Multi-lingual support for Q&A bots

📘

Note

In this section we will be covering only the enhancements to the platform after the introduction of multilingual support in Q&A bots. For further information on any other feature, please refer to other sections in this help page.

Creating multilingual Q&A Bots

Multilingual Q&A bots creation is similar to single lingual Q&A bots creation. Refer to the Q&A bots page for steps.

Configuring languages for the bot

To configure languages in a Q&A bot, follow these steps -

• Go to Settings --> Language

• Click on 'Add Language(s)'

• Select all the required languages

• Click on 'Add'
• Once added, enable/disable languages to activate/deactivate support in the bot and click on 'Update bot'

Default language

The default language for a bot has implications across inference, articles, curation, testing, curation, etc. For an incoming request where the language isn't specified, the bot will respond in the default language. The user can manually set all responses for any of the supported languages to be sent in the default language.

Setting a default language for a bot

Users can edit the default language for a multi-lingual Q&A bot. After another language is set as default, English (en) can be disabled/deleted.

Default languages can't be disabled/deleted

🚧

Note

Changing the default language for a bot will have an impact on articles, curation, testing, and preview experiences.

Multilingual articles

Article creation process can be found in the Q&A bots 'Articles' section. To navigate between different languages, click and expand the 'Language' dropdown and select the needed.

📘

Mandatory fields in an article

Article name and title(s) are mandatory fields. Title(s) is the first variant(s) for each language.

NLU engine switcher controls

With multilingual support, in Swiftmatch, the language encoder is toggled on by default. The user can enable/disable 'Expand contractions', 'Remove special characters', and 'Spellcheck in inference' by selecting/deselecting the selection box and clicking on 'Update'

📘

NLU engine multilingual support

Currently in Q&A bots, only Swiftmatch has multi-lingual support. All features mentioned in this section will be disabled when the engine is not Swiftmatch.

To enhance the bot builder platform’s multi-lingual capabilities in Q&A bots, we have introduced our new multi-lingual model ‘Polymatch’ as an option for training bots. This is available as a part of the Swiftmatch NLU engine’s advanced engine settings. For a bot trained using Polymatch, a distinct model is created for each language added to the bot and inference is performed based on the language of incoming user query. The existing language agnostic model is available as Language encoder in the dropdown. Since Polymatch is not language agnostic and trains the corpus for each language, it requires significant training data for each language unlike Language encoder. Please note that the languages supported by Polymatch and the existing language encoder are different. Please refer to the platform documentation for more information.

📘

Remove stop words, wordform expansion, synonym, and wordforms checkboxes will be disabled when English

isn’t there in bot’s languages

Previewing the bots

For in-depth steps for various ways of previewing a bot please visit the 'Previewing bots' section.

Multilingual bot previews now have the capability of auto-detecting user utterance language. User can select the language in the preview by clicking on the language selector and searching/selecting the language from the list of available options.

Changes to sessions

Along with other transaction information for a session, users can now also see the detected vs bot response language, language encoder status, and the language detection provider in sessions.

Changes to testing

For multilingual bots, a language parameter has been introduced to the test case definition. To define a test case in a language click and select language from the dropdown. For more details on testing please refer to the 'Testing' section.

Language related parameters have also been added to the execution results section.

Model history changes

The model history tab can now log the languages with which the multi-lingual Q&A bots have been trained. Simply add the training comment and the language related data will be automatically appended and shown in model history.

Languages supported by Q&A bots

📘

zh includes support for all zh-cn (simplified), zh-tw(Traditional), zh-hk (Traditional - Cantonese)