Mozilla has released local machine translation tools as part of Project Bergamot

In the year In January 2019, Mozilla joined the University of Edinburgh, Charles University, University of Sheffield and University of Tartu as part of a project funded by the European Union. Project Bergamot. The ultimate goal of this partnership was to build a neural machine translation toolset that would allow Mozilla to develop a native website translation plugin, meaning that the engines, language models, and in-page translation algorithms should and should exist. It’s done entirely inside the user’s computer, so none of the data is sent to the cloud, making it completely private.

In addition, two novel features need to be introduced. The first was Forms Translation, which allows users to enter text in their own language and have it translated into the language of the page on the fly. The second feature was translation quality estimation, where low confidence translations should be automatically highlighted on the page to notify the user of potential errors.

This set of needs created a number of technological challenges for the team: The translation engine was written entirely in programming languages ​​into native code. We needed a way to streamline the distribution of the project to avoid builds that are compatible with all platforms supported by Firefox – impossible to scale and maintain. Also, an engine that needs to run fast enough on CPUs and not rely on GPUs is typically required in deep learning solutions.

Our solution was to create a high-level API around the machine translation engine; Port it to WebAssemblyand optimize the operations for matrix multiplication Run efficiently on CPUs. That allowed us not only to develop translation plugins, but also to integrate local machine translation into each website. Just on this websiteThis allows the user to make free-form translations without using the cloud.

of Translation add-ons are now available to install in the Firefox Add-ons Store. In Firefox Night, Beta and General Release. We’re looking for user feedback and you’ll see a button in the add-on. Fill out a survey That will help Project Bergamot’s collaborators understand which direction to take the product.

We also set it up to allow the community to contribute in new languages. General training line It helps expand the reach of the supplement, allowing fans to easily train new models.

This work is compatible with Mozilla Commitment Making the web accessible to everyone, regardless of language, by building open source projects that are important to our community with a focus on privacy. Please join us and Send suggestions — We need everyone’s voice to make this addition truly accessible to everyone.

This project received funding from the European Union’s Horizon 2020 Research and Innovation Program under Grant Agreement No. 825303. 🇪🇺