21 November schedule
8.30 | Registration and welcome coffee (in the Gallery and Marble Hall) | |
Lecture Theatre |
Education Room |
|
|
Chair: João Esteves-Ferreira |
|
9.00 | Opening addresses, Leadership Talks | |
9.45 | Keynote speaker Jean Senellart (Systran)
The Neural Revolution in the Translation Industry – 3-Year Retrospective and Future Directions |
|
10.45 |
Zydroń has been responsible for the architecture of the essential word and character count GMX-V (Global Information Management Metrics eXchange) standard, as well as the revolutionary xml:tm standard which will change the way in which we view and use translation memory. Zydroń is also head of the OASIS OAXAL (Open Architecture for XML Authoring and Localization technical committee. Zydroń has worked in IT since 1976 and has been responsible for major successful projects at Xerox, SDL, Oxford University Press, Ford of Europe, DocZone and Lingo24. AI has garnered much hype over the past few years. Andrzej Zydroń provides a realistic definition of AI: what is intelligence; how can it be defined; what is the mathematical basis for intelligence, as well as detailing the theoretical limitations of AI and what is actually achievable. The presentation will detail the actual practical potential of AI as well as its limitations and pitfalls when human beings interact with AI systems. |
10.45-11.15 Silver Sponsor Session
interpreterQ Media Player interpreterQ Media Player was specifically designed to service both teachers and students interpreter training. The development was user-driven and inspired by real-life practices in leading interpreter schools, such as Newcastle University, Herriot-Watt University and University of Geneva. The presentation will show how smart features help in creating and evaluating sessions for simultaneous and consecutive interpreting. UK Business Director Since 2001, Dicken has been central to Televic’s activities within the interpreter Education sector, covering the UK, China, Asia Pacific and the USA. Specifically, within the field of high-technology solutions for Interpreter Training, Translation Studies and Language Education, Dicken has delivered and managed small and large-scale projects within Schools, Colleges and Universities around the world. As a founding member of the renowned ‘Talking to the World’ project, Dicken continues to collaborate with Universities on consultancy and commercial projects for interpreter training suites, (interpreterQ) and on new techniques, trends, and developments within Translation and Interpreter Education. Ed Tech Solutions Specialist, Belgium Bert Wylin has both an academic and a business profile. He works at the K.U.Leuven since 1993, first at the Continuing Education service, where he started up the distance education department and the Open University. Since 1996, he led the Education Innovation Center, leading to a growing expertise in integrating technology and the Internet in university education. In 2001, he founded the K.U.Leuven-spin-off company Telraam/Indie Education, developing and servicing educational multimedia and online projects. In 2008 Indie Education merged with Artec to become the actual Televic Education. Today, Bert Wylin leads the research and content divisions for assessmentQ and translationQ, Televic’s own online and offline testing and evaluation platforms. assessmentQ is the most used/spread educational technology in Flanders education and is a very reliable exam tool for the public and the private sector in Europe. |
11.15 | ![]() |
|
Chair: Ruslan Mitkov | ||
11.45 | Aleš Tamchyna (Memsource)
Applying AI to NT and MT In the translation industry, the disruptive effect of AI is not yet apparent. Machine learning/AI has traditionally been associated with one feature: machine translation (MT). It is true that with the recent advancements in neural MT, the output quality is inching closer to human translation. However, neural MT still makes serious mistakes and its quality can be upset by more complex sentences. More importantly, professional translation has different standards than simply passing for human translation; translations might require a specific style, consistent terminology, coherence across sentences and paragraphs, etc. But a translator’s goal is to convey the original meaning as closely as possible. They have to carefully navigate ambiguity and craft wording that best reflects the emphasis in the original text, ensuring that there can be no confusion about the meaning. Consider the severity of mistakes within legal texts or medical records; in marketing, a good translation can be the difference between a successful campaign and an international embarrassment. It’s clear that MT is not going to replace human translation anytime soon, if ever. |
11.45-13.15 Gold Sponsor Session Judith Klein (MA Information Science) has 20 years’ experience in language technology. She joined STAR Germany in 1999 where she works as an expert in support, training and consulting for STAR’s language technology tools. Her most recent interest lies in STAR’s MT technology. Before she came to STAR, she worked in the Language Technology department at the German Research Center for Artificial Intelligence (DFKI) in Saarbrücken. The Swiss Broadcasting Corporation (SRG) has taken a keen interest in expanding their professional translation environment to include machine translation (MT). Switzerland’s typical multilingualism is once more the driving force to introduce state-of-the-art language technology. The focus is on supporting the work of the translator. SRG’s range of topics covers many areas of life with different types of text and a great linguistic diversity. The demands placed on translations are correspondingly diverse. Due to the success of MT in very different areas, SRG is convinced that MT will also be of great value for all involved. Elke Fuchs, a German national, is a certified translator in English and Spanish. She has 19 years of experience in the translation industry working for STAR Group. Her focus is on technical support, software instruction and STAR Group product representation. She also consults with customers, providing guidance on the use of STAR Transit and helping them integrate Machine Translation into their workflow. The aim of alignment is synchronized data. Interactive alignment with Transit NXT and its Alignment mode is easy. Probable hits are marked and so the user does only need to concentrate on questionable segment pairs. Machine alignment does not need any interaction – you can interact – but you do not have to. Machine alignment creates value for outdated data. High volumes of data can be processed in order to use it as reference material.
