Mozilla has released as part of the Bergamot project, the launch of a machine translation system browser-based, something similar to the Chrome translator, but with the difference that will be that the option to translate web pages in Firefox it will work without internet connection.
The launch of this project will allow a standalone page translation engine to be integrated into Firefox not accessing external cloud services and processes data exclusively on the user's system. The main objective of the development of this project is to guarantee confidentiality and protect user data from possible leaks when translating the content of the pages opened in the browser.
Bergamot is being developed at Mozilla's headquarters in Berlin with the participation of researchers from various universities in the UK, Estonia and the Czech Republic. The development is funded by the European Union as part of a grant received under the Horizon 2020 program.
The amount of this support is around three million euros. The project is designed for three years. Mozilla has opened a job as a specialist in machine learning systems to participate in the development of an engine for the translation from one language to another.
Of the developments related to the Bergamot project, the following are mentioned:
- The University of Edinburgh works on the development of Marian machine translation framework, which is built on the basis of a recurrent neural network. The framework is written in C ++, and can make use of the GPU to speed up learning and translation. The project comes under the MIT license.
- The tool Neural Monkey developed at the University of Prague to process natural language information using sequential machine learning methods. The project uses the TensorFlow framework and it can be used to prototype machine translation systems and classify information in natural language. The code is available under the BSD license.
- The QuEst ++ project, which is developed at the University of Sheffield, is used to evaluate and predict the quality of machine translation systems.
- Mozilla is developing on a speech synthesizer (TTS) and speech recognition engine (Deep Speech)
- The project ParaCrawl, funded by the European Union, which accumulates a database of simultaneous translations of several phrases in different languages, which can be used to form machine learning systems.
- The basis of the project is bitextor, which is a bot, which indexes multilingual websites and automatically finds the same texts presented in several languages. The parallel translation examples base is made up of 24 languages.
As shown in the demo video, a new button will appear in Firefox, when the user clicks on this button you can select the language to which you want to translate. Next to it will be placed a button to return to the original language.
The developers say that the new local system will provide better privacy and data protection for users by translating the content of the page in the browser.
A few years ago, Mozilla was already trying to add a translation feature similar to Google Chrome in your browser, but abandoned it due to costs support too high.
Firefox has a built-in feature to translate pages, but it is linked to the use of external cloud services (compatible with Google, Yandex and Bing) and it is not enabled by default. To activate the function it is necessary to go to "about:config”And change the settings of«browser.translation").
The translation mechanism also supports automatic language detection when a page is opened in an unknown language and displays a special indicator with a proposal to translate the page.
The prototype of the translation system that is being developed as part of the project Bergamot uses the same interface to interact with the user. This translation system can be tested in Firefox Nightly by enabling the settings browser.translation.ui.show y browser.translation.detectLanguagee.