Books & Rights Marketplace 
日本語  |  ENGLISH 
Welcome Question?
Privacy Policy | Terms & Conditions  
First time user?
Please register online today !  
Register Now! 
MemberLog-in: Password: Login
 
 
Affiliated Partners
 
SEARCH
Topic
Author(Surname by initial)
 
ACCESS RANKING
This week's Top 10 Ranking 
1.  LOOPS ― The Seven ...
2.  The spiritual way ...
 

Become a fan of BRM


Promote Your Page Too
 
 
News & Events
2016.10.27 Deep Learning adopted to Google Translation
On September 27th, Google Inc. announced the development of the “Google Neural Machine Translation(GNMT)”, their new translation system utilizing deep learning technology.
Google Inc. has adopted the “Phrase Based Machine Translation” (PBMT)system in the past that separates sentences into words and phrases, and translates them each individually. However, regarding several language pairings, the adoption of this new translation system allows for the creation of more natural and accurate translated sentences than PBMT.

A technology called Neural network Machine Translation(NMT)was adopted to this new translation system. NMT is an algorithm model designed by modeling a human brain, and it has been designed to perform pattern recognition, and is suitable for automatically processing complicated, vast amounts of data.

GNMT is an original NMT that Google Inc. realized based off of “TensorFlow”,(the machine translation system which released as open source)and the “Tensor Processing Unit”(TPU)(the hardware for machine translation).

The new translation system Google Inc. adopted was introduced for a Chinese-English pairing. Google Inc. plans to introduce GNMT to other language pairings during the coming months.

The article states that the GNMT system has decreased 55-85% of the translation error in a number of language pairings in comparison with the PBMT system when Google evaluated each of the translated sentences that the PBMT system or the GNMT system had translated. It did so by using articles from Wikipedia or news sites.

However, it states that the GNMT system produced mistakes that humans never make in translation, such as making mistakes of people names, names of object, uncommon words, or forgetting the translation of words.

For more information, please refer to the following websites:
 
http://news.mynavi.jp/news/2016/09/28/072/

https://arxiv.org/abs/1609.08144

https://research.googleblog.com/2016/09/a-neural-network-for-machine.html