google-neural-machine-translation

Google has been making strides with its neural machine translation software, expanding the number of languages it is able to translate with its computer system. The improvement is a welcome addition to the endeavors of the search engine giant to develop artificial intelligence and systems that learn similarly like a human brain.

In addition to an existing 8 languages that Google’s neural translator already knows, the machine will be adding Vietnamese, Hindi and Russian into its repertoire in two weeks, allowing a grand total of 11 languages thus far. Other major languages that Google’s neural translator can translate in 2016 include English, French, German, Spanish, Portuguese, Japanese, Chinese, Korean and Turkish.

machine-translation-search-engines

The software uses neural networks – a long coveted experimental computing method that is based on the human brain and central nervous system, allowing the computer system to translate complete sentences instead of simple phrases, making it a more effective translation software in comparison to anything in the market today.

“Neural translation is a lot better than our previous technology, because we translate whole sentences at a time, instead of pieces of a sentence,” wrote Google Translate product lead Barak Turovsky, in the blog post announcement. “This makes for translations that are usually more accurate and sound closer to the way people speak the language.”

The changes are expected to be rolled out across all Google search platforms, including Search, Translate.Google.com, Google Apps and even automatic language translations. Turovsky has also noted that there will be more neural machine language rollouts in the coming weeks.

Established with Google Brain in 2011, Google’s research project involves deep learning for its artificial intelligence systems. By September last year, Google has been able to utilize its wide array of neural networks for their software to learn languages, with a research team in charge of the Google Neural Machine Translation (GNMT) efforts. By comparing millions upon millions of samples, GNMT improves its own translation, adding ideas such as broad context to improve its translation accuracy.

neural-translation

In addition to the neural networks, the GNMT uses a “zero shot” translation protocol, wherein it directly translates non-English languages to other non-English languages (Japanese to Korean) to get a better translation. In comparison, its old translation software only uses a source language method where it uses its English source language as a middle language before doing the full translation (example, Japanese to English, English to Korean instead of Japanese to Korean).

Google’s Deep Learning project has been making strides in the industry, with its primary artificial intelligence learning Go – an Asian board game considered to be the hardest board game on the planet, and defeating top players by learning from its mistakes and eliminating moves that does not allow it to win. Google’s Deepmind AI has also been tested and is concluded to be competitive in self-preservation situations while being helpful in cooperative situations.