A step-by-step Kickstarter to Power Virtual Agent and Azure Cognitive Service for Language (PVA) is getting better and better from Release to Release, but there are situations where you want to use to improve your user’s experience. Power Virtual Agent Azure Services Why not combine Azure Cognitive Services for Language with PVA to add missing elements? serving your customers and employees immediately in their native language without extra effort and without investing time in translating and maintaining your QnA in other languages. Imagine there is no need for additional resources, but your customers and employees will feel more directly addressed. Imagine your employees can refocus on delivering value to your customers instead of being a voice interface of a QnA collection. Microsoft Power Virtual Agents are perfect for automating customer service, and when combined with Azure Cognitive Services, they are almost infinitely extensible. Imagine A simple Power Automate flow helps your Power Virtual Agent serve your customers in over ninety languages. How easy is that? PVA has a built-in topic that can serve as a hook to the outside world, the “Fallback” system topic. Activate the “Fallback” topic To activate the “Fallback” system topic, go to Settings within the editing canvas (click on the cog symbol), then System fallback, and then “+Add”. After activating the topic, you will find it within “topics”. The “Fallback” topic is ready to connect with the outside world, as everything not recognized by our already in-place topics will end up here as “unrecognized user input”. /Extend Power Virtual Agent to a multilingual QnA Resolver If you have no PVA bot handy, try my Example as a starter. “Power Virtual Agent in a Day” This example will integrate a FAQ Database on “growing chilies”. Data from a random web page, “Greenhouse Garden Shop — will be used. Growing Chillies FAQ I´m not related to this shop but let´s respect their knowledge of growing chilies and use their questions and answers here!! Following my slogan: “Sriracha on everything!” Chilies are the main ingredient of . :) Sriracha Custom question answering We created a new project on and chose “Custom Question Answering”. Azure Language Studio Choose the option “I want to set the language for all projects created in this resource” > select English > Select Next. Enter a project name, “PVAinaDay-CQA”, a description, and leave the default answer with a setting of “No answer found”. Review your choices and select Create project From the Manage sources page, select Add source > URLs. Select Add URL, enter the following values, and then select Add all: URL Name URL Growing Chillies https://www.greenhousesensation.co.uk/growing-chillies-faq/ Language Studio will now import your FAQs into the project and gives you single and multiturn Question and Answer pairs. To use the imported content, we need to clean up the answers and make them usable for a bot. We must remove pictures and links if our bot is a voice bot. To make the Answering more fluent, we will add variations to each question. And deploy the knowledge base. After deploying the knowledge base, we press “Get prediction URL”, and we note down the Site URL(Red) and the Account Key (Yellow); we will use both later in Power Automate. Integrate Custom Question Answering and Microsoft Translator into Power Virtual Agent via Power Automate We jump back into your newly created “Fallback” Topic and remove the “Escalate” and the “Message” node. After deleting the nodes, we create a new “Call an action” node. Power Automate will open, and we can create a new flow. First, we rename the flow and define the input. To determine the language of the question, we will initialize a new variable, “QuestionLanguage”. We will use it to store the language code. Create a new node by searching “Microsoft Translator V2” and selecting “Detect language”, and set Text to the content of the variable “UnrecognisedUserInput”. We use a set Variable Node to store the output of “detect language” into the “QuestionLanguage” Variable with a translate text node (Microsoft Translator V2); we translate the content of “UnrecognizedUserInput” to English. Create a new node by searching “Language — Question Answering” and selecting “Generate answer from Project”. We noted the account Key and Site URL earlier in this guide; we use the values now. Set “Question” to the output of the translation node. Next, Initialize a new variable with a name, e.g., “CQA_Answer” A “Apply to each” Node with a “Translation” and a “set Variable” action will translate back to the language the question was asked in. Finally, we adjust the Return Value to the content of the “CQA_Answer” Variable. After saving the flow, we can select the flow in the action node, adjust the input, and create a message node to display the output of the flow. The flow will result in the following: A question asked to the bot without a topic in PVA will end up in the “Fallback” topic. Custom Question Answering will help with the answer, even to a question not asked in English. Additional Information Custom Question Answering — [successor of QnA Maker] documentation Conversational Language Understanding — if you would like to make the next step — [successor of LUIS] documentation Power Automate Connectors — documentation Featured image by on Unsplash. Sebastian Bill Also published here.

Automating Multilingual Customer Service with Power Virtual Agent and Azure Cognitive Services

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

Generating and Manipulating Text with Azure OpenAI Services

Here's How You Can Architect Advanced Routing Patterns in Symfony

How to Choose and Set Up AI Phone Systems for Small Businesses: A Complete Guide

The New Anatomy of Customer Experience - Part 2

Why We Bet on Clojure and Built Freshcode Around It

Multilingual Isn’t Cross-Lingual: Inside My Benchmark of 11 LLMs on Mid- & Low-Resource Languages

Generating and Manipulating Text with Azure OpenAI Services

Here's How You Can Architect Advanced Routing Patterns in Symfony

How to Choose and Set Up AI Phone Systems for Small Businesses: A Complete Guide

The New Anatomy of Customer Experience - Part 2

Why We Bet on Clojure and Built Freshcode Around It

Multilingual Isn’t Cross-Lingual: Inside My Benchmark of 11 LLMs on Mid- & Low-Resource Languages

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps