Project description
A machine translator that doesn't leak
Machine translation services typically upload text to a server, which reduces user privacy and can break confidentiality. The EU-funded Bergamot project will run translation locally, ensuring privacy because data never leaves the user's computer. Efficiency research enables an ordinary desktop or laptop to translate thousands of words per second. The system also indicates how confident it is in the translations and assists in filling out forms in a foreign language. The software and models will be released as an open-source extension integrated with Mozilla Firefox, which will bring many benefits to non-expert users from both private and public sectors.
Objective
The Bergamot project will add and improve client-side machine translation in a web browser. Unlike current cloud-based options, running directly on users' machines empowers citizens to preserve their privacy and increases the uptake of language technologies in Europe in various sectors that require confidentiality. Free software integrated with an open-source web browser, such as Mozilla Firefox, will enable bottom-up adoption by non-experts, resulting in cost savings for private and public sector users who would otherwise procure translation or operate monolingually.
To understand and support non-expert users, our user experience work package researches their needs and creates the user interface. Rather than simply translating text, this interface will expose improved quality estimates, addressing the rising public debate on algorithmic trust. Building on quality estimation research, we will enable users to confidently generate text in a language they do not speak, enabling cross-lingual online form filling. To improve quality overall, dynamic domain adaptation research addresses the peculiar writing style of a website or user by adapting translation on the fly using local information too private to upload to the cloud. These applications require adaptation and inference to run on desktop hardware with compact model downloads, which we address with neural network efficiency research. Our combined research on user experience, domain adaptation, quality estimation, outbound translation, and efficiency support a broad browser-based innovation plan.
Fields of science
Programme(s)
Funding Scheme
RIA - Research and Innovation actionCoordinator
EH8 9YL Edinburgh
United Kingdom