Invited Talk
Lilja Øvrelid
Professor, University of Oslo - Website
Who do you love? Fine-grained sentiment annotation, aggregation and data augmentation
In this talk I will discuss insights from several large-scale annotation projects aimed at enriching texts with subjective assessments describing their sentiment. In particular, I will describe a trajectory which is shared by many annotation projects aimed at supplying various Information Extraction systems with data: moving from fine-grained annotation of text spans at the sub-sentence level to aggregation of information within or across documents. I will start out by discussing the task of structured sentiment analysis and in particular, the 2022 SemEval shared task devoted to this task. In follow-up work, we assess the usefulness of these annotations for the aggregation of entity-level sentiment annotation for longer documents and provide a study of remaining challenges for the cross-sentence tracking of sentiment. Finally, I will briefly touch on recent work on data augmentation for sentiment analysis evaluating both masked language models and LLM-based prompting in the context of sentiment annotation.
Bio
Lilja Øvrelid is a Professor of Informatics at the University of Oslo, where she leads the Language Technology Group. She has been involved in a number of large annotation projects devoted to a diverse set of NLP tasks, such as UD syntax, NER, negation, sentiment analysis, coreference resolution and question-answering. She has worked on syntactic and semantic parsing, as well as the application of structured prediction approaches to NLP tasks beyond parsing, such as fine-grained sentiment analysis and event extraction.
