Learning to read "Alice in Wonderland" may be the key to teaching machines to understanding the way we speak.
At least that is the tactic Facebook is using to impove the language capabilities of M, the company's virtual assistance.
Facebook began testing M in October, and the early reports of it were ecstatic. People have used M to do everything from purchasing tickets to ordering food. But perhaps the reason M is so good at what it does is because it's not pure AI like Apple's Siri.
Right now, Facebook's M is part-human, part-AI.
But Facebook's team is trying to improve M's language capabilities so it can better cater to users' needs. Mark Zuckerberg put up a Facebook post Thursday night explaining that the company is training a computer to predict the missing words in children's books for that very purpose.
"For this research, our team taught the computer to look at the context of a sentence and much more accurately predict those more difficult words — nouns and names — which are often the most important parts of sentences," Zuckerberg wrote in the post.
So far, the Facebook research team has discovered that its AI is best at predicting words when it's given just the right amount of context. "We call this 'The Goldilocks Principle,'" Zuckerberg explained.
"We still have a long way to go before machines can understand language the way people do, but this research takes us closer to building helpful services like M, our digital assistant in Messenger," Zuckerberg wrote.