An Extraction and Representation Pipeline for Literary Characters

Readers of novels need to identify and learn about characters as they develop an understanding of the plot. With the advancement of natural language processing and increased attention on computational narrative understanding, I present an interdisciplinary inquiry on the automated extraction of literary characters from book-length input as well as dense vector representations for literary characters inspired by narrative theory. Specifically, the thesis presents a character identification pipeline with a named entity recognition (NER) module, a coreference resolution module, and a character cluster disambiguation module using both neural network driven and heuristic approaches. Building on previously proposed definitions of literary character, I have developed two kinds of computational representations for characters: event-driven representations and speechdriven presentations. For character-event representations, I have developed an unsupervised event mining pipeline with contextualized and static embeddings extractions and metrics to calculate the aggregated character embeddings. For character-speech representations, I have developed an attention-score-based ranking algorithm for speech embeddings extractions for each character. I have applied my identification and representation pipelines to Jane Austen's Sense and Sensibility and other Austen novels as in-depth case studies to provide both quantitative and qualitative insights into the model's behavior and significance. This thesis is the first to my knowledge that combines a modular pipeline for automated character identification, representation extraction, and comparisons for full-length English novels. First and foremost, I would like to express my deepest gratitude to my mentor and advisor, Carolyn Anderson, for boldly agreeing to take on a pioneering journey prior to your formal appointment at Wellesley as we navigated the uncharted waters together. Thank you, for your tremendous support ranging from technical expertise in modeling to numerous ways to maintain mental well-being, to coming to many concerts that I have invited you to. Thank you for being there to celebrate the little joys in life, and to embrace some of my darkest moments in the process of the thesis. I'd also like to express my sincere appreciation to all my committee members: Professor Eni Mustafaraj, Professor Peter Mawhorter for being on my committee, and Professor Yoon Sun Lee for kindly agreeing to serve as a visiting member. Professor Mustafaraj, we have come a long way together since my first year. Thank you for believing in me and seeing my potential as a researcher while I was so inexperienced and naive in my second semester and thank you for being my first research mentor and inspiring me on how cool doing CS research can be, and for continuing to always be there for me throughout my Wellesley years and always pushing me to become the better person that I can be. Professor Mawhorter, thank you for your immense patience during the CS115 and CS111 labs and office hours, for encouraging me to believe in myself and pursue more computer science and align it with my literary passion, and for spending so many hours hearing and discussing my research in digital humanities with me. Professor Lee, thank you very much for your support and coaching that extended beyond the courses that you're teaching, and for kindly guiding me into navigating narrative theory for my thesis.