My prediction is that this is wrong. They'll be able to understand widely used terms, but once you start speaking in a more creative way it will fail. At least if there is no change in the construction of AI - because the subtleties of everyday language are not 'understandable' just via empirics.
Reading tip: Douglas Hofstadter's article "The Shallowness of Google Translate".
From the article:
"To my mind, translation is an incredibly subtle art that draws constantly on one’s many years of experience in life, and on one’s creative imagination."
This is true, I just think that we'll arrive at the tools to accomplish this in the future.
Specifically I think(hope) it will be through clever application of GANS and reinforcement learning after a few more applications of moore's law.
Advanced AI would be able to learn about us through replaying years of possible generated experiences.