MT and AI: Probing Near- and Medium-Term Impacts
Organized by: Program Committee – Commercial MT Users and Translators Track
September 19, 09:30-11:00, Toyoda Auditorium Hall
Sharon O’Brien (DCU)/Michel Simard (NRC)
Pascale Fung (Hong Kong University of Science and Technology), Tony Hartley (Rikkyo University, Tokyo), Chris Wendt (Microsoft), Kayoko Takeda (Rikkyo University, Tokyo), Olga Beregovaya (Welocalize / AMTA President), and Minako O’Hagan (University of Auckland)
According to Stephen Hawking, “this is the most dangerous time for our planet” (Hawking, 2016). Hawking was referring to the development and deployment of Artificial Intelligence (AI). He states: “The automation of factories has already decimated jobs in traditional manufacturing, and the rise of artificial intelligence is likely to extend this job destruction deep into the middle classes, with only the most caring, creative or supervisory roles remaining.” Though there is disagreement on the impact that AI will have, there is acknowledgement that AI will change the world we currently live in and this is coupled with considerable worry about redundancy, privacy and control. MT is a form of AI and is likely to improve as AI techniques improve in general. However, there has been little discussion in the realm of Machine Translation about the near- and medium-term impacts of MT. We believe it is time for the ‘awkward public conversation’ (Stilgoe and Maynard, 2017) around MT and AI. Inspired by the Asilomar AI principles (see: https://futureoflife.org/ai-principles/) we pose some questions to our panel members that will commence an important discussion on the future ramifications of AI-driven MT. The panel members will represent different stakeholders in the MT community, commercial developers, R&D, professional translators and translation studies. Panelists have been asked to address a number of probing questions that centre on the topics of shared benefit, shared prosperity, and accountability.
Pascale Fung is a Professor at the Department of Electronic & Computer Engineering at The Hong Kong University of Science & Technology. She is an elected Fellow of the Institute of Electrical and Electronic Engineers (IEEE) for her “contributions to human-machine interactions”, and an elected Fellow of the International Speech Communication Association for “fundamental contributions to the interdisciplinary area of spoken language human-machine interactions”. She is the founding Director of InterACT at HKUST, a joint research and education center with Carnegie Mellon University, University of Karlsruhe (TH) and Waseda University. She co-founded the Human Language Technology Center (HLTC). She is an affiliated faculty with the Robotics Institute and the Big Data Institute at HKUST. She is the founding chair of the Women Faculty Association at HKUST. She is on the Global Future Council on AI and Robotics, a think tank for the World Economic Forum, and blogs for the online WEF publication Agenda.
Tony Hartley trained and worked as a translator and conference interpreter, retrained in cognitive science and became a researcher in natural language generation and MT; established graduate Centre for Translation Studies at Leeds, which was involved in many EU and UK projects with academic and industrial partners; having first introduced MT (Weidner MicroCAT) into the translator training curriculum in 1986, he continues to teach translation technologies to graduate students at Rikkyo.
Chris Wendt graduated as Diplom-Informatiker from the University of Hamburg, Germany, and subsequently spent a decade on software internationalization for a multitude of Microsoft products, including Windows, Internet Explorer, MSN and Bing – bringing these products to market with equal functionality worldwide. Since 2005 he is leading the program management team for Microsoft’s Machine Translation development, responsible for Microsoft Translator services, including Bing Translator, Skype Translator, the Translator API and Microsoft’s self-service MT customization system, the Translator Hub. Chris is responsible for the design of these products, connecting Microsoft’s research activities with its practical use in services and applications That includes Microsoft’s own applications, but more importantly third party applications and enterprise use. Chris’ goal in life is breaking down language barriers between the humans inhabiting earth. He believes it’ll take us a while to get there, but that we are moving in the right direction. Slowly, but occasionally moving a bit faster. We are right in the middle of one of these occasions. He is based at Microsoft headquarters in Redmond, Washington, USA.
Kayoko Takeda is Professor of Translation and Interpreting studies in the College of Intercultural Communication at Rikkyo University in Tokyo. Her research focuses on the history, pedagogy and sociocultural aspects of interpreting and translation.
Olga Beregovaya has over 20 years of experience in the translation and localization industry. She started her career managing domain adaptation projects for rule-based MT systems. Later she managed Language Quality Department at Autodesk, where she oversaw the implementation of LQA automation tools and selection of 3rd party MT solutions. In 2008 Olga moved to PROMT where she was in charge of the company’s enterprise server product development and deployments. Since 2011, Olga is driving the Language Technology and Services strategy for Welocalize. Olga is in charge of the language services, which includes development of in-house content curation tools, Welocalize in-house MT engines and deployment of various partner MT and NLP solutions. Olga is a frequent presenter at various industry conferences and is at the moment a president of the American Machine Translation Association.
