(All times are GMT, see the programme here)
Fabio Alves (Federal University of Minas Gerais, Belo Horizonte)
Translation process research at the interface with human-computer interaction: challenges and perspectives
Date: 6th July, 6:30-7:30pm
Starting in the mid-1980s with studies which solely used think-aloud protocol data, translation process research (henceforth TPR) has grown to incorporate large data sets, robust statistics and computational tools in its research agenda. TPR has drawn on the triangulation paradigm (Alves 2003) as the methodological basis of an interdisciplinary approach and investigated segmentation patterns and translation units, instances of peak performance and sources of external support in translation, among several other topics. Focusing on user activity data and with the aid of data repositories such as the CRITT TPR-DB (Carl, Schaeffer and Bangalore 2016), TPR has provided insights into real-time translation task execution. Recently, advances in machine translation systems have placed the focus on human-computer interaction (henceforth HCI) and contributed to expanding the TPR agenda in a new direction. According to O’Brien (2020), the merging of translation memories and machine translation, as well as the advent of adaptive and interactive neural machine translation systems and the use of multimodal input, have had an impact on the act of translating and, consequently, on the TPR agenda. Building on these trends, this talk will discuss the challenges and perspectives that lie ahead for TPR at the interface with HCI and the prospects of computationally modelling the act of translating as a dynamic cognitive activity.
Lynne Bowker (University of Ottawa)
Machine translation use outside the language industries
Date: 7th July, 7:30-8:30pm
Since the release of the first free online machine translation systems around 15 years ago, machine translation (MT) has been “in the wild”, meaning it is available to users outside the language industries. However, most user-related investigations of MT focus on how this technology is used by language professionals, such as translators. As the user base outside the language professions grows, it is worth understanding more about who is using MT and why as this could help to inform future developments of the tools. Based on my experience of teaching MT to non-translation students in three different university-level contexts in the 2020/2021 academic year, I’ll share some insights about who these users are, why they are using MT, and how satisfied they are with the results.
Sabine Braun (University of Surrey)
Advancing research on distance interpreting for the post-pandemic age
Date: 6th July, 5:00-6:00pm
The Covid-19 pandemic has caused a ‘seismic’ shift in the world of professional interpreting: A large proportion of interpreting assignments has been—and continues to be—delivered online. This has sometimes come with an implicit assumption on the part of the users of interpreting services that the ‘new normal’ of online communication and homeworking have made the transition to distance interpreting easier. Whilst there are obvious benefits, previous research on distance interpreting has also shown that this modality of interpreting entails many difficulties for interpreters including increased fatigue, sometimes marked by a decline in output quality, a lack of rapport with the clients, and/or a deterioration of the communicative dynamics. The broader implications for the quality and effectiveness of the communication in situations of distance interpreting—including, for example, access to justice, healthcare and other public services for linguistic-minority populations—have yet to be fully addressed. Furthermore, before the pandemic, distance interpreting mostly involved either interpreting offsite for clients who were co-present or interpreting for individual remote participants in a meeting, court hearing or other situations. The many new and untested configurations of distance interpreting that have emerged during the pandemic (fully virtual meetings, hybrid video/telephone meetings, types of court hearing conducted online, varying quality of internet connections etc.) have led interpreters to accept working with technologies where the speed of adoption has sometimes side-stepped safeguards introduced through training and regulation. This situation has raised many new research questions. Among them are to what extent adaptation (of interpreters and other participants) is possible, which conditions are (not) conducive to adaptation, how interpreters can be supported in distance interpreting assignments, and how resilient approaches can be achieved in the long term. In this presentation I will review the outcomes of previous research on distance interpreting in light of their relevance for the practice of distance interpreting as it has evolved since the beginning of the pandemic, and identify those aspects that require further or renewed attention in research.
Gloria Corpas Pastor (University of Malaga)
Interpreting and technology: is the sky really the limit?
Date: 7th July, 4:00-5:00pm
This talk will revolve around language technologies applied to interpreting. Nowadays there is a pressing need to develop interpreting-related technologies, with practitioners and other end-users increasingly calling for tools tailored to their needs and their new interpreting scenarios. With the advent of new technology, interpreters can work remotely, deliver interpreting in different modes and contexts, on many devices (phones, tablets, laptops, etc.), and even manage bookings and invoice clients with ease. But interpreting as a human activity has resisted complete automation for various reasons, such as fear, unawareness, communication complexities, lack of dedicated tools, etc.