|
12.15 | Margita Šoštarić, Nataša Pavlović and Filip Boltužić (Univ. of Zagreb) Nataša Pavlović is an associate professor in the Faculty of Humanities and Social Sciences, University of Zagreb, where she teaches translation theory and practice. She holds a PhD in Translation and Intercultural Studies from the University Rovira i Virgili in Tarragona, Spain. Her research interests include translation process research, translator education, and translation technology, in particular machine translation and post-editing. Margita Šoštarić is a recent graduate from the Faculty of Humanities and Social Sciences, University of Zagreb and currently works at Omega Software, a software development company in Zagreb. Her research interests span from the more theoretical approaches to language, such as cognitive linguistics, to the direct applications of language processing, such as machine translation.
Domain Adaptation for Machine Translation Involving a Low-Resource Language: Google AutoML vs. Fom-Scratch NMT System Despite the advances in machine translation achieved with neural models, adaptation of such systems for specialist domains remains a challenge. The problem is particularly acute when it comes to low-resource languages. Additionally, the computational resources and expertise needed to train neural models present barriers to entry for smaller translation companies and freelancers. In such cases, paid but affordable customization services such as Google Cloud AutoML might present a viable solution. In this study, domain adaptation using Google Cloud AutoML Translation is compared to a more traditional scenario, where several neural machine translation systems are trained from scratch using OpenNMT, an open-source toolkit for machine translation. The from-scratch systems are trained using a larger out-of-domain English-Croatian dataset and a smaller in-domain English-Croatian dataset comprised of medical texts. The same in-domain data are used to customize Google Cloud AutoML Translation. The performance of the systems is compared using automatic and human evaluation methods. The resources, skills and time necessary to set up the examined systems are also discussed. |
|
12.45 |
Despite the rich history of research into medical translation, there is a notable lack of empirical studies on the best workflow for this task, especially in a modern translation setting involving post-editing of machine translation. This pilot study was conducted in preparation for a large translation project of medical guidelines for laypeople from Dutch into French. It is meant to shed light on how medical post-editing is best handled. How do medical specialists (doctors) versus language specialists (translators) perform on this task? How can their respective strengths lead to the highest quality translation? To gain more insight into these questions, errors in the machine translation output of medical guidelines were annotated and labelled. Based on these annotations, the product of doctors’ and translators’ post-editing could be analysed and classified into necessary changes (mistakes that were correctly solved), underrevisions (mistakes that were not corrected during post-editing), overrevisions (new errors introduced during post-editing) and hyperrevisions (preferential changes made by the post-editor). The results of this small-scale research illustrate the complexity of the task and reveal some surprising findings (e.g., doctors sometimes struggle with domain-specific terminology, and translators appear to be less efficient because they introduce many hyperrevisions). |
|
13.15 | Lu![]() |
|
Chair: Juliet Macan | ||
14.00 | Rodolfo Maslias (European Parliament)
New Audiences for EU Terminology A short PowerPoint presentation followed by live surfing in the public websites termcoord.eu, YourTerm.org and in the (password protected) EU interinstitutional terminology portal EurTerm, focusing on the terminology management in the European Parliament, the cooperation between EU Institutions at central (EurTerm) and at language levels (wikis), the collaboration with the interpreters in the EU for terminology, the interoperability of the new version of IATE, the efforts of TermCoord for a new terminologist profile in the recruiting procedures in the EU Institutions, the connection of the EU and other terminology resources (like the EP’s GlossaryLinks) with the NMT and the post editing and quality control software, terminology projects with Universities with and for IATE, Master courses on terminology at the Universities of Luxembourg (36 hours), Savoie-Mont Blanc (21 hours) and Orientale Napoli (11 hours) and occasionally in many Universities (Vigo, Germersheim, ISIT Paris a.o.) and the new approach of “plain terminology” projects, adapted to communication needs and addressed to the civil society with the programme “Terminology without Borders” in several fields and in collaboration with specialised EU Agencies and International Organisations and with specialised departments of Universities in several European countries. |
14.00-15.00 Silver Sponsor Session Presentation and live demo of SDL Trados GroupShare 2020 (45 minutes) + Q&A (15 minutes) Laura Saunders-Calvert, Senior Customer Success Executive |
14.30 | Denis Dechandon (EPO), Maria Recort Ruiz (ILO) and Aniko Gerencser (EPO)
