Last year Microsoft announced the release of its Neural Network based translation system for 10 languages: Arabic, Chinese, English, French, German, Italian, Japanese, Portuguese, Russian, and Spanish. Today, Korean is being added to the list.
At a high level, Neural Network translation works in two stages:
- The first stage models the word that needs to be translated based on the context of this word (and its possible translations) within the full sentence, whether the sentence is 5 words or 20 words long.
- The second stage then translates this word model (not the word itself but the model the neural network has built), within the context of the sentence, into the other language.
Neural Network translation uses models of word translations based on what it knows from both languages about a word and the sentence context to find the most appropriate word as well as the most suitable position for this translated word in the sentence.
One way to think about neural network-based translation is to think of a fluent English and French speaker that would read the word “dog” in a sentence: “The dog is happy”. This would create in his or her brain the image of a dog. This image would be associated with “le chien” in French. The Neural Network would intrinsically know that the word “chien” is masculine in French (“le” not “la”). But, if the sentence were to be “the dog just gave birth to six puppies”, it would picture the same dog with puppies nursing and then automatically use “la chienne” (female form of “le chien”) when translating the sentence.
Here’s an example of the benefits of this new technology used in the following sentence: (one of the randomly proposed on our try and compare site: http://translate.ai)
M277dw에 종이 문서를 올려놓고, 스마트폰으로 스캔 명령을 내린 뒤 해당 파일을 스마트폰에 즉시 저장할 수 있다.
Traditional Statistical Machine Translation would offer this translation:
“M277dw, point to the document, the paper off the file scan command Smartphone smartphones can store immediately.”
Neural Network translation, in comparison, generates this clear and fluent sentence:
“You can place a paper document on M277DW, and then save the file to your smartphone immediately after the scan command.”