Multilingual agents

Dialogflow supports many languages. For the complete list, see the languages reference page. The language you choose when creating an agent is set as the default language, and you can add additional languages.

Root and locale-specific languages

There are two categories of languages:

• Root languages: These are languages like English (en) that do not specify a locale.
• Locale-specific languages: These are languages like English-US (en-US) that do specify a locale as a specific region or country.

Some languages act as both root and locale-specific. There are significant differences between locales for these languages, so the locales cannot share a common root language. Examples:

• Chinese-Cantonese (zh-HK)
• Chinese-Simplified (zh-CN)
• Chinese-Traditional (zh-TW)
• Portuguese-Brazil (pt-BR)
• Portuguese-Portugal (pt)

You should primarily design your agent for root languages, and only customize for locale-specific languages as needed.

Add a language

To add a language or locale with the Dialogflow CX console Console:

1. Open the Dialogflow CX console.
2. Choose the Google Cloud project for the agent.
3. Find the agent in the list.
4. Click the agent display name.
5. Click Agent Settings.
6. Click the Languages tab.
7. Add or remove languages as needed.
8. Click Save.

Language-specific data

Most agent data is common for all languages of a multilingual agent. However, text used to interact with an end-user is language-specific. When building a multilingual agent, you must supply this language-specific data for each language. When you select a language from the console or provide a language to the API, you can access this language-specific data for the given language. The following is a complete list of language-specific data:

• intent training phrases
• fulfillment responses
• entity entries

To access language-specific data with the console, click the language button at the top of the console.

For example, the size entity type below is used for ordering t-shirts. The agent was originally configured in English, so the entity has entries like "S" with the synonym "small". When Spanish is added as a language, the entity is created by Dialogflowfor the new language, but you need to supply entity entries in Spanish.

AI generate language-specific data

Conversational Agents (Dialogflow CX) gives you the option to speed up the process of adding language-specific data by using AI generation to provide the data for you, instead of manually entering it by hand. This can be done at the intent, entity, and fulfillment level as a Preview feature.

If you opt not to use AI generation, you can still manually enter the data for non-default languages.

(Preview) Intent AI generation { :#ai-intent}

Auto-generate training phrases in a non-default language:

1. Go to Agent settings > Languages and make sure you have one or more non-default languages enabled for your agent.
2. Navigate to the language drop-down menu at the top of the Console and select your chosen non-default language.
3. Navigate to the Manage tab, then click Intents and choose one of your intents. Make sure that the Description field is populated with a description of the intent's purpose. The description can be in your default language or in the language you are viewing.
4. When you enable a non-default language, Conversational Agents (Dialogflow CX) does not translate any existing training phrases into the new language. To create new training phrases in the new language, click the __ newly LLM-generated phrases radial button. The default value is 3. To copy and automatically translate your existing training phrases, select All training phrases from your Default Language translated by AI generation.
5. Click + Add. Your specified number of training phrases are generated and populated into the table in the language you've selected. Make sure to review all generated training phrases for accuracy.

(Preview) Entity AI generation

Auto-generate and auto-translate reference values and synonyms in a non-default language:

1. Go to Agent settings > Languages and make sure you have one or more non-default languages enabled for your agent.
2. Navigate to the language drop-down at the top of the Console and select your chosen non-default language.
3. Navigate to the Manage tab, then click Entity types and choose one of your entity types.
4. When you enable a non-default language, Conversational Agents (Dialogflow CX) does not automatically translate any existing reference values and synonyms. To auto-translate these values into the new language, click + Copy from default language. All of the original data from the default language entity type will be translated into the new language and populated in the table. Be aware that translating more than 50 items at a time can cause errors. Each reference and synonym value counts as an item.
5. You can also add new reference values by clicking + Add new entry. You can type a reference value in your default language, and both the reference value and its synonyms will be automatically translated into the new language.

(Preview) Fulfillment AI generation

AI generation is available for route, route groups, and event handler fulfillment types. For more information about these route types, see the routes and event handlers documentation.

This feature will only appear if you are viewing the agent in a non-default language, and in one of the route types specified. Navigate to the Fulfillment section of the route. To automatically copy and translate the agent responses from your default language into the language you're viewing, click Copy and translate. Be aware that translating more than 50 responses at a time can lead to errors.

Language-specific data when using the API

To access language-specific data with the API, provide the languageCode parameter when getting, creating, or updating EntityType, Intent, or Page types. Go to the EntityType API reference

Go to the Intent API reference Go to the Page API reference

You must make separate API calls for each language to access all language-specific data. If your API calls update language-generic data, that data is updated for all languages.

Test in the simulator

When using the simulator, choose the language you want to use by selecting it below the agent name.

Supply a language using the API

When sending requests via the API, set the queryInput.languageCode field of the detectIntent method. For more information, see the Sessions type. Go to the Session API reference

Language auto detection

If the end-user's query language differs from the language specified in the request, Dialogflow can detect it and automatically switch to the end-user's language. See the language reference page for a list of languages available for this feature. The LanguageInfo field in a WebhookRequest identifies the input language, resolved language, and confidence score when language detection is enabled.

Enable language auto detection

You must enable language auto detection at both the agent and flow levels.

1. Go to Agent settings and select the Languages tab.

2. Check the box next to Enable language auto detection and switching to enable this feature for the agent. Click Save to save your changes.

3. Select a flow and navigate to Flow settings. In the flow settings panel, check the box next to Enable language auto detection and switching.

4. In the Languages drop-down menu, select the languages you want Dialogflow to be able to respond in. Make sure that you only select languages that are supported by the language auto detect feature.

5. Click Save to save your changes.

Requests that do not supply an agent's language

When Dialogflow receives a request that does not supply a language supported by the agent, the following rules apply:

• When a request does not supply a language, Dialogflow automatically determines the input language and responds in that language if automatic language detection is enabled. If automatic language detection is disabled, Dialogflow handles the request using the default language.
• When a request supplies a locale-specific language that is not supported by the agent, but the associated root language is supported by the agent, Dialogflow handles the request using the root language.
• System entity extraction is an exception to the previous rule. System entity extraction uses the supplied locale-specific language, even if the locale-specific language is not supported by the agent.

Integrations

Agent interactions through text-based integrations only use the default language of the agent.