Sign In

Communications of the ACM

ACM TechNews

To Build a Smarter Chatbot, First Teach It a Second Language

View as: Print Mobile App Share:
A thoughtful robot

Salesforce researchers have found that first teaching a machine-learning algorithm to speak another language makes it perform better in learning other tasks.

Credit: Absolute News

Researchers at Salesforce have developed a method that could improve the performance of modern language programs by teaching a machine-learning algorithm to speak another language before educating it on other tasks.

"We're taking machine translation data, and we're basically teaching the model how to understand words and context," says Salesforce chief scientist Richard Socher.

The team trained a deep-learning system to automatically translate between English and German by feeding a multilayer neural network a large volume of translated documents, and modifying the network's parameters until it could produce a decent translation for itself.

When this bilingual network was trained to perform other tasks, such as text sentiment analysis, its performance significantly surpassed that of a system that had not learned a second language.

Socher thinks this breakthrough could enhance the natural language capabilities of Saleforce's Einstein artificial intelligence platform. "For chatbots and automating customer support, this is super useful," Socher says.

From Technology Review
View Full Article


Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA


No entries found