Posted: 11 Dec 2018 12:06 EST Last activity: 14 Jan 2019 18:13 EST
How to address multiple languages in a Web Chat interface.
I would like to know if there is an out of the box solution to have a web chat interface to speak in the detected language. And what is the best practice for creating the Conversation Flow so that it can adapt to the identified language.
***Edited by Moderator Marissa to update SR Details***
first of all - everything in the WebChat conversation works the same as entire platform, the only difference is that for WebChat the copy of template's operator (configured in webchat channel configuration) is created so if the template has some locales specified, probably they will be used.
WebChat OOTB doesn't fully support multi-language configuration in simple way, there are some limitations but it is possible to handle the runtime in different languages.
If we are talking about the flow, I have some doubts about the "conversation builder" based solution as it uses non-question rule based question shapes. I would suggest to use the Question-Rule based questions in your flow and use the Paragraph rule as the question type - this will allow you to fully and easily localize the questions, what's more - you can nicely format them etc...
Please also keep in mind that for WebChat conversation it is important to have properly configured text analysis, by default text analyser (taxonomy and entities models) are generated for ENGLISH so it is able to detect only English. You would need to find your TextAnalyzer (Clipboard Viewer -> ChannelConfig page -> pyFreeTextModel property) and open it.
In TA rule please find the Advanced tab and check whether "Automatically detect language" is selected. It has to be selected to let the analyser use multiple languages.
Back to first tab and open the taxonomy (topic model) - you will need to define the models for each language you want to support and update them manually. Possibly you will need some assistance from Decisioning experts how to do that from Analytics Center portal or directly from this ruleform.
You will need to define the language-specific models for each analysis used in WebChat (Sentiment, Topic, Entity)
Please also note that some functionallity doesn't support the multi-language configuration
You will be not able to use OOTB NLP Feedback mechanism which is triggered when bot doesn't understand the user's intent
You will be not able to define keywords for topic detection directly from channel command's configuration (Text analysis tab in response/create case configuration)
Hi, is it possible to circumstance the question rules dialog by the .pyLanguage system property. Will that property retain its value for the currently recognized conversation text? Have you ever considered this? Do you think it can work? Your comments are very appreciated. Thanks,