Don't Freak But Google Researchers Discovered Their AI Translation Tool Invented Its Own Secret Language
Don’t panic but Google’s AI may have accidentally (or not so accidentally), sort of, kind of, created its own secret language. This internal self-taught language has apparently been developed by the company's Neural Machine Translation (NMT) and not a human.
How was the secret language created and discovered in the first place? Google announced their new Neural Machine Translation this past September. The Neural Machine Translation translates entire sentences instead of small pieces. It examines the broader context of the sentence in order to produce the more relevant translation. The result is a translation that uses proper grammar and does not sound like a robot. The Neural Machine Translation is an “end-to-end learning system built” that learns over time to develop more natural translations.
NMT has been fairly slow in the past but its accuracy has improved from 55 to 85 percent. Google has been particularly anxious to improve their Chinese to English translations. The NMT is currently available for Google translate in their eight most popular languages: English, French, German, Spanish, Portuguese, Chinese, Japanese, Korean, and Turkish. Neural Machine Translation will eventually be available for all 103 of their current languages.
Google’s NMT developers first translated from English to Korean and vice versa, and then from English to Japanese and vice versa. They were curious to see if the machine could then translate Korean to Japanese without using English as a go between. The answer was yes it in fact could translate directly.
It's how the Google AI achieves this that is a bit of a mystery. It appears that the NMT has created its own internal language or "interlingua." It examines concepts and sentence structures instead of word equivalents. As a result, the NMT has created translations that are more accurate and natural. Google’s NMT creators, however, are unsure how the neural networks work or what exact concepts the NMT has learned translate languages directly. In short, Google's AI has created its own secret language we humans do not fully understand. A white paper that details the researchers' work, including some detail on the mystery "interlingua," can be found here (PDF).
How was the secret language created and discovered in the first place? Google announced their new Neural Machine Translation this past September. The Neural Machine Translation translates entire sentences instead of small pieces. It examines the broader context of the sentence in order to produce the more relevant translation. The result is a translation that uses proper grammar and does not sound like a robot. The Neural Machine Translation is an “end-to-end learning system built” that learns over time to develop more natural translations.
Google’s NMT developers first translated from English to Korean and vice versa, and then from English to Japanese and vice versa. They were curious to see if the machine could then translate Korean to Japanese without using English as a go between. The answer was yes it in fact could translate directly.
Visualization of translation system memory while translating a sentence in multiple languages.
It's how the Google AI achieves this that is a bit of a mystery. It appears that the NMT has created its own internal language or "interlingua." It examines concepts and sentence structures instead of word equivalents. As a result, the NMT has created translations that are more accurate and natural. Google’s NMT creators, however, are unsure how the neural networks work or what exact concepts the NMT has learned translate languages directly. In short, Google's AI has created its own secret language we humans do not fully understand. A white paper that details the researchers' work, including some detail on the mystery "interlingua," can be found here (PDF).
