Call Me: +34 636 02 26 84

GNMT – breakthrough or breakdown?

Just over a month ago I’ve written in my blog about a little known Google Translate option – the augmented reality translation, and it seems like Google was listening to me. At the end of September, to coincide with the 10th anniversary of Google Translate launch, the company has announced it has created a new algorithm called Google Neural Machine Translation system (GNMT). The full details can be found in a report published on 27th of September on Google corporate web, but the company claims it to be a huge step forward in the development of machine-perfect translation. The original algorithm behind the Google Translate launched 10 years ago was based on Phrase-Based Machine Translation (PBMT) and used the mechanism of breaking up input sentences into separate words and phrases and translating them independently, as opposed to analyzing and translation the whole sentence. According to Google, the new neural machine translation system is capable of learning. It considers the whole input sentence as a unit to be translated, simultaneously breaking it up into smaller pieces, that is words and short phrases, and analyzing them. It could be compared to a human eye that can perceive the whole picture without losing sight of smaller brushstrokes.
gnmt_en

Other techniques that made possible vast system improvements include the ability to divide complex words into small pieces and analyze them as separate words thus deducing the meaning of a complex source word from its components. Currently the new system is being tested and so far it’s available for English-Chinese translations, as this combination has proven to be the most difficult and error-prone, but it’s also being tested for French and Spanish which means we could see improvements in these language combinations any time soon. According to Google, the GNMT will reduce translation errors by more than 55%-85%, although it still lags behind a human-made translation. In their own words, “Machine translation is by no means solved. GNMT can still make significant errors that a human translator would never make, like dropping words and mistranslating proper names or rare terms, and translating sentences in isolation rather than considering the context of the paragraph or page. There is still a lot of work we can do to serve our users better. However, GNMT represents a significant milestone.”

It’s clear that Google wastes no time and inverts a lot of research, time and money into its translation algorithm, but as we can see even they admit that their translation mechanisms still have plenty of field for improvement. It’s very likely that in some years ahead we could trust the machine to make a perfect translation of virtually any type of document, but meanwhile a professional translator still remains our best ally in smooth communication with diverse cultures.

Author: anastasia

Leave a Comment

*

code

Uso de cookies

Este sitio web utiliza cookies para que usted tenga la mejor experiencia de usuario. Si continúa navegando está dando su consentimiento para la aceptación de las mencionadas cookies y la aceptación de nuestra política de cookies, pinche el enlace para mayor información. ACEPTAR