Keynote Speakers
Alexandra Birch
Translation and LLMs
26th June 9:15-10:15
Abstract: What is the future of translation research in the era of large language models? Brown et al. in 2020 showed that prompting GPT3 with a few examples of translation could result in translations which were higher quality than SOTA supervised models at the time (into English and only for French, German). Until this point, research on machine translation had been central to the field of natural language processing, often attracting the most submissions in annual NLP conferences and leading to many breakthroughs in the field. Since then, there has been enormous interest in models which can perform a wide variety of tasks and interest in translation as a separate sub-field has somewhat diminished. However, translation remain a compelling and widely used technology. So what is the promise of LLMs for translation and how should we best use them? What opportunities do LLMs unlock and what challenges remain? How can the field of translation still contribute to NLP? I will touch on some of my own research but I focus on these broader questions.
Bio: Alexandra Birch is a Reader in Natural Language Processing in the Institute for Language, Cognition and Computation (ILCC), School of Informatics, University of Edinburgh. She is a leader of the StatMT group and a co-founder of Aveni.ai - an award winning startup in speech analytics and conversational AI. Her main research focuses on machine translation and multilingual dialogue, but she has a broad interest in leveraging NLP to create compelling applications that improve people's lives.
Valter Mavrič
Harnessing the benefits of machine translation at the European Parliament: from current practices to future possibilities
24th June 11:00-12:00
Abstract: Machine translation (MT) is an essential tool for one of the largest institutional translation providers in the world: the European Parliament’s Directorate-General for Translation (DG TRAD). DG TRAD is home to 24 language units that embody and put into practice one of the core democratic principles of the European Union: multilingualism. In this complex environment, MT has become an integral part of DG TRAD’s work, helping it to manage an ever-growing volume of translation requests and allowing it to focus on the unique value that only humans can bring to the translation process.
The MT technology used in DG TRAD is a focal point of cooperation between the EU institutions and is constantly evolving. To best harness the benefits, DG TRAD relies on a dedicated team that carries out tests to explore the best ways of using MT for DG TRAD’s content.
This presentation will tell you, from a user’s perspective, about DG TRAD’s journey to identify the most efficient ways of working with MT. Here are some of the questions we will cover:
How well does MT handle the European Parliament’s content? Do all languages produce the same results? How does MT quality vary based on the type of content?
How does MT improve efficiency? What efforts are still necessary after integrating MT into DG TRAD’s workflow?
What about clear language? How well does MT perform in this area?
Finally, we will look at the new areas DG TRAD is exploring in this age of artificial intelligence (AI) and where we see that further research could provide added value.
Bio: Valter Mavrič is Director-General of the Translation Service (DG TRAD) at the European Parliament (since 2016), where he was previously acting Director-General (from 2014), Director (from 2010) and Head of the Slovenian Translation Unit (from 2004). With an MA in applied linguistics and further training in translation, interpretation, linguistics and management, he has a long experience as manager, translator, interpreter and teacher of languages. He works in Slovenian, Italian, English, French, and Croatian and is currently preparing a PhD in strategic communication.