Exploring storytelling with natural language processing : representation, non-linguistic factors, and narrative engagement

Abstract

Narratives hold significant importance in human societies and have been extensively studied from various perspectives, including literary criticism, cognitive science, and computer science. This PhD thesis employs Natural Language Processing techniques to investigate narratives, focusing on two crucial agents in the narrative comprehension process—the protagonists of the narratives and the readers/audience. The first study introduces a novel approach to character representation in narratives. While previous computational studies have yielded fruitful results by applying sentiment analysis and analysis of social relations on extensive narrative collections, the agency of fictional characters in actively making decisions and influencing their surroundings has not been thoroughly explored. In this thesis, we propose a new entity-based method of analysis called Agency Analysis to measure the extent to which fictional characters proactively shape their lives. We construct a protagonist-centric dataset encompassing 300 protagonists from 261 books based on Project Gutenberg. By integrating Agency Analysis with other commonly measured character features such as sentiment, cognitive process, and interpersonal relationship, we record the time series of each character’s multi- dimensional evolution across the narrative time in this dataset. With this dataset, we aim to enhance our understanding of the narrative structure and character development dynamics. Our research demonstrates the existence of prototypical trajectories in protagonist development. Additionally, we consider non-linguistic factors such as narrative perspectives and authors’ gender and explore their potential influence on narrative arrangements and the overall ratings of the story. The second study delves into the audience’s response to narratives in the form of movies. To what extent movie viewers are swept into a fictional world has always been pondered by psychologists and filmmakers. With the recent development of time-synchronic comments on Chinese online viewing platforms, we can now analyze the immediate responses of viewers toward movies. With over 3 million Chinese time-synchronic comments collected, we apply natural language processing techniques to examine linguistic patterns among comments based on different movie genres. We discovered that people exhibit different commenting styles, as reflected by the linguistic features of their responses when exposed to various types of movie stimuli. Through the assessment of emotion and cognition-related word rates in comments, we also highlight prototypical trajectories of psychological engagement of the audience through time in movie viewing. Finally, we perform a content analysis of the comments with sentence embedding and reveal a number of themes and their distribution across various genres. In conclusion, this thesis contributes to the understanding of narratives by investigating both the roles of literary protagonists and audience engagement. Exploring these two distinct yet complementary perspectives offers an intriguing and comprehensive analysis of narratives. By introducing the Agency Analysis method to assess character agency and exploring time-synchronic comments as a rich source of audience responses, we gain insights into the dynamics of character development and the immediate impact of narratives on viewers. These findings highlight the importance of multidisciplinary approaches, integrating computational analysis with literary, psychological, and film studies, to enhance our comprehension of narratives and their effects on individuals and society.