Google is now re-engineering their translation service and plans to implement a newly developed system, that is more accurate in translating. The system was tested to pit with a professional translator and guess what, the result was surprisingly delightful. Test results confirmed that the new translator system matched closely with the human’s fluency, particularly in the English and Spanish languages.
With this new system aboard, language barrier will never be an issue. We can now travel to other places and effectively communicate with the locals. We can now easily translate web pages and exchange messages with our foreign friends.
Google is now rolling out the implementation of this new system to translate Chinese to English and vice versa. They are also hoping for the sooner implementation of the system in their current translation service.
The huge improvement in the translation service could also help develop software able to carry out complex instructions, such as answering difficult queries or reading Wikipedia.
Google used the “deep learning” technique in their new translation system. This technique uses a network of mathematical functions, roughly inspired by the research of mammalian brains. The innovation will also improve Artificial Intelligence, particularly in speech recognition and image processing.
Google released a paper which presents the result of their experiment on translating a range of languages, from English to Spanish, English to Chinese, English to French. Also, each of these languages was translated back to English. The result of Google’s work was presented to people that can speak these languages fluently. Google asked them to compare the system’s translation of a document to their personal one. Despite slight hiccups on some of the translated phrases, the translators were happy with the result, because most of the translation was correct.
500 translators were asked to rate the fluency of the new system from 0 to 6. The test consisted of translating a 500-sentence long document taken from news articles and random Wikis. Google’s system scored 5.43, while human translation scored 5.5 on English to Spanish translation. They also conducted the French to English translation and got the closest score to human translators. The old translation system wasn’t able to score near human translation.
With their new technology, Google boosted the software’s capability with 87% compared to the old one. With this improvement around, the system is more capable and independent.
The old system uses a particular method, so it can translate the text. Also, it has an algorithm that can learn things, such as the reordering of words in the sentence, so it can understand different languages. Both systems (old and new) looked at many translated documents to learn translation.
This new method doesn’t rely on the developer’s given instruction to translate documents/text, but it can go on by its own to translate the text from source to other languages. It can create or read the text without worrying about the different concept of words. The new system works by breaking the text into fragments, which it then translates separately. After that, it analyzes and processes the text again, to establish the right order of words. The new software’s initiative is a mystery to its developers, they couldn’t believe that it actually works.
“It can be unsettling, but we’ve tested it in a lot of places and it just works,” – Le, Google Engineer
What do you think of this new translation system from Google?