. Each secondary model uses a different language, such as Brazilian Portuguese, Polish, or Finnish.
- Any supported language can be the primary language for a primary model or the secondary language for a secondary model.
- Within a model group, you can't have two secondary models that use the same language.
For more information on the languages available in NLU, see NLU language support.
Implementing language grouping
- You can add a language to a primary model in the model's More options menu, which automatically creates a secondary language model with translation options. For more information on adding and translating a language, see Translate a multilingual model.
- You can access the instance model migration page directly by visiting <instance_name>.service-now.com/$nlu-studio.do#/model-migration. For more information on language grouping using this method, see Model language grouping.
Primary and secondary model interactions
When you add an intent to a primary model, the intent is added to all of its secondary models. Every intent in a secondary model is mapped to its corresponding intent in the primary model. The mapping of intents ensures that any application that uses these intents can access all the secondary intents through their corresponding primary intents.
When you delete an intent or entity in a primary model, its corresponding intents and entities are also deleted in its secondary models. Therefore, the secondary models must always follow the status of the primary model content. Although you can't delete intents in secondary models, you can disable them.
For more information on intent interactions, see Import primary model content to a secondary model.
You can't delete an entity created in a secondary model if it's a copy of a corresponding entity created in the primary model. However, you can add or delete an entity in a secondary model if it doesn't have a corresponding entity in the primary model.
Duplicating grouped models and model groups
Using the nlu_admin role, you can duplicate primary models, secondary models, and entire model groups. You can duplicate just the primary model or a set of secondary models from among the model group.
- If you duplicate a secondary model that's in a model group, the duplicated version becomes a separate primary model that is outside that model group.
- All intents are duplicated. The duplicated intent maintains the same Enabled status as the original intent.
- When duplicating a primary model, you can duplicate a set of secondary models, or all of the secondary models along with it. This action creates a model group comprised of duplicated versions with the respective original models marked as the source models.
- When duplicating a model group, you can choose an existing secondary model to be the primary model for the duplicated group.
- If you select a secondary model in an existing model group as a new primary model while duplicating the group, all the disabled entities are enabled for the duplicated version of the secondary model. The secondary model becomes the primary model in the new duplicated model group.
- If you duplicate a primary model without any secondary model, the duplicated version becomes a separate primary model.
For more information, see Duplicate an NLU model.