Invited talks
Talk 1
Gregory Shreve
The Strange Attractions of Translation: Translation Performance as Complex-Adaptive System
Translation performance can be viewed from the perspective of the complex adaptive systems model (CASM). This model views a translational output as the result of the complex interaction of a variety of system agents interacting at various levels. The action outputs of the system, in this case translation results, are not entirely predictable from the system’s inputs. The translational system is sensitive to its initial inputs and feedback relations between agents can lead to unpredictable results due to dampening or amplifying effects. The system may exhibit emergent properties. Complex adaptive systems are dynamic, and the state of the system can shift dramatically under the influence of both internal and external factors. In this paper we look at how a complex-adaptive model can help explain some elements of the development of translation expertise and explain versioning phenomena in performance. In particular, we will look at the role of strange attractors in establishing certain stable states of the system, such as default translations, and how the notion of introducing new attractors can disrupt the translational performance system and cause it to settle into more desirable states. This latter notion may be useful in translation pedagogy.
About Gregory Shreve
Gregory Monroe Shreve is Professor Emeritus of Translation Studies at Kent State University and Adjunct Full Professor of Translation, Interpreting and Foreign Languages at New York University, where he has taught since 2011. The founding Director of the Institute for Applied Linguistics and past Chair of the Department of Modern and Classical Languages Studies at Kent State, Professor Shreve was instrumental in establishing one of the first comprehensive Translation Studies programs in the United States. Shreve’s research interests include text linguistics and translation, cognitive translation studies, translation expertise, empirical approaches to translation studies, and translation informatics. He is the co-author/co-editor of several books including (with Albrecht Neubert) Translation as Text, (with Joseph Danks) Cognitive Processes in Translation and Interpreting and (with Erik Angelone) Translation and Cognition.
Talk 2
Fabio Alves
Grounding Cognitive Translation Studies within a situated, distributed and extended approach
The study of translation as a cognitive activity goes by many names, including, among others, Translation Process Research (TPR), Cognitive Translatology (CT), Cognitive Translation Studies (CTS) and Cognitive Translation and Interpreting Studies (CTIS). These disparate denominations point, however, to a lack of disciplinary convergence and highlight the need for a common denominator in order to build a coherent framework so that one can strive towards the consolidation of a sub-discipline in the making. Drawing on Alves & Jakobsen (2020), this talk reflects on the commitments and goals which underlie the study of translation as a cognitive activity and introduces a unified approach which brings together the concepts of situated, distributed and extended cognition to plead for an SDE-approach in an attempt to ground the sub-discipline of cognitive translation studies.
About Fabio Alves
Fabio Alves is Professor of Translation Studies at Universidade Federal de Minas Gerais (UFMG), Brazil, and a Principal Investigator at the Laboratory for Experimentation in Translation (LETRA). He holds a PhD in Applied Linguistics from Ruhr-Universität Bochum, Germany, and has been a visiting professor at Universitat Autònoma de Barcelona, Spain; Copenhagen Business School, Denmark; University of Macau, China; and Universität des Saarlandes, Germany. He carries out experimental research in the field of cognitive translation studies with a focus on modelling translation task execution from a process-oriented perspective, translation expertise and human-computer interaction in translation.
Talk 3
Masaru Yamada
Post-editing and a Sustainable Future for Translators
Regardless of its pros and cons, post-editing of machine translation (MTPE) has established itself as a common practice in the translation industry because of its potential to increase productivity and reduce costs. However, the working conditions post-editors experience may not necessarily be optimized under the pressure of achieving these outcomes. This relationship is sometimes unevidenced by rigid empirical data but may be captured through translation process research (TPR).
The purpose of this presentation is to reconsider a social, practical application of TPR in supporting a sustainable future for translators with the integration of new technologies that significantly impact the translation workflow, such as MTPE. The author will review literature on MTPE, including TPR research on MTPE as well as relevant literature on social aspects of professional translators/post-editors, before sharing the outlook of his current interdisciplinary research projects on developing sustainable translation practices in MTPE.
