An assistant for any language

Photo by Hannah Wright on Unsplash

Providing a great customer service experience is tough. To do so, you have to meet your customers where they are and get them to engage, understand their questions accurately, and then solve their problems efficiently. As we’ve discussed in other posts, Watson Assistant does a great job helping you produce this kind of experience.

But what happens when you need to expand your assistant to support more of your customers’ needs? A key concern for any of our enterprise clients in expansion mode is whether the assistant can speak all of the languages that their customers speak. They want confidence that once they start building in the first or second language, that the full set of languages will ultimately be supported as well.

The good news here is that Watson Assistant now supports every global language, but before we jump into that, let’s take a look at how our clients tried to get additional language support before today.

Making a square peg fit in a round hole

Up until now, Watson Assistant supported 13 major different languages natively within the product (you can see the full list here). This means that nordic languages, Russian, Thai, Greek, and others spoken in certain parts of Asia and Africa weren’t officially supported. If clients wanted to create a new assistant to handle an unsupported language, they would have two options:

1. Purchase and use a language translation service in front of their existing assistant
2. Try to find a supported language that was close to the language they wanted (like using a Czech language model for Russian)

Using a language translation service is prone to the effect of compounding errors. If anything goes wrong in the original translation, the language model being used is likely to exacerbate those errors and prevent you from correctly understanding users. And if you decided to go with a similar language model — like using the Czech language model for Russian text — you’re likely to run into issues with the underlying grammatical structure of the language. In essence, each language model is performing a preprocessing like tokenization specific to the supported language. Thus, if you tried to feed in an unsupported language, you might run into major issues and, thus, incorrectly understand users.

Introducing the Universal Language Model

Watson Assistant is upending this paradigm and introducing a Universal Language Model to help! This new feature builds on the breakthroughs in intent classification to create a language model that can be used in over 100 different global languages. Using advanced natural language processing and AutoML, Dialog and Actions skills in Watson Assistant can adapt and train to a previously unsupported language. Now, Watson Assistant clients can build assistants in each and every language that they want to support, taking advantage of the best-in-class machine and deep learning techniques to solve end-user issues anywhere in the world. If you compare this to the 10 to 30 languages most other vendors support, it’s clear Watson Assistant is leading the field.

How can you use this new Universal Language Model to expand support to new languages? In Watson Assistant, you can create a new Dialog or Actions skill and select the Universal Language Model from our drop-down list. Once you do, your assistant will be ready to understand users in a new language outside of the 13 languages we support natively today!

“Another language” as the final option in the language selection list

Set the language when you create a new Actions or Dialog skill

Check out the Universal Language Model in Watson Assistant today! Happy building!

And if you’re new, you can learn more and sign up for Watson Assistant here.