Several computer-assisted interpreting tools and resources for interpreters have been developed, mainly terminology management tools, corpora, and note-taking applications, but they are rather modest in terms of the support they provide. In the same vein, and despite the pressing need to aiding in multilingual mediation, machine interpreting is still under development, with the exception of a few success stories so far.
In this talk, I will present recent R&D projects on interpreting technologies in action. The first one is a speech-to-text system for automating communication of English and Arabic speaking patients in Spanish hospital triage scenarios at A&E services (in progress). The second one is already close to completion. It comprises a suite of NLP-enhanced tools and resources for interpreters and trainees, including but not limited to, terminology tools, corpora building and processing, automatic glossary building, automatic speech recognition and training tools. Final discussion will go back to the idiom included the title of this talk.
Elena Davitti (University of Surrey)
Hybrid practices for real-time interlingual communication via speech recognition: the SMART project
Date: 7th July, 10:30-11:30am
Technologies are changing the landscape of translation and interpreting, blurring the boundaries of traditional practices and opening up the new, hybrid workflows characterised by different degrees of human-machine interaction. On the one hand, the proliferation of live multimedia content, compounded by the rapid move to online/streaming communication in the ‘new normal’, accentuates the demand for such content to be made accessible in real-time, and for a wider audience, across sensory, linguistic and cognitive boundaries. On the other hand, integrating technologies like speech recognition and machine translation in traditional workflows for real time interlingual communication is raising questions in relation to the fitness-for-purpose of the output of increasingly (semi-)automated and hybrid workflows and the place and role of human input in such processes
This talk will provide an overview of human-centric and human-in-the-loop workflows for real time speech-to-text transfer from one language to another, with a particular focus on interlingual respeaking as a the most human-centric practice for live interlingual subtitling. Interlingual respeaking is a speech recognition-based method that relies on human-machine interaction, where the human listens to live input and simultaneously translates it into a target language (with added oral punctuation) to a machine (speech recognition software) that turns it into written text displayed on a screen. As a form of ‘simultaneous interpreting 2.0‘, it relies on advanced cognitive abilities, interpersonal traits and procedural skills across various domains.
Insights on current needs for this service will be drawn from the SMART project (Shaping Multilingual Access through Respeaking Technology), a project funded by the Economic and Social Research Council UK (ES/T002530/1, 2020-2022), which aims to explore how to effectively upskill language professionals from different walks of life to become interlingual respeakers. This talk will discuss the multi-method, multifactorial design developed by the study, which is currently in its data collection stage, to explore the human variables involved in learning and practising interlingual respeaking. Emphasis will be placed on the interdisciplinarity of the approach adopted, which is informed by the expertise not only of leading experts on respeaking and interpreting, but also psychologists who will offer important cognitive insights into the intricacies of the process. Instead of using technological determinism as a starting point to study human-machine interaction (technology-centred perspective), the study uses human performance as a starting point to gain increased awareness of the challenges involved in the process and ultimately lead to more responsible human-machine interaction.
Stephen Doherty (University of New South Wales, Sydney)
Psychological and cognitive advances in translation and interpreting technologies
Date: 6th July, 7:00-8:00am
Technological advances have led to unprecedented changes to translation and interpreting (see Doherty, 2016), including how we access and use translation and interpreting technologies for a diverse and growing range of professional and personal activities at local, national, and international levels. Previous research on translation and interpreting technologies has yielded a wealth of evidence to advance our understanding and usage of these technologies in addition to making them more visible and accessible. Of particular significance amongst this growing body of work is the use of eye tracking in exploring and understanding the psychological and cognitive aspects of translation and interpreting technologies by analysing our eye movements as we interact with these technologies.
In this paper, I will consolidate this body of work by presenting a critical review of the empirical studies of translation and interpreting technologies which have employed eye tracking, including my own recent work in the HAL Language Processing Lab at the University of New South Wales. I will categorise previous research into areas of application, namely: computer-assisted translation tools, quality assessment of machine translation, post-editing machine-translated output, audio-visual translation, and interpreting. In closing, I will discuss the strengths and limitations of eye tracking in such studies and outline avenues for current and future research.