About Masaru Yamada
Masaru Yamada, Ph.D., is professor in the Faculty of Foreign Language Studies at Kansai University in Osaka, Japan. He specializes in translation process research (TPR), including human-computer interaction (HCI) and machine translation plus post-editing (MTPE), and translation in language teaching (TILT). His publication includes “The impact of Google Neural Machine Translation on Post-editing by student translators”, JoSTrans 31.
Talk 4
Mirko Plitt
Towards the transparent translation supply chain?
A transparent translation industry without intermediaries has long been predicted yet not materialized. Now, a number of trends within and outside the industry seem to finally converge towards this promise: the relative success of translation “marketplaces”, the growing recognition of the value added by humans in services previously said to be made redundant by Artificial Intelligence, learnings from attempts to apply the blockchain to the translation industry, the maturing of the market for translation management technology, the increasing prevalence of agile processes among buyers of translation services, the popularity of volunteer-based translation, the juxtaposition of emerging and highly developed translation markets, the popularity of volunteer-based translation, and possibly even the COVID-induced generalization of internet-based remote working, of which the translation industry has been a very early adopter. We will discuss these trends and their impact on the industry practice today, and attempt to propose how the sector can help shape its future practice to the benefit of all actors involved.
About Mirko Plitt
A computational linguist by training, Mirko has contributed to language technology innovation in a wide variety of contexts, from small startups and nonprofits to large enterprises and international organizations. An early achievement of his was the prototype of a translating fax machine, another one of the first implementations of MT in product support. Later, he became known for his pioneering work on the use of MT as a translator productivity tool. Most recently, Mirko was the head of technology at Translators without Borders, where he created the Kató translation environment used by more than 40,000 volunteer translators. In 2018, the spoken translation memory system he developed at TWB received the TAUS Game Changer Innovation Award. Mirko now works on various language technology projects for commercial clients as well as nonprofit partners.
Talk 5
Spence Green
Practical Experience with Interactive Machine Translation
Interactive machine translation (MT) is a mixed-initiative approach to translation where an intelligent system suggests prefix-constrained completions based on the translation the translator has already typed. This technique has been shown to have promising results in small-scale experiments, and can have benefits over post-editing. However, interactive MT has yet to be studied at scale on actual commercial translation systems. This talk presents a series of in-the-wild experiments on Lilt, a commercial language service provider and web-based translation platform that implements the interactive MT paradigm. We use A/B testing and analyze data logged from translator activity to help us quantify how professional translators interact with interactive MT. We will first present a breakdown of what activities translators spend their time on when using interactive MT. We will then analyze how much translators make use of our machine translation suggestions, and how much typing effort is saved. We then discuss the latency-quality tradeoff inherent in interactive MT, and analyze how fast the system needs to be, in terms of latency, to be useful to translators. Finally, we investigate effects of interactive MT on the quality of resulting translations, and the overall time spent by translators.
About Spence Green
Spence Green is Lilt’s Co-Founder and CEO. Prior to founding Lilt, Spence was a fellow at XSeed Capital and a software and research intern at Google, where he worked on Google Translate. He received a PhD and MS in Computer Science from Stanford and a BS from the University of Virginia. He has published papers on machine translation, language parsing, and mixed-initiative systems and given talks on translator productivity.
Talk 6
Alon Lavie
COMET - a Neural Framework for State-of-the-Art MT Evaluation Models
Delivery of high-quality Machine Translation (MT) is only possible with reliable evaluation metrics to inform modelling and system development. The translation workflows we develop at Unbabel require highly-adapted MT systems which are regularly retrained to continuously deliver customer-specific, accurate translations. Unfortunately, with current state-of-the-art neural MT systems, traditional metrics such as BLEU and METEOR have been shown to poorly correlate with human judgments, and in particular, poorly distinguish between fine-grained accuracy distinctions of top performing MT models. This can result in misinformed MT development decisions that affect the quality of translations for our customers.
