What is Google Neural Machine Translation (GNMT)?
In September 2016, Google announced working on Google Neural Machine Translation project. Later in November, Google Translate began using its new Neural Machine Translation. This system uses a large artificial neural network capable of deep learning. Google Neural Machine Translation uses millions of examples and improves the quality of its translation day by day. GNMT doesn’t have any universal interlingua but rather aimed at commonality found in between many languages. In an article on The Keyword with the title Found in translation: More accurate, fluent sentences in Google Translate, Barak Turovsky suggests that:
“In September, our researchers announced Google’s version of this technique. At a high level, the Neural system translates whole sentences at a time, rather than just piece by piece. It uses this broader context to help it figure out the most relevant translation, which it then rearranges and adjusts to be more like a human speaking with proper grammar. Since it’s easier to understand each sentence, translated paragraphs and articles are a lot smoother and easier to read. And this is all possible because of end-to-end learning system built on Neural Machine Translation, which basically means that the system learns over time to create better, more natural translations.”
With this update, Google has made a giant improvement but says that this is only the beginning! Google Neural Machine Translation aims to roll this system into all languages supported by Google Translate. Google also grants access to GNMT via Google Cloud Platform, so developers can use the technology within their own applications.
“…our goal is to eventually roll Neural Machine Translation out to all 103 languages and surfaces where you can access Google Translate.”
Barak Turovsky also suggests that they are relying more on Translate Community and more excited about translating and understanding the world a little bit better.
Google Neural Machine Translation Supported Languages
This new translation engine at first enabled for eight languages. These eight languages include to and from English and French, German, Spanish, Portuguese, Chinese, Japanese, Korean and Turkish. Google also mentioned that these languages cover more than 35% of the world’s Google Translate queries. Then Russian, Hindi, Vietnamese and Thai languages added. about the same month, Hebrew and Arabic languages also supported Google Neural Machin Translation (GNMT). Later on Dutch and some European languages related to English and then in other Indian Languages supported GNMT. For a full list of all supported languages, refer to Wikipedia’s page about languages supported by GNMT.
“The most important aspect of making the web more useful and meaningful for all of India is to make India’s Internet more representative of the India we live in. India today has 234 million Indian Language users who’re online, compared to 175 Million english web users, we expect another 300 million Indian language users to come online in the next four years” – Rajan Anandan, VP, India & South East Asia, Google
GNMT and Zero-shot Translation
While switching to GNMT improved the quality of the languages, scaling up to all supported languages presents a significant challenge. This challenge is said to be overcomed by allowing for a single system to translate between multiple languages. This method enables “Zero-Shot Translation” — translation between language pairs never seen explicitly by the system. Google Neural Machine Translation (GNMT) system is said to be able to handle zero-shot translation. This means that it can translate directly one language into another.