Neural translation technologies and futures
Organized by: Steering committee
September 19, 14:15-16:00, Toyoda Auditorium Hall
Practical Machine Translation
Machine translation technology has been used in practice for decades. Especially the emergence of neural machine translation, achieving significant improvements in recent years, makes it more practical. In this talk, I will first introduce the impacts of the neural machine translation on practical machine translation systems. Then I will analyze the problems in a practical system and the methods used to handle them. In the end, we can see that machine translation with voice recognition and machine vision together are creating new applications and exciting user experiences.
Hua Wu, Ph D, a senior scientist at Baidu, the technical chief of Baidu’s natural language processing department. Her research interests span a wide range of topics including machine translation, dialogue systems, question answering etc., and many of her innovative researches have been applied in Baidu products. She was a leading member of the machine translation project to win the second prize of the State Preeminent Science and Technology Award of China. She was elected to be the Program Co-Chair of ACL 2014, and area chairs of international conferences such as ACL, IJCAI and AAAI.
Google Neural Machine Translation: Status and Challenges
Google Translate started to switch its translation engine from statistical-based to neural network-based system last year. In this talk, I demonstrate how the new system works and new challenges in large scale production neural machine translation systems.
Hideto Kazawa received B.Sc. and M.Sc. from University of Tokyo and Doctor of Engineering from NAIST, Japan. He worked on NLP and ML research in NTT. Now he is a Senior Engineering Manager of Google and working on Google Translate.
The Neural Renaissance: Achieving Critical Mass in Text and Speech Translation
The application of deep learning to speech recognition has had dramatic impacts on the quality of speech recognition systems. Seide et al 2011 showed, for instance, a 32% reduction in Word Error Rate over the previous state of the art Gaussian models, with no changes in training data. More recent work in applying deep learning and neural models to Machine Translation has shown equally dramatic improvements (e.g., Cho et al 2014, Devlin et al 2014, etc.). Advances in both technologies has led to a renaissance in Speech Translation, where improvements in the underlying technologies have truly made Speech Translation feasible. However, it’s not enough to stitch the technologies together; there is significant work in getting the components to talk successfully with each other. Overall, Speech Translation is viable now in a variety of settings, even in highly technical discourse. We have found significant utility in technical talks, such as this, as well as in classroom and lecture settings. In this talk, I will quickly review Microsoft Translator’s work in this space, and will integrate a demo of our speech recognition and translation tech directly into the talk, so that audience members can follow along from their own devices.
Dr. William Lewis is Principal Technical Program Manager with the Microsoft Translator team at Microsoft Research. He has led the Microsoft Translator team’s efforts to build Machine Translation engines for a variety of the world’s languages, including threatened and endangered languages, and has most recently been working on the Translator team’s Speech Translation project. As part of the work on speech translation, Dr. Lewis has been leading the efforts for using the Microsoft Translator app infrastructure and Skype Translator to support the deaf and hard of hearing communities. This work has been extended to the classroom, where “mainstreamed” deaf and hard of hearing students are using MSR’s speech recognition technology to participate fully in the “hearing” classroom, alongside their English Language Learning peers. Before joining Microsoft, Dr. Lewis was Assistant Professor and founding faculty for the Computational Linguistics Master’s Program at the University of Washington, where he continues to hold an Affiliate Appointment, and continues to teach classes on Natural Language Processing.
Semantic and Stylistic Divergences in Machine Translation
September 20, 09:00-10:00, Toyoda Auditorium Hall
Organized by: Program Committee – Research Track
While parallel texts represent invaluable resources for machine translation, they inevitably introduce biases in the cross-lingual
mappings learned by machine translation models. In addition to the domain bias and translationese bias studied in past work,
another form of bias can arise from subtle choices in content and style made by translators to appropriately convey the meaning of the source to their target audience. We will first study the impact of such bias on machine translation training. We argue that it can lead to divergences between source and target texts, and show that these divergences have a substantial impact on the quality of neural machine translation. We then turn to the problem of producing machine translation for a specific audience by controlling not only the content, but also the style of the output.
Marine Carpuat is an Assistant Professor in Computer Science at the University of Maryland. Her research focuses on multilingual natural language processing and machine translation. Before joining Maryland, Marine was a Research Officer at the National Research Council Canada, and a postdoctoral researcher at Columbia University. She received a PhD in Computer Science from the Hong Kong University of Science & Technology, a MPhil in Electrical Engineering from the Hong Kong University of Science & Technology and a Diplome d’Ingenieur from the French Grande Ecole Supelec. Marine currently serves as a board member for the ACL Special Interest Group on Lexical Semantics (SIGLEX) and is on the editorial board of the Computational Linguistics journal. Marine has received best paper awards at *SEM, the Joint Conference on Lexical and Computational Semantics, in 2017 and TALN, the French Conference on
Natural Language Processing in 2010, as well as an Outstanding Teaching Award, and research awards from Google and Amazon.