To address this challenge, we recently developed COMET - a new neural-based framework for training automatic MT evaluation models that is demonstrated to exhibit new state-of-the-art levels of correlation with human judgments. Our framework leverages recent breakthroughs in cross-lingual pretrained language modeling resulting in highly multilingual and adaptable MT evaluation models that exploit information from both the source input and a target-language reference translation in order to more accurately predict MT quality. We showcase our framework by training and evaluating COMET models for three different types of human judgments: Direct Assessments, Human-mediated Translation Edit Rate (HTER) and Multidimensional Quality Metric (MQM). Our models achieve new state-of-the-art performance on the WMT 2019 Metrics shared task and are sensitive to fine distinctions typical of high-performing MT systems.
In this presentation we present an overview of the COMET framework and highlight its capabilities through extensive assessments of the COMET models we have trained, their correlation with human judgments of translation quality, and their utility in practice for evaluating and contrasting MT models developed at Unbabel. We outline the various use cases for automatic MT evaluation metrics at Unbabel, and contrast our assessment of COMET with commonly used traditional metrics such as BLEU across multiple content types and language pairs. We then describe our developed methodology for defining the criteria for where COMET can be adopted as a substitute for existing metrics and/or human MQM annotation.
About Alon Lavie
Dr. Alon Lavie is currently the VP of Language Technologies at Unbabel, where he leads and manages Unbabel’s US AI lab based in Pittsburgh, and provides AI strategic leadership company-wide. Prior to joining Unbabel, Alon was a senior manager at Amazon, where he led and managed the Amazon Machine Translation R&D group in Pittsburgh. Prior to that, Alon was the co-founder, President and CTO of Safaba, an MT technology company that was acquired by Amazon in 2015. For almost 20 years (1996-2015) Alon was a Research Professor at the Language Technologies Institute at Carnegie Mellon University, where he currently serves as an adjunct professor. Alon was President of the International Association for Machine Translation (IAMT) (2013-2015). Prior to that, he was president of the Association for Machine Translation in the Americas (AMTA) (2008-2012), and was General Chair of the AMTA 2010 and 2012 and MT Summit 2013 conferences.
Talk 7
Natasha Tokowicz
The Translation Ambiguity Disadvantage in Language Processing: The Influence of Context and Proficiency
Many words in a given language can be translated in more than one way into another language (e.g., Tokowicz, 2014). For example, the German word “Kiefer” can be translated into English as both “pine” and “jaw”. This mis-mapping of translations across languages, known as “translation ambiguity”, causes difficulty in language processing and language learning. In this presentation, I will discuss how context influences the presence of the translation-ambiguity disadvantage in language processing (e.g., Eddington & Tokowicz, 2014; Tokowicz et al., 2019). I will also discuss how translation ambiguity affects language processing in bilinguals of various proficiencies, including individuals who were raised bilingual (e.g., Boada et al., 2013).
About Natasha Tokowicz
Natasha Tokowicz earned an M.S. and Ph.D. in Cognitive Psychology from The Pennsylvania State University. She completed post-doctoral fellowships at Carnegie Mellon University and the University of Pittsburgh prior to beginning a faculty appointment at the University of Pittsburgh in 2004. She is currently Associate Professor of Psychology and Linguistics and Research Scientist at the Learning Research and Development Center at the University of Pittsburgh. Her research combines behavioral and cognitive neuroscientific methodologies to address questions about adult second language learning and bilingualism. Her book, Lexical Processing and Second Language Acquisition, was published by Routledge in 2014.
Talk 8
Antonio Toral
Human Parity and Translation Universals in Machine Translation
Machine Translation (MT) has improved dramatically since the recent adoption of neural networks (e.g. Badahnau et al., 2014; Vaswani et al., 2017). In fact, there have been recently claims of human parity and even of super-human performance (Hassan et al., 2018; Barrault et al., 2019), although there are potential issues in the way these MT systems were evaluated. Due to this, I deem it timely to (i) reassess these claims using state-of-the-art human evaluation procedures and to (ii) go beyond the common practices used to date for the evaluation and analysis of translations produced by MT systems, for which I propose to borrow concepts from Translation Studies, namely a set of translation universals and laws of translation: simplification, normalisation and interference. In this presentation I will provide an overview of my recent and on-going work on these two intertwined research topics.
