Articoli recenti


by / Monday, 19 December 2016 / Published in DOTWALL

December 2, 2016 – To make the Google Translate automatic translation system more effective, the company has created a new language based on information processed by an artificial neural network.
Google Translate currently supports 103 languages and translates more than 140 billion words a day, and was recently made more efficient by the introduction of a system based on an artificial neural network that provides improved translation for the world’s most widely spoken languages.

Now the official Mountain View blog tells us that Translate will also be able to manage less common languages, for which it has not been directly programmed, thanks to the creation of its own artificial language. Traditional automatic translation systems usually break sentences down into smaller parts and then translate the individual words, an approach that frequently produces meaningless phrases. The new system uses the Google Neural Machine Translation (GNMT) artificial neural network, and works on an entire phrase so that it understands the context and turns out better translations. At the moment the system is used for the eight most widely used languages: English, French, German, Spanish, Portuguese, Chinese, Japanese, Korean and Turkish. Italian is not included but Google says it will be rolling out Neural Machine Translation for all 103 languages it normally handles.

There’s more: Google’s researchers are already talking about an extension to the system so there will no longer be any need to translate with the support of a group of reference languages: previously, translation from Japanese to Korean for example was triangulated through English, with the risk of altering the meaning of the text, but now this will no longer be necessary. The system will use an interlanguage, a sort of new common language that can’t be read or used directly by people, but is perfect for use by translation systems.

So could we do without translators, and hand everything over to Machine Translation (MT)?
Absolutely not
. As many experts in the application of innovation to multilingualism point out, technology is now indispensable to cut times and costs, and enable databases to be built up for companies and professionals, but only a human being can understand fully the context of a text and its lexical and stylistic characteristics, adopting a semantic approach geared to localization rather than simply to translation.

Dotwords co-founder David Orban, who is an advisor at Singularity University, observes that “one of most promising aspects is the ability of a team of people to use Artitifical Intelligence – at the most advanced level – to improve their performance.”

This is why the Dotwords R&D team is working on a project integrating a number of software systems with the focus on Machine Translation, starting from the assumption that the future will be oriented increasingly to the quality of post-editing.