Introduction of MT into industrial-scale translation workflows with translator acceptance
September 20, 13:30-14:30, Toyoda Auditorium Hall
Organized by: Program Committee – Commercial MT Users and Translators Track
MT is often added as an on-top process to existing “translate-edit-proof” translation workflows. Frequently, too little thought is given as to how professional translators are supposed to engage with such systems. This lack of forethought is painfully visible in the way many translators are paid in conjunction with MT. At the same time, expectations of non-translators relating to the effect of MT in terms of time and quality improvements are often exaggerated, leading to misunderstandings and disappointments on both sides.
The talk will briefly outline the history of MT at SAP SE since the early 1990s, before focusing on its current MT projects, and especially the current MT adoption project. This project aims to create efficient translation process scenarios that foster a higher acceptance of MT amongst SAP’s large external translator community. Key topics will be: · Research: what research ideas led us to the process that we chose · Payment: what our findings were when we explored commercial payment models for MT-related translation work · Change-management and socialization of MT: what we did to positively enhance and foster the topic of MT · Ramp-up: how we are launching this initiative · Strategic partnerships: who we are working with · Expectations: what we expect the effects of this project to be · Future directions: where we take MT from here · Productivity and translator-centric working; what working models we can expect to see for technical translators in the future
Chris Pyne studied French and German literature before going on to teach in the UK. After moving permanently to Germany in 1985, he worked as a technical translator for a publishing house, becoming its general manager 7 years later.
His interest in machine translation and translation automation took him to the Siemens corporation in Munich where he worked for several years on the METAL rule-based MT project. He finally left the project to form a partnership in a translation company which became part of the Lionbridge corporation. Chris stayed with Lionbridge to become their managing director in Germany, before moving to SAP SE in 2002 to set up a translation suppliers network – the SAP translation ecosystem. Since then he has driven the establishment of the ecosystem in over 45 countries including China, Russia, India, Iran and Kazakhstan. He has remained closely involved with MT during this time. He is based at SAP headquarters in Walldorf, Germany.
Social innovation based on speech-to-speech translation technology targeting the 2020 Tokyo Olympic/Paralympic Games
September 21, 09:00-10:00, Toyoda Auditorium Hall
Organized by: Program Committee – Research Track
This talk will focus on Speech-to-Speech (S2S) translation, which enables communication between people across the globe who speak different languages, i.e., breaking the language barriers among them. In 1986, fundamental research on S2S began in Japan; the idea of S2S spread worldwide and has been explored by countless numbers of researchers in the past three decades. First, the advance in ICT such as processors, memories, and wireless communication accelerated this computation-intensive technology. Second, accumulation of digital data such as speech and language data encouraged recent approaches based on machine learning, and finally, field experiments enabled the S2S systems to be utilized not only in the lab but by the public. Now, S2S is available on smartphones and tablet devices.
In 2014, the Japanese Government launched the “Global Communication Plan”. The Plan aims to spread easy-to-use S2S devices throughout the country as fruits of successful collaborations between public and private sectors of Japan, with an eye toward the year 2020, when Tokyo will be hosting the Olympic/Paralympic Games.
In a future not so far from today, the S2S technology will automate simultaneous interpretations. S2S technology promises a great impact on the society, including education, where there will be less needs to learn a second language and people can spend their time on other things. Connecting the people, the society, and the economy with S2S, will help demonstrate unprecedented explosive growth in many fields, and this vision of mine will conclude the talk.
Eiichiro Sumita, received his PhD in Engineering from Kyoto University in 1999, and a Master’s and Bachelor’s in Computer Science from the University of Electro-Communications in 1982 and 1980, respectively.
Before joining NICT, he worked for the Advanced Telecommunications Research Institute International (ATR) in Kyoto, Japan and for IBM Research in Tokyo, Japan.
He served as the Chairperson of the Association for Natural Language Processing (NLP) and an Associate Editor for the ACM Transactions on Asian Language Information Processing (TALIP) and serves a Director of Japan Translation Federation (JTF) and a Director of Asia-Pacific Association of Machine Translation (AAMT).
He is a co-recipient of the Commendation for Science and Technology by the Minister of Education, Culture, Sports, Science and Technology, Prizes for Science and Technology in 2010, Minister for Internal Affairs and Communications Award (11th annual Merit Awards for Industry-Academia-Government Collaboration) in 2013 and the AAMT Nagao Awards in 2007 and 2014.