CEF Telecom Project: iADAATPA, the Intelligent, Automatic Domain Adapted Automated Translation for Public Administrations

2019-04-24

CEF Telecom Project: iADAATPA, the Intelligent, Automatic Domain Adapted Automated Translation for Public Administrations

How to provide a high-quality machine translation when several domains are covered in a single document?

eTranslation provides excellent results in specific areas (legal, EU-related issues), but for domains as specific as construction, health, finance or tourism, the quality and coverage can be improved (1). The variety of subject areas dealt with by digital public services across Europe and the number of official languages in the European Union represent serious challenges for public services that implement MT in their translation workflow.

The iADAATPA consortium (2) has contributed to address this issue by designing and developing a highly-secured translation router with language and domain detection mechanisms to switch between MT engines every time a new domain is detected in a document, even from one paragraph to another. This was achieved by providing a number of neural MT engines developed and trained with domain-specific data, and re-using the eDelivery and eTranslation CEF building blocks.

Public services such as the Lithuanian parliament or the Spanish Open Data portal were closely associated to the project’s activities by contributing data and taking part in the testing of the platform. For Lithuanian, the NMT-based approach provided significantly better results than statistical MT engines which reflected in final human translation gain in productivity reaching 40%.

Having been developed as part of EU-funded project, the code for this new technology platform is currently available to the Open Source Community at https://github.com/iADAATPA. It is available under the GNU General Purpose License v3.0.

iADAATPA ran over 18 months, from August 2017 to February 2019.

In this last stage of the project, the consortium partners have looked into a sustainable model for the iADAATPA service platform as a marketplace for customers (Public Administrations) and providers (Machine Translation Companies), which has been launched as MtHub (http://www.mt-hub.eu/) on February 28, 2019.

(1) ELRC initiative

The ELRC consortium is collecting Language Resource to feed and train the eTranslation system to help improve the quality, coverage and performance of automated translation solutions in the context of current and future CEF digital services.

http://www.lr-coordination.eu/home

(2) Project partners

The iADAATPA consortium was made up of Pangeanic (Spain), SEAD – the Secretary of State for Digital Advancement (Spain), Everis (Spain), Prompsit (Spain), Segittur (Spain), Tilde (Latvia), KantanMT (Ireland) and the Dublin City University (Ireland)

More about iADAATPA