The increasing amount of information on the web, including critical e-services (e.g. social benefits), has forced users to move from phone calls and personal visits to performing tasks online. This online content may be challenging for some individuals (e.g. non-native speakers and cognitively impaired), especially the use of complex words and syntactic structures, together with technical terms. In this talk, I will present the task of text simplification (TS), that consists of monolingual text-to-text transformations for simplifying complex texts. Although traditional work on TS is divided in lexical simplification (LS) and syntactic simplification (SS), our research, within the EU H2020 SIMPATICO project, focuses on applications that require a single solution that encompasses all possible simplification operations together. In this talk, I will present our work on data-driven TS that aims to go beyond sequence-to-sequence models. I will show that neural TS models informed with specific information from the TS task and/or from the target audience present significantly better results than previous state-of-the-art approaches. This highlights the importance of personalising TS applications and shows the potential of personalised NLP tasks.

Bio: Dr Carolina Scarton is a Research Associate at the University of Sheffield and a member of the Natural Language Processing Group. She holds a PhD from the University of Sheffield on Quality Estimation of Machine Translation and her research interests are quality estimation, text simplification, misinformation detection/verification, readability assessment, personalised NLP, dialogue systems and NLP for robotics. Currently, Dr Scarton works for the EC H2020 WeVerify project, focusing on automatically detecting and verifying online misinformation. Previously, she worked in the EC H2020 SIMPATICO project, where she developed approaches for sentence simplification. From September 2019, Dr Scarton will start an Academic Fellowship in the University of Sheffield.