Applying natural language processing to narrative discourse theory
Stories play an integral role in our lives, whether as an artistic medium, tradition, or means to share our personal experiences. Although social, cultural, and sentimental aspects are deeply embedded in any story, we often overlook the technical structures that underlie their narratives. Emerging from the intersections of narrative theory and natural language processing (NLP), computational narratology is a field that models narrative structures with computable representations (1). This field is influenced by themes in humanities narratology, linguistics, and computational large language models (1). In textual narratives, computational narratology is employed for testing literary hypotheses or the automatic interpretation and generation of stories (1). In computational narratives, it appears in AI systems employing human-computer interaction or story-based video games (3). The field is especially gaining prominence in computer graphics and games, where interactive narrative is deemed the “holy grail” of design (3). In all these examples, the field explores the algorithmic processes involved in interpreting and creating narratives (1).
Computational narratology is divided into two levels of interaction: interactions between different narrative features, known as the classical theory, and audience interactions with narrative, the postclassical theory (2). Most research is founded on the classical approach introduced in 1972, also named narrative discourse theory, by literary theorist and narratologist Gérard Genette (4). Genette focuses on how a narrative is structured instead of what is told, using a hierarchical classification of terms to describe it (4). He distinguishes the narrative into three components: the story (the contents recounted in a story), the discourse (the order and economy of their telling), and narrating (the narrator’s role in shaping the content) (2). Broadly, the term “narrative discourse” refers to the interactions between these three nodes, and how the narrator’s input affects the story and its structure (5). Genette further elaborates on the relationships between these dimensions through three additional linking functions: tense (the narrative’s temporal aspects), mood (the modalities of expressing the story), and voice (perspectival issues like point of view and dialogue) (4).
Schema showing the relationships in Genette’s theory: https://ars.els-cdn.com/content/image/1-s2.0-S1389041719304656-gr1.jpg
Genette’s work, the classical narrative discourse theory, strongly correlates with what an NLP model requires to produce accurate results about narrative. The most basic structure these models need is a causal chain of events that transforms the initial state of the story world into its final version, which renders Genette’s theory strongly applicable (1). On the flip side, NLP is perfect for tasks like modeling a linear order of events, assembling events into general “frames,” or generating probability-based models (2). Therefore, NLP is applied for two central reasons within this theory: story understanding (detecting story types, character types, or storylines) and structure understanding (plot arcs, turning points, and non-linear events).
- [Photo of Gérard Genette] https://upload.wikimedia.org/wikipedia/commons/thumb/6/67/G%C3%A9rard_Genette.jpg/640px-G%C3%A9rard_Genette.jpg
Although the computer narratology field has its limitations—such as representing humor and irony or when story understanding requires human commonsense knowledge—it also benefits many domains (3). Overall, inspired by Genette’s narrative discourse theory, computational narratology aims to make computers better entertainers, educators, communicators, and understanders of the elements that shape our human world.
Citations
- Mani, I. (2013, September 15). Computational Narratology. the living handbook of narratology. Retrieved from https://www-archiv.fdm.uni-hamburg.de/lhn/node/43.html.
- Piper, A. Jean So, R. Bamman, D. (2021, November 7-11). Narrative Theory for Computational Narrative Understanding. 2021 Association for Computational Linguistics. Retrieved from https://aclanthology.org/2021.emnlp-main.26.pdf.
- Riedl, M. (2017, October 24). Computational Narrative Intelligence: Past, Present, and Future. Medium. Retrieved from https://mark-riedl.medium.com/computational-narrative-intelligence-past-present-and-future-99e58cf25ffa#:~:text=In%20the%20following%20video%2C%20we,when%20they%20go%20to%20pharmacies.&text=If%20playback%20doesn%27t%20begin%20shortly%2C%20try%20restarting%20your%20device.
- Akimoto, T. (2019, December). Narrative structure in the mind: Translating Genette’s narrative discourse theory into a cognitive system. ScienceDirect Cognitive Systems Research. Retrieved from https://www.sciencedirect.com/science/article/pii/S1389041719304656#:~:text=Genette%27s%20narrative%20discourse%20theory%20is,instead%20of%20what%20is%20told.
- Piper, A. Bagga, S. (2024, November 15). Using Large Language Models for Understanding Narrative Discourse. 2024 Association for Computational Linguistics. Retrieved from https://aclanthology.org/2024.wnu-1.4.pdf.