Google announced their Neural Machine Translation System (GNMT) recently, which is now in use for Chinese-to-English machine translations. The GNMT uses neural network technology to improve machine language translation. Current machine translations operate by breaking sentences up into words and phrases from one language and finding equivalents in another.
Sometimes the phrase-to-phrase method doesn’t work, so translations “are difficult for existing systems to handle and translate correctly without human intervention, or dense programming requirements,” says The Stack . GNMT fixes this problem by looking at the entire sentence while still paying attention to individual words. Neural networks have been trained to look at images and words like the human brain, breaking big pictures into little sections. TechCrunch reports that the system doesn’t know tenses (past, future, imperfect..); instead, the translations are rooted in math and statistics.
GNMT has several new technical advances including the translation of uncommon words (breaking them into pieces), shorter computing time, and better accuracy. Inputs and outputs in the system “trains” the system. Neural networks design themselves through many iterations, so it is constantly learning.