About Antonio Toral
Antonio Toral is assistant professor in Language Technology at the University of Groningen (The Netherlands). Previously, he held research positions at Dublin City University (Ireland), the Institute for Computational Linguistics (Italy) and the University of Alicante (Spain). He is the author of over 100 peer-reviewed publications, was awarded the best paper at MT Summit 2019 and was the coordinator of Abu-MaTran, a Marie Curie Industry-Academia project flagged as a success story by the European Commission. His research interests include the analysis of translations produced by machines and humans and the application of Machine Translation to literary texts and under-resourced languages.
Talk 9
Maeve Olohan
Translation Practice, Technology and Power
Practice theory focuses conceptually and analytically on social practices and their constituent elements, including human bodies, technologies, practical know-how and symbolic meanings (see Schatzki 1996, Reckwitz 2002, Shove et al. 2012). In Translation and Practice Theory (Olohan 2021) I consider some benefits of reconceptualising translation as a practice, examining how the translation practice is constituted and how its elements, and the practice itself, change over time. The practice-oriented approach also enhances our understanding of connections between translation and other practices.
An aspect of practice that has been given relatively less research attention concerns the tensions, contradictions and power imbalances that keep practices in flux (Nicolini and Monteiro 2017). Issues of interest here for the translation practice, particularly in the context of today’s technologies, include the questions of what activities are favoured by a particular configuration of the practice, and what alternatives could emerge from alternative configurations. Who is empowered or disempowered by particular configurations of the translation practice? In this paper I respond to calls for analyses of practice to be integrated with theories of power when we consider translation practices and the constituent role of technologies.
About Maeve Olohan
Maeve Olohan is Co-Director of the Centre for Translation and Intercultural Studies, University of Manchester, where she delivers courses on the MA in Translation and Interpreting Studies and supervises doctoral candidates. She is author of Translation and Practice Theory (2021), Scientific and Technical Translation (2016) and Introducing Corpora in Translation Studies (2004), and has published widely on translation from sociological, historical, linguistic, methodological and pedagogical perspectives. She is Associate Editor of Translation Spaces (John Benjamins) and Series Editor for Routledge’s Translation Theories Explored monograph series.
Talk 10
Andy Way
The Effect of NMT on Translators and the Translation Process
With recent improvements in the quality achievable by MT engines, we have seen an accompanying rise in the amount of hype surrounding their capabilities. As with previous paradigm shifts, this has led to increased uncertainty among translators about their role in future translation pipelines. In this talk, we will present the basic neural MT model, and demonstrate what it can and still cannot do well. We will underline the continued vital role played by human translators in the translation process, and present new roles that are emerging which can only be filled by qualified human translation experts.
About Andy Way
Andy Way is Professor of Computing at Dublin City University and Deputy Director of the €100 million ADAPT Centre for Digital Content Technology. He has over 350 peer-reviewed papers, most of which are on MT. He was President of the EAMT from 2009-15, and President of the IAMT from 2011-13. He has been editor of the Machine Translation journal from 2007. He received the IAMT Award of Honour in 2019.
Round Table
Translation in Transition: Implications, extensions and applications for the workplace, pedagogy and research
Machine translation and artificial intelligence are bringing about a sea change in the language industry in terms of the scope and contour of project workflows, the roles of those engaged in them, and the manners in which their performance is assessed. Translation is truly in transition to an unprecedented extent. This roundtable will focus on discussing the implications of these changes in terms of their impact on the workplace, translator pedagogy, and future directions of empirical research.
Format:
Each speaker will prepare a brief statement (2 to 3 minutes) in which they share their general thoughts on the topic. This will be followed by a Question/Answer session based on questions asked by those in attendance.
Moderator: Isabel Lacruz
Participants:
Alon Lavie
Antonio Toral
Fabio Alves
Gregory Shreve
Katarzyna Stachowiak-Szymczak
Lucas Nunes Vieira
Maeve Olohan
Spence Green