Florian Faes (Slator)
Inside the USD 25 Billion Global Language Industry
Date: 6th July, 1:00-2:00pm
- Market overview and key players
- Follow the money. Where investors play their bets in translation and localization
- How the language Industry is pioneering the Expert-in-the-Loop Model
- Educating a new generation of language industry professional
- Beyond Machine Translation: The rapid rise of generative language technologies
Marcello Federico (Amazon)
Challenges and Progress in Automatic Dubbing
Date: 6th July, 3:00-4:00pm
Automatic dubbing (AD) is an extension of speech-to-speech translation such that the resulting target speech is carefully aligned in terms of duration, lip movements, timbre, emotion, and prosody of the original speaker in order to achieve audiovisual coherence. Dubbing quality strongly depends on isochrony, i.e., arranging the translation of the original speech to optimally match its sequence of phrases and pauses. In my talk, I will overview ongoing research on AD at Amazon, while focusing on the following aspects: verbosity and style control of machine translation, and prosodic alignment. Controlling the output length of MT is crucial in order to generate utterances of the same duration of the original speech. Style control, on the other hand, aims at generating translations that are stylistically coherent. Finally, the goal of prosodic alignment is to segment the translation of a sentence into phrases and pauses of the same duration of the original phrases. Along my talk, I will present experimental results and demo videos on four dubbing directions – English to French, Italian, German and Spanish.
Stephanie Labroue (Systran)
Monetize your linguistic assets: Train your own MT Model !
Date: 7th July, 11:30am-12:30pm
Quality of Machine Translation has never been higher and translation customization is now made easy by the learning capacity of neural translation engines. Training an NMT engine is becoming a new talent, and requires the right tools, high quality data and industry knowledge.
SYSTRAN provides training tools to enable worldwide linguistic experts to produce their own expert NMT models, either for their own use or to sell them to professional users through the SYSTRAN Marketplace.
During this talk, the following questions will be addressed:
- What is the process to train a translation model and what are the requirements to become a SYSTRAN Marketplace “Trainer”?
- What is the translation model monetization model and the benefits for the trainers?
- What are ensured linguistic data property and model ownership?
Adam LaMontagne (RWS)
Getting meta with MT metadata: how to extract value from the process of MT processing
Date: 6th July, 9:00-10:00am
Machine translation is a data-driven discipline. As such, before we can even get started deploying machine translation into our localization workflows, we focus on collecting information about the MT use case(s), linguistic data to train and evaluate MT, and financial data to calculate MT ROI. But what happens to the data that is produced passively as a result of the inclusion of machine translation? In this session, we will discuss what this metadata is, as well as some ideas of what can be done with it and what value it can add to MT-enhanced pipelines and to the business as a whole.
William D. Lewis (University of Washington and Microsoft Translator)
Speech Translation in Education: Use and Impacts
Date: 7th July, 3:00-4:00pm
Most education in the United States is conducted in English. Educators present in English, have course materials in English, and communicate with students and parents in English. For schools in regions hosting large immigrant populations, English-only instruction is not adequate since it leaves students who do not know English behind. The numbers of non-English speaking students in the US are staggering: in 2015, 9.5 percent of the student population in the US, some 4.8 million students, were listed as English Language Learners (ELL)—meaning that the dominant language spoken at home was not English—and some school districts in the country have more than 100 languages spoken by students and parents at their schools. Schools have increasingly looked for technological solutions to break down language barriers and make education more accessible. With the recent significant improvements in Speech Recognition and Machine Translation technologies, it is now possible lecture in one language and have students follow along in another. Further, the technology offers access to parents who may not speak English, say, at Parent-Teacher conferences or 1:1 meetings, and gives them the ability to be involved in their children’s education where they might otherwise be blocked. Although the technologies aren’t perfect, they are able to bridge the language gap and provide a means for educators to communicate more effectively with students and parents, and improve learning outcomes.
André Martins (Unbabel)
Human-Aided Translation and Quality Estimation
Date: 6th July, 4:00-5:00pm
Despite the amazing progress in machine translation (MT) quality in the last years, off-the-shelf MT systems in the real world still struggle too often with critical mistakes, inadequate translation of out-of-domain content, and overly literal translations lacking a “human touch.”
In this talk, I will describe the Unbabel translation pipeline, which combines machine translation and a community of human post-editors, intermediated by quality estimation (QE). This leads to a system that “knows” when MT is not good enough and a human is needed, ensuring high translation quality while reducing human translation effort to a bare minimum. I will describe other AI components, including automatic editor evaluation and representation (Translator2Vec), and an automatic MT evaluation metric (COMET), trained on multi-dimensional quality metric (MQM) annotations. The human post-edits and the MQM quality annotations are an organic part of our pipeline, enabling continuous improvement of the MT and QE systems through the incorporation of this human feedback — translations get better and better with more data, with less and less human intervention.
Some of Unbabel’s technology has been released as open source software and is widely used by the research community, including a framework for quality estimation (OpenKiwi) and MT evaluation (COMET).
