Orphée is a senior researcher at the LT3 team with great expertise in readability prediction and text mining of user-generated content. She obtained her PhD in 2015 for which she devised the first state-of-the-art classification-based readability prediction system for Dutch and the first end-to-end system for fine-grained sentiment analysis. For both systems she investigated the added value of incorporating deep semantic knowledge in the form of coreferential relations, semantic roles and linked open data.
During her postdoc, Orphée has been further elaborating her readability research by investigating domain portability and collaborated on papers researching translation quality and post-editing. At the same time, she is exploring new techniques for porting her sentiment analysis pipeline to different domains and languages and has successfully finalized a valorization project with the industry. Currently, she is enthusiastic to undertake the next logical step and employ her expertise on deep semantic processing and readability prediction to the closely related domain of automated writing evaluation.
Orphée is teaching courses on digital communication using blended learning techniques, guides several PhD students and is actively involved in various research projects.
After obtaining her Master's at the Faculty of Applied Language Studies, Orphée became a member of the LT3 research team where she was introduced to the fascinating fields of computational linguistics and machine learning. During her first work on various research projects at the LT3 team she was responsible for the compilation of prestigious Dutch corpus projects and the (semi)-automatic annotation of various semantic layers. She has devised new tools for coreference resolution and semantic role labeling using machine learning techniques. The latter while she was partially employed at the University of Utrecht.
Working on the HENDI project, Orphée developed a full-fledged readability prediction system for Dutch generic text with a focus on the added value of incorporating more semantic knowledge. On the subsequent PARIS project, she worked on the deep linguistic processing of user-generated content by developing normalization and sentiment analysis systems. In the spring of 2014, she spent a semester at the University of Mannheim where she performed research on adding deep semantics to content-based book recommenders and combined sentiment analysis techniques with information derived from linked open data. In the summer of 2015 all this work culminated in the successful defense of her dissertation “Tipping the scales: exploring the added value of deep semantic processing on readability prediction and sentiment analysis.”
As a postdoctoral research assistant, Orphée has been able to continue working on her two favorite research areas. Regarding readability prediction, she has optimized her system for both English and Dutch generic text and has been exploring domain portability. Regarding sentiment analysis, her fine-grained pipeline is in production thanks to a successful valorization project with the industry.