Previously in his last role, Denis was responsible for leading a service dedicated to the linguistic and technical support provided to translators, revisers, editors, captioners and subtitlers (Computer Assisted Translation, corpus management, formatting and layouting, machine translation and terminology). Additionally, he supervised the maintenance and development of tools and linguistic resources at the Translation Centre for the Bodies of the European Union. Committed to further changes and evolutions in these fields, Denis took over the role of InterActive Terminology for Europe (IATE) Tool Manager from May 2015 to August 2017. In his current role as Head of the Metadata sector of the Publications Office of the European Union, he is leading the activities in standardization (in particular: EuroVoc and registry of metadata) as well as intensely involved in the field of linked open data at the Publications Office of the European Union. Latest projects involve the development of synergies between several different stakeholders, such as EU institutions, agencies and bodies, international organisations and national public services. Maria Recort Ruiz is a philologist, translator and terminologist who works as Document Services Coordinator and Terminology Manager at the International Labour Organization in Geneva. She is responsible for the production and management of official documents, management of terminology work and the use of new CATT tools to improve working methods. She holds a Degree in Slavic Philology from the University of Barcelona, where she specialized in Russian and Polish Language and Literature, and Linguistics; a Master in French and Comparative Literature (19th-20th centuries) from the University of Montpellier, where she conducted research on the roman populaire at the beginning of the 20th century; and a Master in Specialized Translation from the University of Geneva. Before joining the ILO, she worked as a freelance translator and editor for international organizations and the private sector.
Since joining the Publications Office of the European Union she has been working in the field of metadata standardisation and linked open data management. Her particular area of responsibility is the co-maintenance of the EuroVoc multilingual thesaurus and its alignment with other controlled vocabularies. She currently works on the optimisation of the thesaurus management tool Vocbench3 which involves analysing users` needs and improving collaborative features. She is in charge of providing presentations, consultancy and trainings regarding the use of VocBench3 for EU institutions. In addition she strongly contributes to an on-going project that aims to achieve interoperability between controlled vocabularies by sharing common tools and formats for the creation, use and maintenance of vocabularies and taxonomies. Terminology: Towards a Systematic Integration of Semantics and Metadata |
|
15.00 | Jean-Francois Richard (Terminotix)
Terminotix is a software development company that helps linguistic services and language service providers increase performance by automating and enhancing the translation process. Terminotix helps translators, revisers, coordinators and terminologists increase their productivity by up to 50% with a complete suite of products designed and optimized for language professionals. Terminology Extraction as a Tool for MT Output Assessment and Improvement The present paper proposes the use of a Terminology Recall Index (TRI) calculated on retaining nominal groups’ frequencies and stemming info only. Though this paper proposes to demonstrate the utility of a TRI calculation between a human translated document and neural machine translated document, it also attempts to demonstrate that a broader use of the TRI calculation has many other surprising applications inside a linguistic service’s translation workflow. |
15.00-15.30 Silver Sponsor Session XTM Workbench Elizabeth Butters Business Development Manager |
15.30 | ![]() |
|
Chair: Ruslan Mitkov | ||
16.00 | Maria Stasimioti and Vilelmini Sosoni (Ionian University)
She holds a BA in English Language and Linguistics from the National and Kapodistrian University of Athens, an MA in Translation and a PhD in Translation and Text Linguistics from the University of Surrey. Her research interests lie in the areas of Corpus Linguistics, Machine Translation (MT), Cognitive Studies, Translation of Institutional Texts and AVT. She is a founding member of the Research Lab “Language and Politics” of the Ionian University and a member of the “Centre for Research in Translation and Transcultural Studies” of Roehampton University. She has participated in several EU-funded projects, notably TraMOOC, Eurolect Observatory and Training Action for Legal Practitioners: Linguistic Skills and Translation in EU Competition Law, while she has edited several volumes and books on translation and published numerous articles in international journals and collective volumes. Undergraduate Translation Students’ Performance and Attitude towards Machine Translation and Post-editing: Does Training Play a Role? In an effort to meet the demands in speed and productivity, while keeping the cost low, the translation industry has turned to Machine Translation (MT) and Post-Editing (PE). Nowadays, it is common practice to include MT in the translation workflow by using MT output as raw translation to be further post-edited by a translator (Lommel and DePalma, 2016). Yet, translators still approach PE with caution and skepticism and question its real benefits (Koponen 2012; Gaspari et al 2014; Moorkens 2018). In addition, attitudes to MT and PE seem to affect PE effort and performance (Witczak, 2016; Çetiner and İşisağ, 2019). Under that light, this study aims to investigate the attitudes and perceptions of undergraduate translation students towards MT and PE as well as their performance before and after they receive training in MT and PE. Questionnaires are used to capture their attitudes and perceptions, while a human evaluation of their post-edited MT output is used to assess their performance and the quality of the post-edited texts. The analysis reveals a change in the students’ attitudes and perceptions; they report a more positive attitude toward MT and PE, they are more confident and faster, while they avoid over-editing. |
