Browser-based Multilingual Translation

Project Information

Bergamot

Grant agreement ID: 825303

DOI

10.3030/825303

Project closed

EC signature date 12 November 2018

Start date 1 January 2019

End date 30 June 2022

Funded under

• INDUSTRIAL LEADERSHIP - Leadership in enabling and industrial technologies - Information and Communication Technologies (ICT)

Total cost

€ 2 999 096,25

EU contribution

€ 2 999 096,25

Coordinated by

THE UNIVERSITY OF EDINBURGH
United Kingdom

Periodic Reporting for period 2 - Bergamot (Browser-based Multilingual Translation)

Reporting period: 2020-07-01 to 2022-06-30

Summary of the context and overall objectives of the project

Machine translation, such as from French to English, is commonly used in travel, e-commerce, government, and to support Ukrainian refugees. Typically text is sent the cloud for translation but this endangers user privacy. In a business or government context, cloud translation may be banned to protect confidentiality. Our machine translation system runs locally, preserving privacy and removing the risk of disclosure. Moreover, it does not require Internet access.

Running locally requires speed to run on desktop CPUs, low memory consumption, and small download sizes. The experience should be integrated with a desktop application and in the browser. A Firefox extension translates web pages locally. Going beyond simply translating pages, an assistant will help users fill out forms that expect responses in another language. The system should also estimate its own confidence and display it to users, warning of potentially incorrect translations.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

The project has a Firefox extension with 34,000 users that translates locally and a desktop application at https://translatelocally.com. The applications support translation for Bulgarian, Czech, English, Estonian, French, Polish, and Spanish. On a range of consumer laptops and desktops dating from 2012 to 2019, the desktop application translates 1000-9000 words per second.

An assistant supports filling forms where users can type in their own language and translations are entered as responses. Users can tick a box "highlight potential errors in red" to see where translation may be lower quality according to the system's self-assessment.

Our project featured in over 100 news articles including TechCrunch, Yahoo News, making #1 on Hacker News twice.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

Free local translation preserves privacy and empowers users who work with sensitive data. Extending Firefox means the project will be widely available. Open source enables extension and the possibility for the community to contribute translation models for underserved languages. The form filling assistant enables users to respond in languages they do not speak, going beyond ordinary web translation.

Our project has pushed boundaries of efficient neural network models, making them faster or supporting larger models on consumer devices. These methods have implications the speed and quality of other systems like captioning and summarization.

Estimating the system's own quality ensures trust in algorithms, a rising societal issue, is placed more appropriately. Moreover, it requires less labeling, so the technology can be applied more widely for languages with less data available.

browser-after-crop.png

big-translation.png

qe.png

beautiful-person.png

bafin.png

bar.png