Structure- and Context-aware NLP Approaches to Emotion and Subjectivity

Incorporating hierarchical structures of language such as trees has been shown to be effective for various Natural Language Processing (NLP) tasks. The utilization of hierarchical and structural information of text can provide insights into context, compensating for data shortage in many situations. In this dissertation, I investigate and propose novel structure- and context-aware models that aim for effective analysis of emotionality, intentionality, and subjectivity at various levels in text. Further, I explore how pre-trained language models and auto-completion generative algorithms built into modern writing tools and word processors alter the practice and experience of writing for the human user. Lastly, I address the gap that exists in pedagogical approaches in computer science education by proposing a framework that cultivates deeper computational thinking skills that have their roots in Human-Centered AI principles, thus fostering more humane, responsible, and critical approaches to computation.