Dr. Helena Moniz
Title: Crowdsourcing and Related Tools for Quality Monitoring in Post-Editing Machine Translation
Abstract: Crowdsourcing applied in contexts of post-editing Machine Translation allows for substantial improvements in the quality of the MT, but it also poses several challenges in terms of scalable assessment, assurance, and quality monitoring. This talk with focus on the crowdsourcing ecosystem composed of MT leveraged by human edition. In order to monitor this complex ecosystem in both technological and human components, several AI tools were created in an industry scenario. The analytics generated by the tools are then used to provide feedback to the editors and improve the tools themselves, always aiming at a balance between the two components.
Short Bio: Dr. Helena Moniz is a researcher at INESC-ID. Her research interests include social prosody and communication, speech recognition, audio analytics, multimodal analysis, and paralinguistic information, and, more recently, post-editing Machine Translation processes balanced with AI tools. She received a PhD degree in Linguistics from the University of Lisbon, in cooperation with the Technical University of Lisbon (IST), in 2013. She has been working at INESC-ID since 2000, in 15 international and national projects. Since 2015, she has also been a consultant for Linguistic Quality Assurance at Unbabel, a translation company focused on AI leveraged by humans, to deliver quality in a time efficient way, translating distinct content types, e.g., product descriptions, chat, social media, and video subtitles.
Laura Casanellas Luri
Title: Machine Translation is only part of it
Abstract: Digital Transformation surrounds us. At great speed, technology is entering every part of our lives; without us even realising it. Machine Translation seems like a great breakthrough in the world of translation, and indeed it is. But it is only part of the new way we interact with ourselves and the world around us. We might be more successful if we approach it with the fresh eyes of a new born; by doing so we will be able to see the full spectrum of possibilities. If the decision of implementing the technology has been made, fear needs to be left behind. Traditional methods will need to be questioned and some of them might not survive. Are we ready to do that? If we do, we might need to rethink the legacy of best practices that was our handbook until now.
Short Bio: Laura Casanellas specialises in deployment and customization of Machine Translation (MT) programs. Previously she worked in a variety of roles (Language Quality, Vendor Management, Content Management) and verticals (Games, Travel,IT, Automotive, Legal) and acquired extensive experience in all aspects related to Localization. Since 2011 Laura’s focus is on Language Technology and MT. She currently helps companies implement MT in their organizations.
Dr. Arianna Bisazza
Title: Understanding syntactic and semantic transfer in multilingual neural network models
Abstract: Recent work has shown that state-of-the-art models of language and translation can be successfully trained on multiple languages simultaneously without changes to the underlying neural network architectures. Besides the practical advantage of having fewer and smaller models to maintain, this approach has the potential to dramatically improve the quality of MT and other NLP applications in low-resource languages by exploiting cross-language commonalities. As a result, the adoption of multilingual models is quickly catching on, however the mechanisms explaining knowledge sharing in these models remain largely unknown: What kind of knowledge is really shared among languages? Does multilingual training mostly lead to better semantic modeling of the lexicon or does it also enable the sharing of more abstract grammatical categories? And furthermore, what are the conditions for cross-lingual transfer to happen at various levels? This talk will present several answers to these questions based on a variety of models and probing tasks, and discuss the future of multilinguality in NLP.
Short Bio: Dr. Arianna Bisazza is Assistant Professor in natural language processing at Leiden University, Netherlands. Her research aims at identifying intrinsic limitations of current language modeling paradigms as well as improving the quality of machine translation for challenging language pairs. She previously worked as a postdoc at the University of Amsterdam and as a research assistant at Fondazione Bruno Kessler. She obtained her PhD from the University of Trento, Italy, in 2013 and is fully funded by a VENI (NWO's starting grant) since 2016.