16.00-17.30 Gold Sponsor Session
Terminotix is a software development company that helps linguistic services and language service providers increase performance by automating and enhancing the translation process. Terminotix helps translators, revisers, coordinators and terminologists increase their productivity by up to 50% with a complete suite of products designed and optimized for language professionals. |
16.30 | Christopher Gledhill and Maria Zimina (Univ. Paris-Diderot)
The Impact of Machine Translation on a Masters Course in Web Translation: From Disrupted Practice to a Qualitative Translation/Revision Workflow The introduction of technology into translation curricula is a complex task in terms of translation competences and their acquisition. Computer tools and MT directly affect trainee translators. This study investigates the impact of technology on students on a Master’s in Specialised Translation and Language Industries at Université Paris Diderot. We present the results of a teaching project “Website translation into English” which places strong emphasis on hands-on applications of MT. The aim of the project is to provide students with a semi-professional work experience in which they face real-life website translation problems. Students are expected to translate and revise webpages from French into English using a professional platform SystranLINKS. The first results of our study show that a more equipped translator’s workstation results in assisted but also disrupted translation practice, and requires additional learning/teaching time. Intensive practice of MT raises students’ awareness of the importance of a revision workflow, and gives students a broader understanding of translation quality. Our methodology involves the analysis of project reporting forms, which students write at the end of the course as a record of their learning experience. We examine both their explicit comments and their implicit metalanguage, in order to explore how they conceptualise MT. |
|
17.00 | Marion Kaczmarek and Michael Filhol (LIMSI)
He chose to focus on Sign Language for his PhD, the most exotic linguistic system he knew, and which he had been learning since high school. Largely still unknown to science and virtually absent in the NLP field, he addressed and proposed a formal description model of signs for Sign Language synthesis by 3D avatars. He defended his PhD in 2008 at Université Paris Sud (Orsay, France), and continued his research career on Sign Language processing. He stayed for a post-doctoral year at Gallaudet University (Washington, DC, USA), where all classes and services are accessible in Sign Language, and some of the most famous researchers on Sign linguistics are hosted. Back in France, he got his permanent researcher position at CNRS (the French national scientific research centre), where he kept working on the formal description and computer implementation of Sign Language. He eventually proposed AZee, a formal approach capturing all levels of discourse and capable of driving a 3D avatar to animate Sign from a combination of semantic operations. It is now used by the world leaders in Sign synthesis, as input for their animation platforms. While always improving and extending the coverage of the AZee approach, his research interests have grown to encompass more topics like graphical writing systems for Sign Language, or automatic and assisted text-to-Sign translation. Assisting Sign Language Translation: what Interface Given the Lack of Written Form and the Spatial Grammar? Computer-assisted translation (CAT) software offers tools for the translators to ease their tasks, and gain time as well as comfort. However, despite the growing need for Sign Language content, there has been no effort to equip Sign Language translation with CAT software. The problem we address here is the specification of such software. Sign Languages are visual and iconic, with grammar and discourse organisation, but also no written form. This is problematic when it comes to CAT, for it relies on editable written structures and the fact that the concatenation of the translated segments will result in the translation of the concatenated source segments (we call it the linearity assumption). In this paper, we explain that Sign Language cannot follow those rules. We address those differences by means of new adapted modules which would be more flexible, and by considering new tools based on professionals’ feedback towards their actual practice as well as the problems they encounter during the translation process. We will detail those results along with the presentation of how we envisage a sign language concordancer, and its database. |
|
17.30 | End of day 1 | |
20.00 | Networking Dinner (optional) |