Konstantin Savenkov (Intento)
Machine Translation Today and Tomorrow: A Retrospective and a Look at the Future
Date: 7th July, 5:00-6:00pm
Recent technology advances bring Machine Translation to its maturity, enabling outstanding performance across a broad range of use-cases. In this talk, we discuss what makes machine translation excel for a particular project and what it takes to turn the productivity gains into monetary value and competitive advantage. Also, we will touch the remaining obstacles that may MT perform poorly and available options to overcome them.
Rico Sennrich (University of Zürich)
How Contextual is Neural Machine Translation?
Date: 7th July, 8:00-9:00am
Contextual knowledge is of central importance for machine translation to resolve ambiguities and ensure consistency in reference, formality, etc. In this talk, I discuss the capability of neural machine translation (NMT) to take into account contextual information. I first show results in word sense disambiguation showing the possibility of integrating linguistic knowledge to address this problem, but also the ability and limitations of models learning from raw text data. I then shift to the challenge of taking into account wider context beyond the current sentence. While even simple model architectures are promising for this, the relative lack of document-level parallel data, and the insensitivity of standard metrics to wider context, hamper the development of systems that go beyond the current sentence. I discuss evaluation with contrastive test sets to obtain targeted results for contextual phenomena, and the use of automatic post-editing to increase translation consistency with only monolingual data.
Elsa Sklavounou (RWS)
What a future talent would say about translation automation?
Date: 6th July, 10:30-11:30am
RWS inspires young talents to start their journey in the localization industry by focusing on developing their individual expertise and business. Through its Campus, RWS commits to contribute to the community skilling, reskilling, and upskilling ensuring that young talents are given the best possible offset for starting their own, personal journey in the localization industry embracing technology. In this session we derive insights from the future talents vision on translation automation in Enterprise data management, protecting ideas and intellectual property.
Josef van Genabith (German Research Centre for Artificial Intelligence)
MMPE Multi-Modal Interface Support for Post-Editing
Date: 7th July, 12:30-1:30pm
Machine translation is increasingly becoming part of human professional translation workflows, where raw translation outputs produced by a machine are reviewed, corrected and certified by human translation professionals. Compared to translation from scratch, this changes text interaction from production to revision and review. This raises the question whether traditional text interaction modalities such as keyboard and mouse are best suited to modern postediting (PE) tasks? In our research, we investigate the use of touch, pen, gestures, and speech in addition to keyboard and mouse, as well as combined interaction modalities. We also explored the use of language technologies such as translation quality estimation in PE and automatic PE.
- Rashad Albo Jamara, Nico Herbig, Antonio Krüger and Josef van Genabith. “Mid-Air Hand Gestures for Post-Editing of Machine Translation”. (to appear) ACL-IJCNLP 2021
- Herbig, N., Düwel, T., Pal, S., Meladaki, K., Monshizadeh, M., Krüger, A., van Genabith, J. “MMPE: A Multi-Modal Interface for Post-Editing Machine Translation”. ACL 2020
- Herbig, N., Pal, S., Düwel, T., Meladaki, K., Monshizadeh, M., Hnatovskiy, V., Krüger, A., van Genabith, J. “MMPE: A Multi-Modal Interface using Handwriting, Touch Reordering, and Speech Commands for Post-Editing Machine Translation”. ACL 2020
- Pal, S., Xu, H., Herbig, N., Naskar, S., Krüger, A., van Genabith, J. “The Transference Architecture for Automatic Post-Editing“. COLING 2020
- Herbig, N., Pal, S., van Genabith, J., Krüger, A. “Multi-Modal Approaches for Post-Editing Machine Translation”. CHI 2019
Rosanna Villani (European Central Bank)
The changing profile of the translator profession at the European Central Bank – how translators mastered the digital transformation
Date: 7th July: 9:00-10:00am
The evolution of the translator profile at the ECB has seen a marked acceleration in the last years. Staff of the linguistic services moved from their traditional set of tasks – translation, editing and proofreading, often done on print-ups or in plain Word documents – to dealing with a multitude of functions, such as project management, procurement, recruitment and training, and lately cultural mediation. But the real game changer in this transformation has been the quantum leap that technology applied to translation (CAT and MT) has enabled in complex organisational structures. This brought about significant challenges in terms of people and change management, but also presented huge opportunities and the chance to broadcast the ECB’s language services as a cutting edge professional unit within the organisation. The journey is still ongoing, but we are not in unchartered waters anymore and are currently expanding our interest in new technologies to widen accessibility of communication even beyond the pure translation function.