¸£ÀûÔÚÏßÃâ·Ñ

Linguistics Circle Seminar - Ms Agnes Farkas

This event will be taking place on Friday 7 November 2025 at 12:00-13:00, face-to-face, at the FEMA Building, Room 419.

The special guest will be Ms Agnes Farkas (AI Language ¸£ÀûÔÚÏßÃâ·Ñ Advisory, Directorate-General for Translation, European Commission).

Title

Digital Europe AI-Based Multilingual ¸£ÀûÔÚÏßÃâ·Ñ: The journey from early machine translation to an EU institutional large language model

Abstract

The journey from early machine learning to the development of an EU institutional large language model is a testament to the evolution of language technology within the Directorate-General for Translation (DGT).

This progression began in the 1970s with the introduction of a rule-based machine translation approach which laid the foundation for future advancements. By the 1990s, the implementation of translation memory technology revolutionised translation processes, enabling the storage and reuse of translations across EU institutions, significantly impacting machine translation practices.

The transition to a statistical machine system in 2013 and subsequently to a neural machine translation system in 2017 marked a pivotal leap incapabilities, culminating in advanced neural models that allow for more nuanced translations across the EU's 24 official languages and additional languages.

Innovation continued through the early 2020s with the incorporation of AI-based multilingual services, including speech-to-text technology, an anonymisation tool, and a summarisation tool. With the advent of large language models, innovation was pushed even further.

The introduction of LLM-based tools such as eBriefing, eReply, WebText, and Accessible Text marked a new era, offering AI-powered tools to support daily tasks for the European Commission staff and eligible users in member states.

The development of the EU institutional LLM represents the peak of this transformation. Built on core European technologies, this model aims to address limitations in language diversity and cater to the specific needs of public sectors and low-resource languages. By leveraging the powerful processors of European supercomputers and the extensive multilingual datasets of Euramis, the European institutions' translation memory database, the DGT is actively working on training a high-quality EU institutional large language model to further strengthen multilingualism in Europe.

Bio:
Ágnes Farkas – Team Leader – AI Language ¸£ÀûÔÚÏßÃâ·Ñ Advisory, Directorate-General for Translation, European Commission 

Ágnes Farkas leads the Directorate-General for Translation's AI Language ¸£ÀûÔÚÏßÃâ·Ñ Advisory, an advisory body tasked with promoting eTranslation and other AI-based multilingual services developed by the European Commission for EU institutions and eligible users across EU Member States. The advisory body also provides user support and expert guidance to clients. Previously, she worked as a policy officer in the field of multilingualism and as acting head of unit and translator in the Hungarian translation department of the Directorate-General for Translation. She holds master’s degrees in translation and interpretation, as well as in French language and political sciences.