Build Multilingual Chatbots with Watson Language Translator & Watson Assistant
The Pain Point
I asked the question, ‘What Language Barrier?’, stressing the importance of multilingual communication and how Watson Language Translator offers a solution. Following up, I wanted to showcase the potential of Watson Language Translator when paired with other Watson services, namely Watson Assistant.
Often conversational interfaces use only one language and switching between languages is difficult. Adding to that, a barrier enterprises face when developing chatbots is: industry specific multilingual capabilities. These types of problem are more complex, but when approached the right way, can open new markets!
How do these technologies work together?
Watson Assistant is a natural language chatbot that can be trained on intents, entities, and dialog for specific industries.
Say you want to expand the use case to be multilingual. You can train a new chatbot in another language or integrate Watson Language Translator for a faster approach.
Watson Language Translator can be set up to translate incoming user requests to the language the chatbot is trained in. A chatbot handles phrases and matches them with intents. Thus, even if there is a slight change in translation, Watson Assistant will be able to correctly identify the intent.
Once identified, the chatbot identify the intent to return a response. Before the response is seen by the user, Watson Language Translator will translate the response to the language in which the question was asked.
To improve translation accuracy, a forced glossary can be created with Watson Language Translator. A forced glossary allows for you to customize your translations. Given that the responses back to the user are fixed, a forced glossary can be set up so each of the responses returned by the chatbot are customized.
Try the Demo & Build your own!
We created a sample code pattern and demo allowing you to switch between languages in the same conversation. Try the demo above or follow the steps below to build your own.
Interested in building your own multilingual chatbot? Check out the Github repo to get started.
1. Train chatbot in English (or other language of your choice).
2. Translate the output responses defined in the chatbot using Watson Language Translator.
3. Review the responses.
4. If the translated responses seem less fluid or not domain specific, create a forced glossary for those responses. Each response customized with a forced glossary will always be translated in the defined manner.
5. This forced glossary ensures that the responses are always accurate.
6. You now have a multilingual chatbot!
Zia Mohammad is on the product management team for Watson Language Translator and Watson Natural Language Classifier. His passions center around: AI, emerging technologies, astronomy, and neuroengineering.
Feel free to comment or reach out for more information!