loading

Language Convertor Service – Build Multilingual Virtual Agents With Ease


Introduction

As companies expand globally, language can act as a barrier but drive opportunities at the same time in the Customer Service industry. We need to converse in the local language to ensure a smooth experience for the customers.

This is why companies are inching towards implementing virtual agents that can support multiple languages as they scale their businesses globally. With a multilingual virtual agent, organizations can interact with users in multiple languages or the language they’re familiar with. However, building multilingual virtual agents requires more than just processing text or dialogue in a particular language through a language translator. A virtual agent must be aware of the end user’s cultural and regional nuances that seep into conversation and this needs additional efforts. Quantiphi’s language converter for Conversational AI solutions aims to simplify this process of developing a multilingual virtual agent.

Building Multilingual Virtual Agents

The Virtual Agent Development Lifecycle consists of the following stages

  • Conversation flow design
  • Training data generation
  • Creation of intents and entities
  • Context management
  • Backend system integration
  • Fine-tuning and optimization
  • Testing

We observed that these stages consumed a major chunk of time for our engineers when providing support for a new language – Training data generation, context management, fine-tuning and optimizing, and testing. The idea to automate these stages occurred to us when we were working on a Rapid Response Virtual Agent solution for the Department of Labor of a large US state, at the beginning of the pandemic. We had to build a multilingual virtual agent in 5+ languages as part of the engagement in less than a month. During this time, we prototyped the language convertor solution platform and successfully used it to help the Conversation Bot Engineers focus on other stages of the life cycle.

Curious about how we did this? Let’s do a technical deep dive into the service.

Building the Tech Stack

We used Google Cloud Platform services to develop the platform. The platform allows us to upload the existing virtual agent, choose the required languages, and download the translated agent supporting all the chosen languages.

High Level Components of Language Convertor Platform

Application Services

We follow a microservices pattern internally for all our platforms, language converter service being one of them. The application services are responsible for authentication, authorization, handling all the incoming translation requests and managing the translation queue, interaction with storage systems, etc.

Content Storage

We use Google Cloud Object Storage for storing the uploaded virtual agents and Google Cloud Firestore to monitor and track the translation process in real-time.

Translation Service

Translation service plays a critical role here. It does the job of processing the virtual agent and making it multilingual. We have broken down the entire process into various stages as follows:

  • Data Extraction: As the name suggests, this stage extracts intents, training phrase, and entities from the uploaded virtual agent
  • Language Translation: This stage handles the language conversion from source to target language and entity annotation post the translation is done
  • Agent Builder: This is the last stage that binds together the existing virtual agent with the new generated data and bundles it into a shareable zip file

The current features supported by the language convertor platform are:

  • Conversion to a new language with context management
  • Intent, entity, response, and other translation support
  • Entity Annotation in the new language
  • Generation of ready to import the translated agent zip file

Implementation

The solution has been used to implement 5+ omni-channel (Chat, Voice) virtual agents for various industries (public sector, healthcare). Some of the translated agents helped us speed up the development process by weeks and we were able to release agents to production ahead of schedule.

In an average scenario with medium size agents (50 intents), the entire translation process from one language to another takes about 10-15 mins. We have been using this tool internally as well to develop multilingual cross channel virtual agent templates across industries and add them to our repository.

Leveraging Language Convertor Services for Your Organization

With Quantiphi’s Language Convertor Service, you can now convert your existing virtual agent to 60+ languages within minutes. The ease and flexibility of using the User Interface makes weeks of efforts seamless. It helps businesses with:

  • Hassle-free translation with a few clicks
  • Reduced development time
  • Reduced efforts
  • No prior knowledge of language required to translate the agent
  • Increased developer productivity

Get in touch with our experts to explore our Language Convertor Service and Conversational AI capabilities.

Written byTwisha Saraiya
Software Engineer, Qonversational AI Team

Get your digital transformation started

Let's Talk